Release v3.8.4

.github/workflows/update-api-md.yml

@@ -0,0 +1,57 @@
+name: Update API Reference
+
+on:
+  push:
+    branches: [ "main" ]
+    paths:
+      - 'src/praisonai-agents/praisonaiagents/**/*.py'
+      - 'src/praisonai/praisonai/**/*.py'
+      - 'src/praisonai-ts/src/**/*.ts'
+      - '.github/workflows/update-api-md.yml'
+
+permissions:
+  contents: write
+
+jobs:
+  update-api-md:
+    runs-on: ubuntu-latest
+    
+    steps:
+    - name: Checkout repository
+      uses: actions/checkout@v4
+      with:
+        ref: main
+        persist-credentials: true
+        fetch-depth: 0
+
+    - name: Set up Python
+      uses: actions/setup-python@v5
+      with:
+        python-version: '3.11'
+
+    - name: Generate api.md
+      run: |
+        cd $GITHUB_WORKSPACE
+        python3 src/praisonai/scripts/generate_api_md.py
+        
+    - name: Check for changes
+      id: check_changes
+      run: |
+        if git diff --quiet api.md; then
+          echo "changes=false" >> $GITHUB_OUTPUT
+          echo "No changes to api.md"
+        else
+          echo "changes=true" >> $GITHUB_OUTPUT
+          echo "api.md has been updated"
+          git diff api.md | head -50
+        fi
+
+    - name: Commit and push changes
+      if: steps.check_changes.outputs.changes == 'true'
+      uses: stefanzweifel/git-auto-commit-action@v5
+      with:
+        commit_message: "docs: auto-update api.md [skip ci]"
+        file_pattern: 'api.md'
+        commit_user_name: "github-actions[bot]"
+        commit_user_email: "41898282+github-actions[bot]@users.noreply.github.com"
+        commit_author: "github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>"

.gitignore

@@ -83,4 +83,5 @@ issue_details.json
 # MCP Registry tokens
 .mcpregistry_*
 
-*.log
+*.log
+.internal

api.md

@@ -60,6 +60,7 @@ Methods:
 * <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">get_skills_prompt</a>() -> str</code>
 * <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">handoff_to</a>(target_agent: 'Agent', prompt: str, context: Optional[Dict[str, Any]] = None, config: Optional['HandoffConfig'] = None) -> 'HandoffResult'</code>
 * <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">handoff_to_async</a>(target_agent: 'Agent', prompt: str, context: Optional[Dict[str, Any]] = None, config: Optional['HandoffConfig'] = None) -> 'HandoffResult'</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">iter_stream</a>(prompt: str, **kwargs)</code>
 * <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">launch</a>(path: str = '/', port: int = 8000, host: str = '0.0.0.0', debug: bool = False, protocol: str = 'http')</code>
 * <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">llm_model</a>()</code>
 * <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">output_style</a>()</code>
@@ -76,7 +77,7 @@ Methods:
 * <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">run_autonomous</a>(prompt: str, max_iterations: Optional[int] = None, timeout_seconds: Optional[float] = None)</code>
 * <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">session_id</a>() -> Optional[str]</code>
 * <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">skill_manager</a>()</code>
-* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">start</a>(prompt: str, **kwargs)</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">start</a>(prompt: str = None, **kwargs)</code>
 * <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">store_memory</a>(content: str, memory_type: str = 'short_term', **kwargs)</code>
 * <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">switch_model</a>(new_model: str) -> None</code>
 * <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">thinking_budget</a>()</code>
@@ -116,7 +117,7 @@ Methods:
 * <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">save_output_to_file</a>(task, task_output)</code>
 * <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">save_session_state</a>(session_id: str, include_memory: bool = True) -> None</code>
 * <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">set_state</a>(key: str, value: Any) -> None</code>
-* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">start</a>(content = None, return_dict = False, **kwargs)</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">start</a>(content = None, return_dict = False, output = None, **kwargs)</code>
 * <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">todo_list</a>()</code>
 * <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">update_plan_step_status</a>(step_id: str, status: str) -> bool</code>
 * <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">update_state</a>(updates: Dict) -> None</code>
@@ -193,7 +194,7 @@ Methods:
 * <code title="class PraisonAIAgents">PraisonAIAgents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">save_output_to_file</a>(task, task_output)</code>
 * <code title="class PraisonAIAgents">PraisonAIAgents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">save_session_state</a>(session_id: str, include_memory: bool = True) -> None</code>
 * <code title="class PraisonAIAgents">PraisonAIAgents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">set_state</a>(key: str, value: Any) -> None</code>
-* <code title="class PraisonAIAgents">PraisonAIAgents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">start</a>(content = None, return_dict = False, **kwargs)</code>
+* <code title="class PraisonAIAgents">PraisonAIAgents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">start</a>(content = None, return_dict = False, output = None, **kwargs)</code>
 * <code title="class PraisonAIAgents">PraisonAIAgents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">todo_list</a>()</code>
 * <code title="class PraisonAIAgents">PraisonAIAgents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">update_plan_step_status</a>(step_id: str, status: str) -> bool</code>
 * <code title="class PraisonAIAgents">PraisonAIAgents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">update_state</a>(updates: Dict) -> None</code>
@@ -314,6 +315,8 @@ Methods:
 * <code title="class Memory">Memory.<a href="./src/praisonai-agents/praisonaiagents/memory/memory.py">compute_quality_score</a>(completeness: float, relevance: float, clarity: float, accuracy: float, weights: Dict[str, float] = None) -> float</code>
 * <code title="class Memory">Memory.<a href="./src/praisonai-agents/praisonaiagents/memory/memory.py">finalize_task_output</a>(content: str, agent_name: str, quality_score: float, threshold: float = 0.7, metrics: Dict[str, Any] = None, task_id: str = None)</code>
 * <code title="class Memory">Memory.<a href="./src/praisonai-agents/praisonaiagents/memory/memory.py">get_all_memories</a>() -> List[Dict[str, Any]]</code>
+* <code title="class Memory">Memory.<a href="./src/praisonai-agents/praisonaiagents/memory/memory.py">get_learn_context</a>() -> str</code>
+* <code title="class Memory">Memory.<a href="./src/praisonai-agents/praisonaiagents/memory/memory.py">learn</a>()</code>
 * <code title="class Memory">Memory.<a href="./src/praisonai-agents/praisonaiagents/memory/memory.py">reset_all</a>()</code>
 * <code title="class Memory">Memory.<a href="./src/praisonai-agents/praisonaiagents/memory/memory.py">reset_entity_only</a>()</code>
 * <code title="class Memory">Memory.<a href="./src/praisonai-agents/praisonaiagents/memory/memory.py">reset_long_term</a>()</code>
@@ -543,16 +546,16 @@ Methods:
 * <code title="function">praisonaiagents.<a href="./src/praisonai-agents/praisonaiagents/main.py">display_error</a>(message: str, console = None)</code>
 * <code title="function">praisonaiagents.<a href="./src/praisonai-agents/praisonaiagents/main.py">display_generating</a>(content: str = '', start_time: Optional[float] = None)</code>
 * <code title="function">praisonaiagents.<a href="./src/praisonai-agents/praisonaiagents/main.py">display_instruction</a>(message: str, console = None, agent_name: str = None, agent_role: str = None, agent_tools: List[str] = None)</code>
-* <code title="function">praisonaiagents.<a href="./src/praisonai-agents/praisonaiagents/main.py">display_interaction</a>(message, response, markdown = True, generation_time = None, console = None, agent_name = None, agent_role = None, agent_tools = None, task_name = None, task_description = None, task_id = None)</code>
+* <code title="function">praisonaiagents.<a href="./src/praisonai-agents/praisonaiagents/main.py">display_interaction</a>(message, response, markdown = True, generation_time = None, console = None, agent_name = None, agent_role = None, agent_tools = None, task_name = None, task_description = None, task_id = None, metrics = None)</code>
 * <code title="function">praisonaiagents.<a href="./src/praisonai-agents/praisonaiagents/main.py">display_self_reflection</a>(message: str, console = None)</code>
-* <code title="function">praisonaiagents.<a href="./src/praisonai-agents/praisonaiagents/main.py">display_tool_call</a>(message: str, console = None, tool_name: str = None, tool_input: dict = None, tool_output: str = None)</code>
+* <code title="function">praisonaiagents.<a href="./src/praisonai-agents/praisonaiagents/main.py">display_tool_call</a>(message: str, console = None, tool_name: str = None, tool_input: dict = None, tool_output: str = None, elapsed_time: float = None, success: bool = True)</code>
 * <code title="function">praisonaiagents.<a href="./src/praisonai-agents/praisonaiagents/main.py">register_display_callback</a>(display_type: str, callback_fn, is_async: bool = False)</code>
 
 # Other
 
 Types:
 ```python
-from praisonaiagents import AutoRagConfig, Citation, CodeExecutionStep, DeepResearchResponse, ExpandResult, ExpandStrategy, FileSearchCall, MCPCall, Provider, RagRetrievalPolicy, ReasoningStep, RewriteResult, RewriteStrategy, Task, WebSearchCall
+from praisonaiagents import AutoRagConfig, Citation, CodeExecutionStep, DeepResearchResponse, ExpandResult, ExpandStrategy, FileSearchCall, MCPCall, Provider, RagRetrievalPolicy, ReasoningStep, RewriteResult, RewriteStrategy, Task, WebSearchCall, warmup
 ```
 
 Methods:
@@ -569,6 +572,7 @@ Methods:
 * <code title="class Task">Task.<a href="./src/praisonai-agents/praisonaiagents/task/task.py">execute_callback_sync</a>(task_output: TaskOutput) -> None</code>
 * <code title="class Task">Task.<a href="./src/praisonai-agents/praisonaiagents/task/task.py">initialize_memory</a>()</code>
 * <code title="class Task">Task.<a href="./src/praisonai-agents/praisonaiagents/task/task.py">store_in_memory</a>(content: str, agent_name: str = None, task_id: str = None)</code>
+* <code title="function">praisonaiagents.<a href="./src/praisonai-agents/praisonaiagents/__init__.py">warmup</a>(include_litellm: bool = False, include_openai: bool = True) -> dict</code>
 
 # Wrapper (praisonai)
 
@@ -614,8 +618,11 @@ Methods:
 * <code title="cli">praisonai context clear <a href="./src/praisonai/praisonai/cli/commands/context.py">--help</a></code>
 * <code title="cli">praisonai context compact <a href="./src/praisonai/praisonai/cli/commands/context.py">--help</a></code>
 * <code title="cli">praisonai context export <a href="./src/praisonai/praisonai/cli/commands/context.py">--help</a></code>
+* <code title="cli">praisonai context grep <a href="./src/praisonai/praisonai/cli/commands/context.py">--help</a></code>
+* <code title="cli">praisonai context list <a href="./src/praisonai/praisonai/cli/commands/context.py">--help</a></code>
 * <code title="cli">praisonai context show <a href="./src/praisonai/praisonai/cli/commands/context.py">--help</a></code>
 * <code title="cli">praisonai context stats <a href="./src/praisonai/praisonai/cli/commands/context.py">--help</a></code>
+* <code title="cli">praisonai context tail <a href="./src/praisonai/praisonai/cli/commands/context.py">--help</a></code>
 * <code title="cli">praisonai debug acp <a href="./src/praisonai/praisonai/cli/commands/debug.py">--help</a></code>
 * <code title="cli">praisonai debug debug-callback <a href="./src/praisonai/praisonai/cli/commands/debug.py">--help</a></code>
 * <code title="cli">praisonai debug interactive <a href="./src/praisonai/praisonai/cli/commands/debug.py">--help</a></code>
@@ -670,14 +677,19 @@ Methods:
 * <code title="cli">praisonai lsp status <a href="./src/praisonai/praisonai/cli/commands/lsp.py">--help</a></code>
 * <code title="cli">praisonai lsp stop <a href="./src/praisonai/praisonai/cli/commands/lsp.py">--help</a></code>
 * <code title="cli">praisonai mcp add <a href="./src/praisonai/praisonai/cli/commands/mcp.py">--help</a></code>
+* <code title="cli">praisonai mcp describe <a href="./src/praisonai/praisonai/cli/commands/mcp.py">--help</a></code>
 * <code title="cli">praisonai mcp list <a href="./src/praisonai/praisonai/cli/commands/mcp.py">--help</a></code>
 * <code title="cli">praisonai mcp mcp-callback <a href="./src/praisonai/praisonai/cli/commands/mcp.py">--help</a></code>
 * <code title="cli">praisonai mcp remove <a href="./src/praisonai/praisonai/cli/commands/mcp.py">--help</a></code>
+* <code title="cli">praisonai mcp status <a href="./src/praisonai/praisonai/cli/commands/mcp.py">--help</a></code>
+* <code title="cli">praisonai mcp sync <a href="./src/praisonai/praisonai/cli/commands/mcp.py">--help</a></code>
 * <code title="cli">praisonai mcp test <a href="./src/praisonai/praisonai/cli/commands/mcp.py">--help</a></code>
+* <code title="cli">praisonai mcp tools <a href="./src/praisonai/praisonai/cli/commands/mcp.py">--help</a></code>
 * <code title="cli">praisonai memory add <a href="./src/praisonai/praisonai/cli/commands/memory.py">--help</a></code>
 * <code title="cli">praisonai memory clear <a href="./src/praisonai/praisonai/cli/commands/memory.py">--help</a></code>
 * <code title="cli">praisonai memory search <a href="./src/praisonai/praisonai/cli/commands/memory.py">--help</a></code>
 * <code title="cli">praisonai memory show <a href="./src/praisonai/praisonai/cli/commands/memory.py">--help</a></code>
+* <code title="cli">praisonai memory status <a href="./src/praisonai/praisonai/cli/commands/memory.py">--help</a></code>
 * <code title="cli">praisonai package install <a href="./src/praisonai/praisonai/cli/commands/package.py">--help</a></code>
 * <code title="cli">praisonai package list <a href="./src/praisonai/praisonai/cli/commands/package.py">--help</a></code>
 * <code title="cli">praisonai package uninstall <a href="./src/praisonai/praisonai/cli/commands/package.py">--help</a></code>
@@ -728,12 +740,19 @@ Methods:
 * <code title="cli">praisonai session resume <a href="./src/praisonai/praisonai/cli/commands/session.py">--help</a></code>
 * <code title="cli">praisonai session show <a href="./src/praisonai/praisonai/cli/commands/session.py">--help</a></code>
 * <code title="cli">praisonai skills create <a href="./src/praisonai/praisonai/cli/commands/skills.py">--help</a></code>
+* <code title="cli">praisonai skills info <a href="./src/praisonai/praisonai/cli/commands/skills.py">--help</a></code>
 * <code title="cli">praisonai skills install <a href="./src/praisonai/praisonai/cli/commands/skills.py">--help</a></code>
 * <code title="cli">praisonai skills list <a href="./src/praisonai/praisonai/cli/commands/skills.py">--help</a></code>
+* <code title="cli">praisonai skills search <a href="./src/praisonai/praisonai/cli/commands/skills.py">--help</a></code>
 * <code title="cli">praisonai skills validate <a href="./src/praisonai/praisonai/cli/commands/skills.py">--help</a></code>
+* <code title="cli">praisonai standardise check <a href="./src/praisonai/praisonai/cli/commands/standardise.py">--help</a></code>
+* <code title="cli">praisonai standardise fix <a href="./src/praisonai/praisonai/cli/commands/standardise.py">--help</a></code>
+* <code title="cli">praisonai standardise init <a href="./src/praisonai/praisonai/cli/commands/standardise.py">--help</a></code>
+* <code title="cli">praisonai standardise report <a href="./src/praisonai/praisonai/cli/commands/standardise.py">--help</a></code>
 * <code title="cli">praisonai templates create <a href="./src/praisonai/praisonai/cli/commands/templates.py">--help</a></code>
 * <code title="cli">praisonai templates list <a href="./src/praisonai/praisonai/cli/commands/templates.py">--help</a></code>
 * <code title="cli">praisonai test info <a href="./src/praisonai/praisonai/cli/commands/test.py">--help</a></code>
+* <code title="cli">praisonai test interactive <a href="./src/praisonai/praisonai/cli/commands/test.py">--help</a></code>
 * <code title="cli">praisonai test run <a href="./src/praisonai/praisonai/cli/commands/test.py">--help</a></code>
 * <code title="cli">praisonai todo add <a href="./src/praisonai/praisonai/cli/commands/todo.py">--help</a></code>
 * <code title="cli">praisonai todo done <a href="./src/praisonai/praisonai/cli/commands/todo.py">--help</a></code>

docker/Dockerfile.chat

@@ -16,7 +16,7 @@ RUN mkdir -p /root/.praison
 # Install Python packages (using latest versions)
 RUN pip install --no-cache-dir \
     praisonai_tools \
-    "praisonai>=3.7.2" \
+    "praisonai>=3.8.4" \
     "praisonai[chat]" \
     "embedchain[github,youtube]"
 

docker/Dockerfile.dev

@@ -20,7 +20,7 @@ RUN mkdir -p /root/.praison
 # Install Python packages (using latest versions)
 RUN pip install --no-cache-dir \
     praisonai_tools \
-    "praisonai>=3.7.2" \
+    "praisonai>=3.8.4" \
     "praisonai[ui]" \
     "praisonai[chat]" \
     "praisonai[realtime]" \

docker/Dockerfile.ui

@@ -16,7 +16,7 @@ RUN mkdir -p /root/.praison
 # Install Python packages (using latest versions)
 RUN pip install --no-cache-dir \
     praisonai_tools \
-    "praisonai>=3.7.2" \
+    "praisonai>=3.8.4" \
     "praisonai[ui]" \
     "praisonai[crewai]"
 

src/praisonai-agents/01_basic_token_tracking.py

@@ -13,7 +13,7 @@ def main():
         role="Research Specialist", 
         goal="Find information about a topic",
         backstory="You are an expert researcher who finds comprehensive information.",
-        verbose=True,
+        output="verbose",
         llm="gpt-5-nano",  # Use a cost-effective model for examples
         metrics=True  # 🎯 SIMPLIFIED: Just add this parameter!
     )
@@ -29,7 +29,7 @@ def main():
     agents = Agents(
         agents=[research_agent],
         tasks=[research_task],
-        verbose=True
+        output="verbose"
     )
     
     print("🚀 Running agent task...")

src/praisonai-agents/02_multi_agent_workflow_monitoring.py

@@ -104,8 +104,7 @@ def main():
         workflow = Agents(
             agents=[researcher, analyst, writer],
             tasks=[research_task, analysis_task, report_task], 
-            process="sequential",
-            verbose=True
+            process="sequential"
         )
         
         result = workflow.start()

src/praisonai-agents/03_agent_with_tools_monitoring.py

@@ -11,7 +11,7 @@ Shows how to monitor agent performance when using external tools.
 """
 
 from praisonaiagents import Agent
-from praisonaiagents.tools import duckduckgo, wikipedia
+from praisonaiagents.tools import duckduckgo, wikipedia_tools
 from praisonaiagents.telemetry import (
     monitor_function, track_api_call, get_api_stats,
     get_slowest_functions, performance_monitor
@@ -49,7 +49,7 @@ def main():
         Use the available tools to search for information and provide comprehensive answers.
         Be thorough in your research and cite your sources.""",
         llm="gpt-5-nano",
-        tools=[duckduckgo, wikipedia]  # Add search tools
+        tools=[duckduckgo, wikipedia_tools]  # Add search tools
     )
     
     # Test queries to demonstrate tool performance monitoring
@@ -97,7 +97,9 @@ def main():
     # Slowest functions analysis
     print("\n🐌 Slowest Functions:")
     slowest = get_slowest_functions()
-    for func_name, avg_time in slowest:
+    for item in slowest[:5]:
+        func_name = item.get('function', 'unknown')
+        avg_time = item.get('average_time', 0)
         print(f"  {func_name}: {avg_time:.3f}s average")
     
     # Real-time performance data

src/praisonai-agents/06_memory_agent_monitoring.py

@@ -165,7 +165,9 @@ def main():
     # Slowest operations
     print("\n🐌 Slowest Memory Operations:")
     slowest = get_slowest_functions()
-    for func_name, avg_time in slowest[:5]:  # Top 5 slowest
+    for item in slowest[:5]:  # Top 5 slowest
+        func_name = item.get('function', 'unknown')
+        avg_time = item.get('average_time', 0)
         print(f"  {func_name}: {avg_time:.3f}s average")
     
     # Memory operation breakdown

src/praisonai-agents/08_custom_tools_monitoring.py

@@ -263,7 +263,7 @@ def main():
     # Slowest operations
     print("\n🐌 Slowest Tool Operations:")
     slowest = get_slowest_functions()
-    tool_slowest = [(name, time) for name, time in slowest if 'tool' in name.lower()]
+    tool_slowest = [(item.get('function', 'unknown'), item.get('average_time', 0)) for item in slowest if 'tool' in item.get('function', '').lower()]
     
     for tool_name, avg_time in tool_slowest[:3]:
         print(f"  {tool_name}: {avg_time:.3f}s")

src/praisonai-agents/10_comprehensive_dashboard.py

@@ -302,13 +302,17 @@ def display_dashboard(dashboard_data, benchmark_results):
     perf_data = dashboard_data['performance_data']
     
     print("  Fastest Functions:")
-    fastest_functions = sorted(perf_data['slowest_functions'], key=lambda x: x[1])[:3]
-    for name, time in fastest_functions:
-        print(f"    {name}: {time:.3f}s")
+    fastest_functions = sorted(perf_data['slowest_functions'], key=lambda x: x.get('average_time', float('inf')))[:3]
+    for item in fastest_functions:
+        name = item.get('function', 'unknown')
+        time_val = item.get('average_time', 0)
+        print(f"    {name}: {time_val:.3f}s")
     
     print("  Slowest Functions:")
-    for name, time in perf_data['slowest_functions'][:3]:
-        print(f"    {name}: {time:.3f}s")
+    for item in perf_data['slowest_functions'][:3]:
+        name = item.get('function', 'unknown')
+        time_val = item.get('average_time', 0)
+        print(f"    {name}: {time_val:.3f}s")
     
     # Benchmark Results
     print(f"\n⚡ BENCHMARK RESULTS")

src/praisonai-agents/agent-profile.py

@@ -0,0 +1,63 @@
+"""
+Streaming Example with Profiling
+
+Key Points:
+1. output="stream" preset enables streaming by default
+2. stream=True in start() is redundant when using output="stream" (but doesn't hurt)
+3. Profiler.enable() MUST be called before profiling works
+4. Use Profiler.streaming() context manager for TTFT metrics
+"""
+from praisonai.profiler import Profiler, profile, profile_imports
+
+# IMPORTANT: Enable profiler BEFORE any profiling operations
+Profiler.enable()
+
+# Profile imports
+with profile_imports():
+    from praisonaiagents import Agent
+
+@profile
+def main():
+    agent = Agent(
+        instructions="You are a helpful assistant",
+        output="stream"  # This already enables streaming
+    )
+
+    print("\nStreaming response:")
+    print("-" * 40)
+    
+    # Use Profiler.streaming() to track TTFT (Time To First Token)
+    with Profiler.streaming("llm_stream") as tracker:
+        first_chunk = True
+        for chunk in agent.start("Write a short paragraph about the history of computing"):
+            if first_chunk:
+                tracker.first_token()  # Mark TTFT
+                first_chunk = False
+            tracker.chunk()  # Count chunks
+            print(chunk, end="", flush=True)
+    
+    print("\n" + "-" * 40)
+
+# Run with profiling blocks
+with Profiler.block("full_workflow"):
+    with Profiler.block("setup"):
+        pass
+    
+    with Profiler.block("execution"):
+        result = main()
+    
+    with Profiler.block("cleanup"):
+        pass
+
+# Generate report
+Profiler.report()
+
+# Show streaming-specific stats
+streaming_records = Profiler.get_streaming_records()
+if streaming_records:
+    print("\nStreaming Metrics:")
+    for record in streaming_records:
+        print(f"  {record.name}: TTFT={record.ttft_ms:.2f}ms, Total={record.total_ms:.2f}ms, Chunks={record.chunk_count}")
+
+stats = Profiler.get_statistics()
+print(f"\nMedian execution time: {stats['p50']:.2f}ms")

src/praisonai-agents/agentic_parallelisation.py

@@ -21,7 +21,7 @@ article_agent = Agent(
     goal="Fetch the user's favorite morning article",
     tools=[fetch_favorite_article],
     llm="gemini/gemini-2.5-flash-lite-preview-06-17",
-    verbose=True
+    output="verbose"
 )
 
 news_agent = Agent(
@@ -30,7 +30,7 @@ news_agent = Agent(
     goal="Search for trending news in Kenya",
     tools=[search_trending_kenya],
     llm="gemini/gemini-2.5-flash-lite-preview-06-17",
-    verbose=True
+    output="verbose"
 )
 
 twitter_agent = Agent(
@@ -39,7 +39,7 @@ twitter_agent = Agent(
     goal="Fetch the latest Twitter feed",
     tools=[fetch_twitter_feed],
     llm="gemini/gemini-2.5-flash-lite-preview-06-17",
-    verbose=True
+    output="verbose"
 )
 
 aggregator = Agent(
@@ -47,7 +47,7 @@ aggregator = Agent(
     role="Result Aggregator",
     goal="Aggregate and summarize all results",
     llm="gemini/gemini-2.5-flash-lite-preview-06-17",
-    verbose=True
+    output="verbose"
 )
 
 # Tasks for each agent
@@ -92,7 +92,7 @@ async def main():
         agents=[article_agent, news_agent, twitter_agent, aggregator],
         tasks=[article_task, news_task, twitter_task, aggregate_task],
         process="workflow",
-        verbose=True
+        output="verbose"
     )
     results = await workflow.astart(dict_output=True)
 

src/praisonai-agents/any-llm-agents-async.py

@@ -5,8 +5,8 @@ async def main():
     agent = Agent(
         instructions="You are a helpful assistant",
         llm="gemini/gemini-1.5-flash-8b",
-        self_reflect=True,
-        verbose=True
+        reflection=True,
+        output="verbose"
     )
 
     # Use achat instead of start/chat for async operation

src/praisonai-agents/async-tavily-agents-gemini.py

@@ -79,7 +79,7 @@ async def run_parallel_searches():
         backstory="Expert in web search and information retrieval. You have access to the tavily_search tool to search the web.",
         tools=[tavily_search],
         llm=llm_config,
-        verbose=True
+        output="verbose"
     )
     
     # Create summary agent

src/praisonai-agents/basic-agents.py

@@ -1,14 +1,3 @@
-#!/usr/bin/env python3
-
 from praisonaiagents import Agent
-
-print("Creating agent...")
-agent = Agent(
-    instructions="You are a helpful assistant"
-)
-
-print("Starting agent...")
-result = agent.start("Why is the sky blue?")
-
-print("Agent completed successfully!")
-print(f"Result: {result}") 
+agent = Agent(instructions="You are a helpful assistant")
+result = agent.start("Why is the sky blue?")

src/praisonai-agents/benchmarks/EXECUTION_BENCHMARK_RESULTS.md

@@ -1,6 +1,6 @@
 # PraisonAI Agents - Real Execution Benchmark
 
-**Generated:** 2025-12-18 14:43:52
+**Generated:** 2026-01-13 06:07:28
 **Model:** gpt-4o-mini
 **Iterations:** 3
 **Prompt:** "What is 2+2? Reply with just the number."
@@ -9,18 +9,18 @@
 
 | Framework | Method | Avg Time | Relative |
 |-----------|--------|----------|----------|
-| **PraisonAI** | `agent.start()` | **0.45s** | **1.00x (fastest)** |
-| PraisonAI (LiteLLM) | `agent.start()` | 0.55s | 1.22x |
-| CrewAI | `crew.kickoff()` | 0.58s | 1.28x |
-| Agno | `agent.run()` | 0.92s | 2.05x |
+| **PraisonAI (LiteLLM)** | `agent.start()` | **0.55s** | **1.00x (fastest)** |
+| PraisonAI | `agent.start()` | 0.64s | 1.15x |
+| CrewAI | `crew.kickoff()` | 0.95s | 1.71x |
+| Agno | `agent.run()` | 1.05s | 1.90x |
 
 ## Package Versions
 
 | Package | Version |
 |---------|--------|
-| PraisonAI | 0.1.5 |
-| Agno | 2.3.14 |
-| CrewAI | 1.6.1 |
+| PraisonAI | 0.11.7 |
+| Agno | 2.3.25 |
+| CrewAI | 1.8.0 |
 
 ## How to Reproduce
 

src/praisonai-agents/benchmarks/REAL_BENCHMARK_RESULTS.md

@@ -1,6 +1,6 @@
 # PraisonAI Agents - Real-World Execution Benchmark
 
-**Generated:** 2025-12-18 14:33:49
+**Generated:** 2026-01-13 06:05:09
 **Model:** gpt-4o-mini
 **Iterations:** 3 per test
 
@@ -8,60 +8,60 @@
 
 | Test | PraisonAI (avg) | Agno (avg) | Difference |
 |------|-----------------|------------|------------|
-| Single Agent | 2.04s | 3.18s | PraisonAI 1.56x faster |
-| 2 Agents | 2.49s | 20.49s | PraisonAI 8.21x faster |
-| 3 Agents | 6.52s | 13.97s | PraisonAI 2.14x faster |
+| Single Agent | 2.09s | 3.25s | PraisonAI 1.55x faster |
+| 2 Agents | 2.50s | 9.73s | PraisonAI 3.89x faster |
+| 3 Agents | 6.25s | 13.56s | PraisonAI 2.17x faster |
 
 ## Detailed Results
 
 ### praisonai_single
-- Average: 2.04s
-- Min: 1.95s
-- Max: 2.19s
-- Std Dev: 0.13s
-- Times: ['2.19s', '1.95s', '2.00s']
+- Average: 2.09s
+- Min: 1.64s
+- Max: 2.84s
+- Std Dev: 0.66s
+- Times: ['2.84s', '1.79s', '1.64s']
 
 ### agno_single
-- Average: 3.18s
-- Min: 3.00s
-- Max: 3.51s
-- Std Dev: 0.28s
-- Times: ['3.51s', '3.00s', '3.04s']
+- Average: 3.25s
+- Min: 3.14s
+- Max: 3.35s
+- Std Dev: 0.10s
+- Times: ['3.35s', '3.25s', '3.14s']
 
 ### praisonai_two_agents
-- Average: 2.49s
-- Min: 2.22s
-- Max: 2.65s
-- Std Dev: 0.24s
-- Times: ['2.65s', '2.22s', '2.62s']
+- Average: 2.50s
+- Min: 2.38s
+- Max: 2.75s
+- Std Dev: 0.21s
+- Times: ['2.38s', '2.75s', '2.38s']
 
 ### agno_two_agents
-- Average: 20.49s
-- Min: 9.96s
-- Max: 35.05s
-- Std Dev: 13.02s
-- Times: ['16.46s', '9.96s', '35.05s']
+- Average: 9.73s
+- Min: 9.34s
+- Max: 10.53s
+- Std Dev: 0.69s
+- Times: ['10.53s', '9.34s', '9.34s']
 
 ### praisonai_three_agents
-- Average: 6.52s
-- Min: 5.62s
-- Max: 7.39s
-- Std Dev: 0.88s
-- Times: ['5.62s', '6.54s', '7.39s']
+- Average: 6.25s
+- Min: 5.86s
+- Max: 6.80s
+- Std Dev: 0.49s
+- Times: ['6.80s', '6.09s', '5.86s']
 
 ### agno_three_agents
-- Average: 13.97s
-- Min: 11.69s
-- Max: 15.32s
-- Std Dev: 1.99s
-- Times: ['14.91s', '11.69s', '15.32s']
+- Average: 13.56s
+- Min: 12.26s
+- Max: 14.90s
+- Std Dev: 1.32s
+- Times: ['14.90s', '13.51s', '12.26s']
 
 ## Package Versions
 
 | Package | Version |
 |---------|--------|
-| PraisonAI | 0.1.5 |
-| Agno | 2.3.14 |
+| PraisonAI | 0.11.7 |
+| Agno | 2.3.25 |
 
 ## How to Reproduce
 

src/praisonai-agents/benchmarks/execution_benchmark.py

@@ -52,12 +52,11 @@ def benchmark_praisonai(model, iterations, prompt):
         agent = Agent(
             name="Calculator",
             instructions="You are a helpful assistant. Be very brief.",
-            llm=model,
-            verbose=False
+            llm=model
         )
         
         start = time.perf_counter()
-        response = agent.start(prompt)
+        response = agent.start(prompt, output="silent")
         elapsed = time.perf_counter() - start
         
         times.append(elapsed)
@@ -80,12 +79,11 @@ def benchmark_praisonai_litellm(model, iterations, prompt):
         agent = Agent(
             name="Calculator",
             instructions="You are a helpful assistant. Be very brief.",
-            llm=f"openai/{model}",
-            verbose=False
+            llm=f"openai/{model}"
         )
         
         start = time.perf_counter()
-        response = agent.start(prompt)
+        response = agent.start(prompt, output="silent")
         elapsed = time.perf_counter() - start
         
         times.append(elapsed)

src/praisonai-agents/benchmarks/real_benchmark.py

@@ -103,12 +103,11 @@ def benchmark_praisonai_single(iterations: int = 3) -> ExecutionResult:
             name="Assistant",
             instructions="You are a helpful assistant. Be very brief.",
             llm="gpt-4o-mini",
-            tools=[get_weather],
-            verbose=False
+            tools=[get_weather]
         )
         
         start = time.perf_counter()
-        response = agent.start("What's the weather in NYC? Reply in one sentence.")
+        response = agent.start("What's the weather in NYC? Reply in one sentence.", output="silent")
         elapsed = time.perf_counter() - start
         
         result.times.append(elapsed)
@@ -132,8 +131,7 @@ def benchmark_praisonai_two_agents(iterations: int = 3) -> ExecutionResult:
             goal="Gather weather information",
             instructions="Get weather data and pass to reporter. Be brief.",
             llm="gpt-4o-mini",
-            tools=[get_weather],
-            verbose=False
+            tools=[get_weather]
         )
         
         reporter = Agent(
@@ -141,18 +139,16 @@ def benchmark_praisonai_two_agents(iterations: int = 3) -> ExecutionResult:
             role="News Reporter",
             goal="Create weather report",
             instructions="Create a one-sentence weather summary.",
-            llm="gpt-4o-mini",
-            verbose=False
+            llm="gpt-4o-mini"
         )
         
         agents = Agents(
             agents=[researcher, reporter],
-            process="sequential",
-            verbose=False
+            process="sequential"
         )
         
         start = time.perf_counter()
-        response = agents.start("Get NYC weather and create a brief report.")
+        response = agents.start("Get NYC weather and create a brief report.", output="silent")
         elapsed = time.perf_counter() - start
         
         result.times.append(elapsed)
@@ -176,8 +172,7 @@ def benchmark_praisonai_three_agents(iterations: int = 3) -> ExecutionResult:
             goal="Get weather data",
             instructions="Fetch weather information. Be brief.",
             llm="gpt-4o-mini",
-            tools=[get_weather],
-            verbose=False
+            tools=[get_weather]
         )
         
         time_agent = Agent(
@@ -186,8 +181,7 @@ def benchmark_praisonai_three_agents(iterations: int = 3) -> ExecutionResult:
             goal="Get time information",
             instructions="Fetch timezone information. Be brief.",
             llm="gpt-4o-mini",
-            tools=[get_time],
-            verbose=False
+            tools=[get_time]
         )
         
         summarizer = Agent(
@@ -195,18 +189,16 @@ def benchmark_praisonai_three_agents(iterations: int = 3) -> ExecutionResult:
             role="Summary Writer",
             goal="Combine information",
             instructions="Create a one-sentence summary combining weather and time.",
-            llm="gpt-4o-mini",
-            verbose=False
+            llm="gpt-4o-mini"
         )
         
         agents = Agents(
             agents=[weather_agent, time_agent, summarizer],
-            process="sequential",
-            verbose=False
+            process="sequential"
         )
         
         start = time.perf_counter()
-        response = agents.start("Get NYC weather and EST time, then summarize.")
+        response = agents.start("Get NYC weather and EST time, then summarize.", output="silent")
         elapsed = time.perf_counter() - start
         
         result.times.append(elapsed)

src/praisonai-agents/gemini_self_reflection.py

@@ -14,7 +14,7 @@ llm_config = {
 agent = Agent(
     instructions="You are a helpful assistant. Be concise and clear.",
     llm=llm_config,
-    verbose=True,
+    output="verbose",
     self_reflect=True,
     max_reflect=2,
     min_reflect=1

src/praisonai-agents/guardrail_agent_example.py

@@ -7,8 +7,7 @@ def validate_content(data):
 
 agent = Agent(
     instructions="You are a writer",
-    guardrail=validate_content,
-    max_guardrail_retries=1
+    guardrails=validate_content
 )
 
 agent.start("Write a welcome message with 5 words")

src/praisonai-agents/llm-gemini-advanced.py

@@ -32,7 +32,7 @@ agent = Agent(
     instructions="You are a helpful Assistant specialized in scientific explanations. "
                 "Provide clear, accurate, and engaging responses.",
     llm=llm_config,                  # Pass the detailed configuration
-    verbose=True,                    # Enable detailed output
+    output="verbose",                    # Enable detailed output
     markdown=True,                   # Format responses in markdown
     self_reflect=True,              # Enable self-reflection
     max_reflect=3,                  # Maximum reflection iterations

src/praisonai-agents/mongodb_knowledge_basic.py

@@ -55,10 +55,8 @@ def main():
         retrieve information from a comprehensive MongoDB knowledge base. You excel 
         at finding relevant information, synthesizing knowledge from multiple sources, 
         and providing accurate, context-aware responses.""",
-        knowledge_config=mongodb_knowledge_config,
         knowledge=[os.path.join(os.path.dirname(__file__), "llms.md")],
         memory=True,
-        verbose=True,
         llm="gpt-5-nano"
     )
     
@@ -72,7 +70,6 @@ def main():
         knowledge base. You ensure information is accurate, well-structured, 
         and properly categorized.""",
         memory=True,
-        verbose=True,
         llm="gpt-5-nano"
     )
     
@@ -128,7 +125,7 @@ def main():
         agents=[research_agent, knowledge_agent],
         tasks=knowledge_tasks,
         memory=True,
-        verbose=True
+        output="verbose"
     )
     
     # Execute the knowledge management pipeline

src/praisonai-agents/multi-agents.py

@@ -0,0 +1,7 @@
+from praisonaiagents import Agent, Agents
+
+research_agent = Agent(instructions="Research about AI")
+summarise_agent = Agent(instructions="Summarise research agent's findings")
+
+agents = Agents(agents=[research_agent, summarise_agent])
+response = agents.start()

src/praisonai-agents/praisonaiagents/__init__.py

@@ -476,6 +476,62 @@ def _init_telemetry():
 _init_telemetry()
 
 
+def warmup(include_litellm: bool = False, include_openai: bool = True) -> dict:
+    """
+    Pre-import heavy dependencies to reduce first-call latency.
+    
+    NOTE: For default OpenAI usage (llm="gpt-4o-mini"), warmup is NOT needed.
+    The default path uses the native OpenAI SDK which is fast (~100ms import).
+    
+    Warmup is only beneficial when using LiteLLM backend, which is triggered by:
+    - Using "/" in model name (e.g., llm="openai/gpt-4o-mini")
+    - Passing a dict config (e.g., llm={"model": "gpt-4o-mini"})
+    - Using base_url parameter
+    
+    Args:
+        include_litellm: Pre-import LiteLLM (~2-3s). Only needed for multi-provider support.
+        include_openai: Pre-import OpenAI SDK (~100ms). Default path, usually fast.
+    
+    Returns:
+        dict: Timing information for each component warmed up
+    
+    Example:
+        # For LiteLLM multi-provider usage:
+        from praisonaiagents import warmup
+        warmup(include_litellm=True)  # Pre-load LiteLLM
+        
+        agent = Agent(llm="anthropic/claude-3-sonnet")  # Now faster
+        
+        # For default OpenAI usage, no warmup needed:
+        agent = Agent(llm="gpt-4o-mini")  # Already fast!
+    """
+    import time
+    timings = {}
+    
+    if include_openai:
+        start = time.perf_counter()
+        try:
+            import openai
+            timings['openai'] = (time.perf_counter() - start) * 1000
+        except ImportError:
+            timings['openai'] = -1  # Not available
+    
+    if include_litellm:
+        start = time.perf_counter()
+        try:
+            import litellm
+            # Also configure litellm to avoid first-call overhead
+            litellm.telemetry = False
+            litellm.set_verbose = False
+            litellm.drop_params = True
+            litellm.modify_params = True
+            timings['litellm'] = (time.perf_counter() - start) * 1000
+        except ImportError:
+            timings['litellm'] = -1  # Not available
+    
+    return timings
+
+
 __all__ = [
     # Core classes (lazy loaded)
     'Agent',
@@ -646,4 +702,19 @@ __all__ = [
     # Parse Utilities
     'is_policy_string',
     'parse_policy_string',
+    # Performance utilities
+    'warmup',
+    # Cortex - Continuous Improvement System
+    'CortexConfig',
+    'CaptureStore',
+    'CaptureMode',
+    'CaptureScope',
+    'PersonaConfig',
+    'InsightConfig',
+    'ThreadConfig',
+    'EntityConfig',
+    'PatternConfig',
+    'DecisionConfig',
+    'FeedbackConfig',
+    'ImprovementConfig',
 ]

src/praisonai-agents/praisonaiagents/agent/agent.py

@@ -44,7 +44,7 @@ _shared_apps = {}  # Dict of port -> FastAPI app
 if TYPE_CHECKING:
     from ..task.task import Task
     from ..main import TaskOutput
-    from ..handoff import Handoff
+    from .handoff import Handoff, HandoffConfig, HandoffResult
     from ..rag.models import RAGResult, ContextPack
 
 class Agent:
@@ -291,96 +291,87 @@ class Agent:
         """Initialize an Agent instance.
 
         Args:
-            name (Optional[str], optional): Name of the agent used for identification and logging.
-                If None, defaults to "Agent". Defaults to None.
-            role (Optional[str], optional): Role or job title that defines the agent's expertise
-                and behavior patterns. Examples: "Data Analyst", "Content Writer". Defaults to None.
-            goal (Optional[str], optional): Primary objective or goal the agent aims to achieve.
-                Defines the agent's purpose and success criteria. Defaults to None.
-            backstory (Optional[str], optional): Background story or context that shapes the agent's
-                personality and decision-making approach. Defaults to None.
-            instructions (Optional[str], optional): Direct instructions that override role, goal,
-                and backstory when provided. Used for simple, task-specific agents. Defaults to None.
-            llm (Optional[Union[str, Any]], optional): Language model configuration. Can be a model
-                name string (e.g., "gpt-5-nano", "anthropic/claude-3-sonnet") or a configured LLM object.
-                Defaults to environment variable OPENAI_MODEL_NAME or "gpt-5-nano".
-            tools (Optional[List[Any]], optional): List of tools, functions, or capabilities
-                available to the agent for task execution. Can include callables, tool objects,
-                or MCP instances. Defaults to None.
-            function_calling_llm (Optional[Any], optional): Dedicated language model for function
-                calling operations. If None, uses the main llm parameter. Defaults to None.
-            max_iter (int, optional): Maximum number of iterations the agent can perform during
-                task execution to prevent infinite loops. Defaults to 20.
-            max_rpm (Optional[int], optional): Maximum requests per minute to rate limit API calls
-                and prevent quota exhaustion. If None, no rate limiting is applied. Defaults to None.
-            max_execution_time (Optional[int], optional): Maximum execution time in seconds for
-                agent operations before timeout. If None, no time limit is enforced. Defaults to None.
-            memory (Optional[Any], optional): Memory system for storing and retrieving information
-                across conversations. Requires memory dependencies to be installed. Defaults to None.
-            verbose (bool, optional): Enable detailed logging and status updates during agent
-                execution for debugging and monitoring. Defaults to True.
-            allow_delegation (bool, optional): Allow the agent to delegate tasks to other agents
-                or sub-processes when appropriate. Defaults to False.
-            step_callback (Optional[Any], optional): Callback function called after each step
-                of agent execution for custom monitoring or intervention. Defaults to None.
-            cache (bool, optional): Enable caching of responses and computations to improve
-                performance and reduce API costs. Defaults to True.
-            system_template (Optional[str], optional): Custom template for system prompts that
-                overrides the default system prompt generation. Defaults to None.
-            prompt_template (Optional[str], optional): Template for formatting user prompts
-                before sending to the language model. Defaults to None.
-            response_template (Optional[str], optional): Template for formatting agent responses
-                before returning to the user. Defaults to None.
-            allow_code_execution (Optional[bool], optional): Enable the agent to execute code
-                snippets during task completion. Use with caution for security. Defaults to False.
-            max_retry_limit (int, optional): Maximum number of retry attempts for failed operations
-                before giving up. Helps handle transient errors. Defaults to 2.
-            code_execution_mode (Literal["safe", "unsafe"], optional): Safety mode for code execution.
-                "safe" restricts dangerous operations, "unsafe" allows full code execution. Defaults to "safe".
-            embedder_config (Optional[Dict[str, Any]], optional): Configuration dictionary for
-                text embedding models used in knowledge retrieval and similarity search. Defaults to None.
-            knowledge (Optional[List[str]], optional): List of knowledge sources (file paths, URLs,
-                or text content) to be processed and made available to the agent. Defaults to None.
-            knowledge_config (Optional[Dict[str, Any]], optional): Configuration for knowledge
-                processing and retrieval system including chunking and indexing parameters. Defaults to None.
-            use_system_prompt (Optional[bool], optional): Whether to include system prompts in
-                conversations to establish agent behavior and context. Defaults to True.
-            markdown (bool, optional): Enable markdown formatting in agent responses for better
-                readability and structure. Defaults to True.
-            stream (bool, optional): Enable streaming responses from the language model for real-time
-                output when using Agent.start() method. Defaults to False for backward compatibility.
-            metrics (bool, optional): Enable automatic token usage tracking and display summary
-                when tasks complete. Simplifies token monitoring for cost optimization. Defaults to False.
-            self_reflect (bool, optional): Enable self-reflection capabilities where the agent
-                evaluates and improves its own responses. Defaults to False.
-            max_reflect (int, optional): Maximum number of self-reflection iterations to prevent
-                excessive reflection loops. Defaults to 3.
-            min_reflect (int, optional): Minimum number of self-reflection iterations required
-                before accepting a response as satisfactory. Defaults to 1.
-            reflect_llm (Optional[str], optional): Dedicated language model for self-reflection
-                operations. If None, uses the main llm parameter. Defaults to None.
-            reflect_prompt (Optional[str], optional): Custom prompt template for self-reflection
-                that guides the agent's self-evaluation process. Defaults to None.
-            user_id (Optional[str], optional): Unique identifier for the user or session to
-                enable personalized responses and memory isolation. Defaults to "praison".
-            reasoning_steps (bool, optional): Enable step-by-step reasoning output to show the
-                agent's thought process during problem solving. Defaults to False.
-            guardrail (Optional[Union[Callable[['TaskOutput'], Tuple[bool, Any]], str]], optional):
-                Safety mechanism to validate agent outputs. Can be a validation function or
-                description string for LLM-based validation. Defaults to None.
-            max_guardrail_retries (int, optional): Maximum number of retry attempts when guardrail
-                validation fails before giving up. Defaults to 3.
-            handoffs (Optional[List[Union['Agent', 'Handoff']]], optional): List of agents or
-                handoff configurations that this agent can delegate tasks to. Enables agent-to-agent
-                collaboration and task specialization. Defaults to None.
-            base_url (Optional[str], optional): Base URL for custom LLM endpoints (e.g., Ollama).
-                If provided, automatically creates a custom LLM instance. Defaults to None.
-            api_key (Optional[str], optional): API key for LLM provider. If not provided,
-                falls back to environment variables. Defaults to None.
+            name: Agent name for identification and logging. Defaults to "Agent".
+            role: Role/job title defining expertise (e.g., "Data Analyst").
+            goal: Primary objective the agent aims to achieve.
+            backstory: Background context shaping personality and decisions.
+            instructions: Direct instructions (overrides role/goal/backstory). Recommended for simple agents.
+            llm: Model name string ("gpt-4o", "anthropic/claude-3-sonnet") or LLM object.
+                Defaults to OPENAI_MODEL_NAME env var or "gpt-4o-mini".
+            model: Alias for llm parameter.
+            function_calling_llm: Dedicated LLM for function calling. Deprecated: use llm=.
+            llm_config: LLM configuration dict. Deprecated: use llm=.
+            base_url: Custom LLM endpoint URL (e.g., for Ollama). Kept separate for auth.
+            api_key: API key for LLM provider. Kept separate for auth.
+            tools: List of tools, functions, callables, or MCP instances.
+            allow_delegation: Allow task delegation to other agents. Defaults to False.
+            allow_code_execution: Enable code execution during tasks. Defaults to False.
+            code_execution_mode: "safe" (restricted) or "unsafe" (full access). Defaults to "safe".
+            handoffs: List of Agent or Handoff objects for agent-to-agent collaboration.
+            auto_save: Session name for automatic session saving.
+            rate_limiter: Rate limiter instance for API call throttling.
+            memory: Memory system configuration. Accepts:
+                - bool: True enables defaults, False disables
+                - MemoryConfig: Custom configuration
+                - MemoryManager: Pre-configured instance
+            knowledge: Knowledge sources. Accepts:
+                - bool: True enables defaults
+                - List[str]: File paths, URLs, or text content
+                - KnowledgeConfig: Custom configuration
+            planning: Planning mode. Accepts:
+                - bool: True enables with defaults
+                - PlanningConfig: Custom configuration
+            reflection: Self-reflection. Accepts:
+                - bool: True enables with defaults
+                - ReflectionConfig: Custom configuration
+            guardrails: Output validation. Accepts:
+                - bool: True enables with defaults
+                - Callable: Validation function
+                - GuardrailConfig: Custom configuration
+            web: Web search/fetch. Accepts:
+                - bool: True enables with defaults
+                - WebConfig: Custom configuration
+            context: Context management. Accepts:
+                - bool: True enables with defaults
+                - ManagerConfig: Custom configuration
+            autonomy: Autonomy settings. Accepts:
+                - bool: True enables with defaults
+                - Dict: Configuration dict
+                - AutonomyConfig: Custom configuration
+            verification_hooks: List of VerificationHook instances for output verification.
+            output: Output configuration. Accepts:
+                - str: Preset name ("silent", "actions", "verbose", "json", "stream")
+                - OutputConfig: Custom configuration
+                Controls: verbose, markdown, stream, metrics, reasoning_steps
+            execution: Execution configuration. Accepts:
+                - str: Preset name ("fast", "balanced", "thorough")
+                - ExecutionConfig: Custom configuration
+                Controls: max_iter, max_rpm, max_execution_time, max_retry_limit
+            templates: Template configuration (TemplateConfig).
+                Controls: system_template, prompt_template, response_template
+            caching: Caching configuration. Accepts:
+                - bool: True enables with defaults
+                - CachingConfig: Custom configuration
+            hooks: Event hooks. Accepts:
+                - List: List of hook callables
+                - HooksConfig: Custom configuration
+            skills: Agent skills. Accepts:
+                - List[str]: Skill directory paths
+                - SkillsConfig: Custom configuration
 
         Raises:
             ValueError: If all of name, role, goal, backstory, and instructions are None.
             ImportError: If memory or LLM features are requested but dependencies are not installed.
+
+        Note:
+            Many legacy parameters have been consolidated into config objects:
+            - verbose, markdown, stream, metrics, reasoning_steps → output=
+            - max_iter, max_rpm, max_execution_time, max_retry_limit → execution=
+            - self_reflect, max_reflect, min_reflect, reflect_llm → reflection=
+            - guardrail, max_guardrail_retries → guardrails=
+            - system_template, prompt_template, response_template → templates=
+            - cache, prompt_caching → caching=
+            - web_search, web_fetch → web=
         """
         # Add check at start if memory is requested
         if memory is not None:
@@ -447,14 +438,27 @@ class Agent:
         # ─────────────────────────────────────────────────────────────────────
         # Resolve OUTPUT param using unified resolver
         # Supports: None, str preset, list [preset, overrides], OutputConfig
+        # DEFAULT: "silent" mode (zero overhead, fastest performance)
+        # Can be overridden by PRAISONAI_OUTPUT env var
         # ─────────────────────────────────────────────────────────────────────
+        from ..config.presets import DEFAULT_OUTPUT_MODE
+        
+        # Check for env var override
+        env_output = os.environ.get('PRAISONAI_OUTPUT', '').lower()
+        if env_output and env_output in OUTPUT_PRESETS and output is None:
+            output = env_output
+        
+        # Use default output mode if not specified
+        if output is None:
+            output = DEFAULT_OUTPUT_MODE
+        
         _output_config = resolve(
             value=output,
             param_name="output",
             config_class=OutputConfig,
             presets=OUTPUT_PRESETS,
             array_mode=ArrayMode.PRESET_OVERRIDE,
-            default=OutputConfig(),
+            default=OutputConfig(),  # OutputConfig defaults to silent mode (zero overhead)
         )
         if _output_config:
             verbose = _output_config.verbose
@@ -463,8 +467,24 @@ class Agent:
             metrics = _output_config.metrics
             reasoning_steps = _output_config.reasoning_steps
             output_style = getattr(_output_config, 'style', None)
+            actions_trace = getattr(_output_config, 'actions_trace', False)  # Default False (silent)
+            json_output = getattr(_output_config, 'json_output', False)
         else:
-            verbose, markdown, stream, metrics, reasoning_steps = True, True, False, False, False
+            # Fallback defaults match silent mode (zero overhead)
+            verbose, markdown, stream, metrics, reasoning_steps = False, False, False, False, False
+            actions_trace = False  # No callbacks by default
+            json_output = False
+        
+        # Enable actions output mode if configured
+        # This registers callbacks to capture tool calls and final output
+        if actions_trace:
+            try:
+                from ..output.actions import enable_actions_mode, is_actions_mode_enabled
+                if not is_actions_mode_enabled():
+                    output_format = "jsonl" if json_output else "text"
+                    enable_actions_mode(redact=True, use_color=True, format=output_format)
+            except ImportError:
+                pass  # Actions module not available
         
         # ─────────────────────────────────────────────────────────────────────
         # Resolve EXECUTION param using unified resolver
@@ -755,7 +775,21 @@ class Agent:
 
         # If instructions are provided, use them to set role, goal, and backstory
         if instructions:
-            self.name = name or "Agent"
+            # Auto-generate descriptive name from instructions if not provided
+            if name:
+                self.name = name
+            else:
+                # Generate name from first 2 meaningful words of instructions
+                # "Research about AI" → "Research Agent"
+                # "Summarise research agent's findings" → "Summarise Agent"
+                words = instructions.split()
+                # Filter out common stop words
+                stop_words = {'the', 'a', 'an', 'and', 'or', 'but', 'in', 'on', 'at', 'to', 'for', 'of', 'with', 'by', 'about'}
+                meaningful = [w for w in words if w.lower() not in stop_words][:2]
+                if meaningful:
+                    self.name = " ".join(meaningful).title() + " Agent"
+                else:
+                    self.name = words[0].title() + " Agent" if words else "Agent"
             self.role = role or "Assistant"
             self.goal = goal or instructions
             self.backstory = backstory or instructions
@@ -824,7 +858,7 @@ class Agent:
                     self.llm_instance = LLM(**llm_config)
                 else:
                     # Create LLM with model string and base_url
-                    model_name = llm or os.getenv('OPENAI_MODEL_NAME', 'gpt-5-nano')
+                    model_name = llm or os.getenv('OPENAI_MODEL_NAME', 'gpt-4o-mini')
                     self.llm_instance = LLM(
                         model=model_name,
                         base_url=base_url,
@@ -887,7 +921,7 @@ class Agent:
                 ) from e
         # Otherwise, fall back to OpenAI environment/name
         else:
-            self.llm = llm or os.getenv('OPENAI_MODEL_NAME', 'gpt-5-nano')
+            self.llm = llm or os.getenv('OPENAI_MODEL_NAME', 'gpt-4o-mini')
         # Handle tools parameter - ensure it's always a list
         if callable(tools):
             # If a single function/callable is passed, wrap it in a list
@@ -933,7 +967,7 @@ class Agent:
         self.min_reflect = min_reflect
         self.reflect_prompt = reflect_prompt
         # Use the same model selection logic for reflect_llm
-        self.reflect_llm = reflect_llm or os.getenv('OPENAI_MODEL_NAME', 'gpt-5-nano')
+        self.reflect_llm = reflect_llm or os.getenv('OPENAI_MODEL_NAME', 'gpt-4o-mini')
         self._console = None  # Lazy load console when needed
         
         # Initialize system prompt
@@ -1065,6 +1099,9 @@ Your Goal: {self.goal}
         self._context_param = context  # Store raw param for lazy init
         self._context_manager = None  # Lazy initialized on first use
         self._context_manager_initialized = False
+        
+        # Action trace mode - handled via display callbacks, not separate emitter
+        self._actions_trace = actions_trace
 
     @property
     def auto_memory(self):
@@ -1720,7 +1757,7 @@ Your Goal: {self.goal}
         elif hasattr(self, 'llm') and self.llm:
             model_name = self.llm
         else:
-            model_name = "gpt-5-nano"
+            model_name = "gpt-4o-mini"
         
         return supports_web_search(model_name)
     
@@ -1742,7 +1779,7 @@ Your Goal: {self.goal}
         elif hasattr(self, 'llm') and self.llm:
             model_name = self.llm
         else:
-            model_name = "gpt-5-nano"
+            model_name = "gpt-4o-mini"
         
         return supports_web_fetch(model_name)
     
@@ -1986,7 +2023,7 @@ Your Goal: {self.goal}
         
         Returns:
             The LLM model/instance being used by this agent.
-            - For standard models: returns the model string (e.g., "gpt-5-nano")
+            - For standard models: returns the model string (e.g., "gpt-4o-mini")
             - For custom LLM instances: returns the LLM instance object
             - For provider models: returns the LLM instance object
         """
@@ -1996,7 +2033,7 @@ Your Goal: {self.goal}
             return self.llm
         else:
             # Default fallback
-            return "gpt-5-nano"
+            return "gpt-4o-mini"
 
     def _ensure_knowledge_processed(self):
         """Ensure knowledge is initialized and processed when first accessed."""
@@ -2817,6 +2854,10 @@ Your Goal: {self.goal}"""
         """
         logging.debug(f"{self.name} executing tool {function_name} with arguments: {arguments}")
         
+        # Notify callbacks that tool is being called (for dynamic status updates)
+        from ..main import execute_sync_callback
+        execute_sync_callback('tool_call', tool_name=function_name, tool_input=arguments)
+        
         # Set up injection context for tools with Injected parameters
         from ..tools.injected import AgentState, with_injection_context
         state = AgentState(
@@ -3696,7 +3737,7 @@ Your Goal: {self.goal}"""
                             if self.verbose:
                                 logging.debug(f"Agent {self.name} final response: {response_text}")
                             # Return only reasoning content if reasoning_steps is True
-                            if reasoning_steps and hasattr(response.choices[0].message, 'reasoning_content'):
+                            if reasoning_steps and hasattr(response.choices[0].message, 'reasoning_content') and response.choices[0].message.reasoning_content:
                                 # Apply guardrail to reasoning content
                                 try:
                                     validated_reasoning = self._apply_guardrail_with_retry(response.choices[0].message.reasoning_content, original_prompt, temperature, tools, task_name, task_description, task_id)
@@ -4221,7 +4262,7 @@ Output MUST be JSON with 'reflection' and 'satisfactory'.
                         
                         final_response = process_stream_chunks(chunks)
                         # Return only reasoning content if reasoning_steps is True
-                        if reasoning_steps and hasattr(final_response.choices[0].message, 'reasoning_content'):
+                        if reasoning_steps and hasattr(final_response.choices[0].message, 'reasoning_content') and final_response.choices[0].message.reasoning_content:
                             return final_response.choices[0].message.reasoning_content
                         return final_response.choices[0].message.content if final_response else full_response_text
 
@@ -4235,16 +4276,94 @@ Output MUST be JSON with 'reflection' and 'satisfactory'.
             return None
 
     async def arun(self, prompt: str, **kwargs):
-        """Async alias for astart() method"""
-        return await self.astart(prompt, **kwargs)
+        """Async version of run() - silent, non-streaming, returns structured result.
+        
+        Production-friendly async execution. Does not stream or display output.
+        
+        Args:
+            prompt: The input prompt to process
+            **kwargs: Additional arguments passed to achat()
+            
+        Returns:
+            The agent's response as a string
+        """
+        # Force non-streaming, non-display for production use
+        kwargs['stream'] = False
+        return await self.achat(prompt, **kwargs)
 
     async def astart(self, prompt: str, **kwargs):
-        """Async version of start method"""
+        """Async version of start() - interactive, streaming-aware.
+        
+        Beginner-friendly async execution. Streams by default when in TTY.
+        
+        Args:
+            prompt: The input prompt to process
+            **kwargs: Additional arguments passed to achat()
+            
+        Returns:
+            The agent's response as a string
+        """
+        import sys
+        
+        # Determine streaming behavior (same logic as start())
+        stream_requested = kwargs.get('stream')
+        if stream_requested is None:
+            if getattr(self, 'stream', None) is not None:
+                stream_requested = self.stream
+            else:
+                stream_requested = sys.stdout.isatty()
+        
+        kwargs['stream'] = stream_requested
         return await self.achat(prompt, **kwargs)
 
     def run(self, prompt: str, **kwargs):
-        """Alias for start() method"""
-        return self.start(prompt, **kwargs)
+        """Execute agent silently and return structured result.
+        
+        Production-friendly execution. Always uses silent mode with no streaming
+        or verbose display, regardless of TTY status. Use this for programmatic,
+        scripted, or automated usage where you want just the result.
+        
+        Args:
+            prompt: The input prompt to process
+            **kwargs: Additional arguments:
+                - stream (bool): Force streaming if True. Default: False
+                - output (str): Output preset override (rarely needed)
+                
+        Returns:
+            The agent's response as a string
+            
+        Example:
+            ```python
+            agent = Agent(instructions="You are helpful")
+            result = agent.run("What is 2+2?")  # Silent, returns "4"
+            print(result)
+            ```
+            
+        Note:
+            Unlike .start() which enables verbose output in TTY for interactive
+            use, .run() is always silent. This makes it suitable for:
+            - Production pipelines
+            - Automated scripts
+            - Background processing
+            - API endpoints
+        """
+        # Production defaults: no streaming, no display
+        if 'stream' not in kwargs:
+            kwargs['stream'] = False
+        
+        # Load history context
+        self._load_history_context()
+        
+        # Check if planning mode is enabled
+        if self.planning:
+            result = self._start_with_planning(prompt, **kwargs)
+        else:
+            result = self.chat(prompt, **kwargs)
+        
+        # Auto-save session if enabled
+        self._auto_save_session()
+        
+        return result
     
     def _get_planning_agent(self):
         """Lazy load PlanningAgent for planning mode."""
@@ -4378,26 +4497,290 @@ Write the complete compiled report:"""
         
         # Chat history is preserved in self.chat_history (no action needed)
 
-    def start(self, prompt: str, **kwargs):
-        """Start the agent with a prompt. This is a convenience method that wraps chat()."""
-        # Load history from past sessions (now handled via context= param)
+    def start(self, prompt: str = None, **kwargs):
+        """Start the agent interactively with verbose output.
+        
+        Beginner-friendly execution. Defaults to verbose output with streaming
+        when running in a TTY. Use this for interactive/terminal usage where 
+        you want to see output in real-time with rich formatting.
+        
+        Args:
+            prompt: The input prompt to process. If not provided, uses the 
+                    agent's instructions as the task (useful when instructions
+                    already describe what the agent should do).
+            **kwargs: Additional arguments:
+                - stream (bool | None): Override streaming. None = auto-detect TTY
+                - output (str): Output preset override (e.g., "silent", "verbose")
+                
+        Returns:
+            - If streaming: Generator yielding response chunks
+            - If not streaming: The complete response as a string
+            
+        Example:
+            ```python
+            # Minimal usage - instructions IS the task
+            agent = Agent(instructions="Research AI trends and summarize")
+            result = agent.start()  # Uses instructions as task
+            
+            # With explicit prompt (overrides/adds to instructions)
+            agent = Agent(instructions="You are a helpful assistant")
+            result = agent.start("What is 2+2?")  # Uses prompt as task
+            ```
+            
+        Note:
+            Unlike .run() which is always silent (production use), .start()
+            enables verbose output by default when in a TTY for beginner-friendly
+            interactive use. Use .run() for programmatic/scripted usage.
+        """
+        import sys
+        
+        # If no prompt provided, use instructions as the task
+        if prompt is None:
+            prompt = self.instructions or "Hello"
+        
+        # Load history from past sessions
         self._load_history_context()
         
-        # Check if planning mode is enabled
-        if self.planning:
-            result = self._start_with_planning(prompt, **kwargs)
-        elif kwargs.get('stream', getattr(self, 'stream', False)):
-            # Return a generator for streaming response
-            result = self._start_stream(prompt, **kwargs)
-        else:
-            # Return regular chat response for backward compatibility
-            kwargs['stream'] = False
-            result = self.chat(prompt, **kwargs)
+        # Determine if we're in an interactive TTY
+        is_tty = sys.stdout.isatty()
+        
+        # Determine streaming behavior
+        # Priority: explicit kwarg > agent's stream attribute > TTY detection
+        stream_requested = kwargs.get('stream')
+        if stream_requested is None:
+            # Check agent's stream attribute first
+            if getattr(self, 'stream', None) is not None:
+                stream_requested = self.stream
+            else:
+                # Auto-detect: stream if stdout is a TTY (interactive terminal)
+                stream_requested = is_tty
+        
+        # ─────────────────────────────────────────────────────────────────────
+        # Enable verbose output in TTY for beginner-friendly interactive use
+        # Priority: explicit output= kwarg > TTY auto-verbose > agent's default
+        # ─────────────────────────────────────────────────────────────────────
+        original_verbose = self.verbose
+        original_markdown = self.markdown
+        output_override = kwargs.pop('output', None)  # Pop to prevent passing to chat()
+        
+        try:
+            # If running in TTY and no explicit output override, enable verbose
+            if is_tty and output_override is None:
+                self.verbose = True
+                self.markdown = True
+            elif output_override:
+                # Apply explicit output preset for this call
+                from ..config.presets import OUTPUT_PRESETS
+                if output_override in OUTPUT_PRESETS:
+                    preset = OUTPUT_PRESETS[output_override]
+                    self.verbose = preset.get('verbose', False)
+                    self.markdown = preset.get('markdown', False)
+            
+            # Check if planning mode is enabled
+            if self.planning:
+                result = self._start_with_planning(prompt, **kwargs)
+            elif stream_requested:
+                # Return a generator for streaming response
+                kwargs['stream'] = True
+                result = self._start_stream(prompt, **kwargs)
+            else:
+                # Return regular chat response with animated working status
+                kwargs['stream'] = False
+                
+                # Show animated status during LLM call if verbose
+                if self.verbose and is_tty:
+                    from rich.live import Live
+                    from rich.console import Group
+                    from rich.status import Status
+                    from ..main import PRAISON_COLORS, sync_display_callbacks
+                    import threading
+                    import time as time_module
+                    
+                    console = Console()
+                    start_time = time_module.time()
+                    
+                    # ─────────────────────────────────────────────────────────────
+                    # Shared state for dynamic status messages (thread-safe)
+                    # Updated by callbacks during tool execution
+                    # ─────────────────────────────────────────────────────────────
+                    current_status = ["Analyzing query..."]
+                    tools_called = []
+                    
+                    # Register a temporary callback to track tool calls
+                    def status_tool_callback(**kwargs):
+                        tool_name = kwargs.get('tool_name', '')
+                        if tool_name:
+                            tools_called.append(tool_name)
+                            current_status[0] = f"Calling tool: {tool_name}..."
+                    
+                    # Store original callback and register ours
+                    original_tool_callback = sync_display_callbacks.get('tool_call')
+                    sync_display_callbacks['tool_call'] = status_tool_callback
+                    
+                    # Animation state
+                    result_holder = [None]
+                    error_holder = [None]
+                    
+                    # Temporarily disable verbose in chat to prevent duplicate output
+                    original_verbose_chat = self.verbose
+                    
+                    def run_chat():
+                        try:
+                            # Suppress verbose during animation - we'll display result ourselves
+                            self.verbose = False
+                            current_status[0] = "Sending to LLM..."
+                            result_holder[0] = self.chat(prompt, **kwargs)
+                            current_status[0] = "Finalizing response..."
+                        except Exception as e:
+                            error_holder[0] = e
+                        finally:
+                            self.verbose = original_verbose_chat
+                            # Restore original callback
+                            if original_tool_callback:
+                                sync_display_callbacks['tool_call'] = original_tool_callback
+                            elif 'tool_call' in sync_display_callbacks:
+                                del sync_display_callbacks['tool_call']
+                    
+                    # Start chat in background thread
+                    chat_thread = threading.Thread(target=run_chat)
+                    chat_thread.start()
+                    
+                    from rich.panel import Panel
+                    from rich.text import Text
+                    from rich.markdown import Markdown
+                    
+                    # ─────────────────────────────────────────────────────────────
+                    # Smart Agent Info: Only show if user provided meaningful info
+                    # Skip if using defaults ("Agent"/"Assistant") with single agent
+                    # ─────────────────────────────────────────────────────────────
+                    has_custom_name = self.name and self.name not in ("Agent", "agent", None, "")
+                    has_custom_role = self.role and self.role not in ("Assistant", "AI Assistant", "assistant", None, "")
+                    has_tools = bool(self.tools)
+                    
+                    show_agent_info = has_custom_name or has_custom_role or has_tools
+                    
+                    agent_panel = None
+                    if show_agent_info:
+                        agent_info_parts = []
+                        if has_custom_name:
+                            agent_info_parts.append(f"[bold {PRAISON_COLORS['task']}]👤 Agent:[/] [{PRAISON_COLORS['agent_text']}]{self.name}[/]")
+                        if has_custom_role:
+                            agent_info_parts.append(f"[bold {PRAISON_COLORS['metrics']}]Role:[/] [{PRAISON_COLORS['agent_text']}]{self.role}[/]")
+                        if has_tools:
+                            tools_list = [t.__name__ if hasattr(t, '__name__') else str(t) for t in self.tools][:5]
+                            tools_str = ", ".join(f"[italic {PRAISON_COLORS['response']}]{tool}[/]" for tool in tools_list)
+                            agent_info_parts.append(f"[bold {PRAISON_COLORS['agent']}]Tools:[/] {tools_str}")
+                        
+                        agent_panel = Panel(
+                            "\n".join(agent_info_parts), 
+                            border_style=PRAISON_COLORS["agent"], 
+                            title="[bold]Agent Info[/]", 
+                            title_align="left", 
+                            padding=(1, 2)
+                        )
+                    
+                    # Create task panel
+                    task_panel = Panel.fit(
+                        Markdown(prompt) if self.markdown else Text(prompt),
+                        title="Task",
+                        border_style=PRAISON_COLORS["task"]
+                    )
+                    
+                    # Show initial panels (agent info if applicable, then task)
+                    if agent_panel:
+                        console.print(agent_panel)
+                    console.print(task_panel)
+                    
+                    # ─────────────────────────────────────────────────────────────
+                    # Animate with Rich.Status showing DYNAMIC status from callbacks
+                    # ─────────────────────────────────────────────────────────────
+                    with console.status(
+                        f"[bold yellow]Working...[/]  {current_status[0]}", 
+                        spinner="dots",
+                        spinner_style="yellow"
+                    ) as status:
+                        last_status = current_status[0]
+                        while chat_thread.is_alive():
+                            time_module.sleep(0.15)  # More responsive updates
+                            # Only update if status changed (reduces flicker)
+                            if current_status[0] != last_status:
+                                last_status = current_status[0]
+                            status.update(f"[bold yellow]Working...[/]  {current_status[0]}")
+                    
+                    # Calculate elapsed time
+                    elapsed = time_module.time() - start_time
+                    
+                    # Re-raise any error from chat
+                    if error_holder[0]:
+                        raise error_holder[0]
+                    
+                    result = result_holder[0]
+                    
+                    # Display response panel
+                    response_panel = Panel.fit(
+                        Markdown(str(result)) if self.markdown else Text(str(result)),
+                        title=f"Response ({elapsed:.1f}s)",
+                        border_style=PRAISON_COLORS["response"]
+                    )
+                    console.print(response_panel)
+                    
+                    # Show tool activity summary if tools were called
+                    if tools_called:
+                        tools_summary = ", ".join(tools_called)
+                        console.print(f"[dim]🔧 Tools used: {tools_summary}[/dim]")
+                else:
+                    result = self.chat(prompt, **kwargs)
+            
+            # Auto-save session if enabled
+            self._auto_save_session()
+            
+            return result
+        finally:
+            # Restore original output settings
+            self.verbose = original_verbose
+            self.markdown = original_markdown
+    
+    def iter_stream(self, prompt: str, **kwargs):
+        """Stream agent response as an iterator of chunks.
+        
+        App-friendly streaming. Yields response chunks without terminal display.
+        Use this for building custom UIs or processing streams programmatically.
+        
+        Args:
+            prompt: The input prompt to process
+            **kwargs: Additional arguments:
+                - display (bool): Show terminal output. Default: False
+                - output (str): Output preset override
+                
+        Yields:
+            str: Response chunks as they are generated
+            
+        Example:
+            ```python
+            agent = Agent(instructions="You are helpful")
+            
+            # Process stream programmatically
+            full_response = ""
+            for chunk in agent.iter_stream("Tell me a story"):
+                full_response += chunk
+                # Custom processing here
+            
+            # Or collect all at once
+            response = "".join(agent.iter_stream("Hello"))
+            ```
+        """
+        # Load history context
+        self._load_history_context()
+        
+        # Force streaming, no display by default (app-friendly)
+        kwargs['stream'] = True
+        
+        # Use the internal streaming generator
+        for chunk in self._start_stream(prompt, **kwargs):
+            yield chunk
         
         # Auto-save session if enabled
         self._auto_save_session()
-        
-        return result
     
     def _load_history_context(self):
         """Load history from past sessions into context.

src/praisonai-agents/praisonaiagents/agent/context_agent.py

@@ -434,8 +434,7 @@ This report contains all agent interactions and outputs from a complete ContextA
                     
                     Provide comprehensive analysis that follows the PRD template principles and enables 
                     AI assistants to implement features that perfectly match existing codebase patterns.""",
-                    llm=self.llm if hasattr(self, 'llm') else "gpt-5-nano",
-                    verbose=getattr(self, 'verbose', True)
+                    llm=self.llm if hasattr(self, 'llm') else "gpt-4o-mini",
                 )
                 
                 prompt = f"""Analyze this gitingest codebase digest following PRD template methodology:
@@ -515,8 +514,7 @@ codebase style and architecture following PRD template principles."""
             role="Expert Manual Codebase Analysis Specialist",
             goal="Perform comprehensive manual codebase analysis following PRD methodology",
             instructions="""Analyze the codebase samples following PRD template methodology for complete understanding.""",
-            llm=self.llm if hasattr(self, 'llm') else "gpt-5-nano",
-            verbose=getattr(self, 'verbose', True)
+            llm=self.llm if hasattr(self, 'llm') else "gpt-4o-mini",
         )
         
         # Format comprehensive sample
@@ -566,8 +564,7 @@ Analyze following PRD principles to extract patterns, conventions, and architect
             6. Design pattern implementations
             7. Code complexity metrics
             8. API and interface patterns""",
-            llm=self.llm if hasattr(self, 'llm') else "gpt-5-nano",
-            verbose=getattr(self, 'verbose', True)
+            llm=self.llm if hasattr(self, 'llm') else "gpt-4o-mini",
         )
         
         # Perform AST analysis on Python files
@@ -624,8 +621,7 @@ Extract comprehensive patterns that follow PRD template principles for implement
             
             For each pattern, provide the pattern name, where it's used, and how to replicate it 
             following PRD template principles.""",
-            llm=self.llm if hasattr(self, 'llm') else "gpt-5-nano",
-            verbose=getattr(self, 'verbose', True)
+            llm=self.llm if hasattr(self, 'llm') else "gpt-4o-mini",
         )
         
         # Get representative code samples
@@ -674,8 +670,7 @@ patterns and best practices for first-try success."""
             goal="Analyze testing patterns for comprehensive validation framework design",
             instructions="""Analyze testing patterns to understand validation approaches and create 
             comprehensive test frameworks following PRD methodology.""",
-            llm=self.llm if hasattr(self, 'llm') else "gpt-5-nano",
-            verbose=getattr(self, 'verbose', True)
+            llm=self.llm if hasattr(self, 'llm') else "gpt-4o-mini",
         )
         
         # Get test files
@@ -801,8 +796,7 @@ Extract testing patterns for validation framework creation following PRD princip
             Confidence level for one-pass implementation
             
             Generate PRPs following this EXACT structure for first-try implementation success.""",
-            llm=self.llm if hasattr(self, 'llm') else "gpt-5-nano",
-            verbose=getattr(self, 'verbose', True)
+            llm=self.llm if hasattr(self, 'llm') else "gpt-4o-mini",
         )
         
         prompt = f"""Generate a comprehensive Product Requirements Prompt (PRP) following the EXACT PRD template structure:
@@ -850,8 +844,7 @@ on the first try following PRD template principles."""
             6. CODE QUALITY: Complexity analysis, maintainability
             7. DOCUMENTATION VALIDATION: Documentation completeness
             8. DEPENDENCY VALIDATION: Dependency analysis and security""",
-            llm=self.llm if hasattr(self, 'llm') else "gpt-5-nano",
-            verbose=getattr(self, 'verbose', True)
+            llm=self.llm if hasattr(self, 'llm') else "gpt-4o-mini",
         )
         
         # Analyze existing validation patterns
@@ -897,8 +890,7 @@ following PRD template principles."""
             instructions="""Compile all available documentation following PRD methodology including:
             README files, API documentation, setup guides, architecture docs, and any other 
             relevant documentation that provides context for implementation.""",
-            llm=self.llm if hasattr(self, 'llm') else "gpt-5-nano",
-            verbose=getattr(self, 'verbose', True)
+            llm=self.llm if hasattr(self, 'llm') else "gpt-4o-mini",
         )
         
         # Get documentation files
@@ -943,8 +935,7 @@ following PRD template principles."""
             instructions="""Analyze integration points following PRD methodology including:
             APIs, databases, external services, configuration points, and any other 
             integration requirements that affect implementation.""",
-            llm=self.llm if hasattr(self, 'llm') else "gpt-5-nano",
-            verbose=getattr(self, 'verbose', True)
+            llm=self.llm if hasattr(self, 'llm') else "gpt-4o-mini",
         )
         
         # Get configuration and dependency files
@@ -1004,8 +995,7 @@ following PRD template principles."""
             8. DOCUMENTATION UPDATES: Documentation to create/update
             9. INTEGRATION STEPS: How to integrate with existing systems
             10. VALIDATION CHECKPOINTS: Validation steps at each phase""",
-            llm=self.llm if hasattr(self, 'llm') else "gpt-5-nano",
-            verbose=getattr(self, 'verbose', True)
+            llm=self.llm if hasattr(self, 'llm') else "gpt-4o-mini",
         )
         
         prompt = f"""Create detailed implementation blueprint following PRD methodology:
@@ -1302,8 +1292,7 @@ Every agent interaction has been saved for full audit trail and reproducibility.
             GOAL: [extracted implementation goal]
             
             Be precise and extract only what is explicitly mentioned or clearly implied.""",
-            llm=self.llm if hasattr(self, 'llm') else "gpt-5-nano",
-            verbose=getattr(self, 'verbose', True)
+            llm=self.llm if hasattr(self, 'llm') else "gpt-4o-mini",
         )
         
         prompt = f"""Parse this user input to extract codebase and goal:
@@ -1536,8 +1525,7 @@ Note: Detailed function/class metadata not available due to content access limit
             5. Documentation topics
             
             Make the output easy for a file selection agent to understand which files contain what functionality.""",
-            llm=self.llm if hasattr(self, 'llm') else "gpt-5-nano",
-            verbose=getattr(self, 'verbose', True)
+            llm=self.llm if hasattr(self, 'llm') else "gpt-4o-mini",
         )
         
         structure_text = basic_structure.get('structure', '')
@@ -1770,8 +1758,7 @@ Focus on creating clear, structured metadata that will help with intelligent fil
             ["README.md", "src/auth/login.py", "config/settings.py", ...]
             
             Maximum 50 files for efficient analysis.""",
-            llm=self.llm if hasattr(self, 'llm') else "gpt-5-nano",
-            verbose=getattr(self, 'verbose', True)
+            llm=self.llm if hasattr(self, 'llm') else "gpt-4o-mini",
         )
         
         # Prepare enhanced structure information
@@ -1923,8 +1910,7 @@ Maximum 50 files.""".format(goal=goal)
             8. EXAMPLES: Similar features that can guide {goal} implementation
             
             Since these files were pre-selected for relevance, provide deep analysis of how each contributes to implementing: {goal}""",
-            llm=self.llm if hasattr(self, 'llm') else "gpt-5-nano",
-            verbose=getattr(self, 'verbose', True)
+            llm=self.llm if hasattr(self, 'llm') else "gpt-4o-mini",
         )
         
         prompt = f"""Analyze these carefully selected repository files for implementing: {goal}
@@ -2024,8 +2010,7 @@ Since these files were pre-selected for relevance, explain how each contributes
             - Success criteria for {goal}
             
             Focus everything on successfully implementing: {goal}""",
-            llm=self.llm if hasattr(self, 'llm') else "gpt-5-nano",
-            verbose=getattr(self, 'verbose', True)
+            llm=self.llm if hasattr(self, 'llm') else "gpt-4o-mini",
         )
         
         chunk_info = ""
@@ -2303,13 +2288,13 @@ def create_context_agent(llm: Optional[Union[str, Any]] = None, **kwargs) -> Con
     Factory function to create a ContextAgent following Context Engineering and PRD methodology.
     
     Args:
-        llm: Language model to use (e.g., "gpt-5-nano", "claude-3-haiku")
+        llm: Language model to use (e.g., "gpt-4o-mini", "claude-3-haiku")
         **kwargs: Additional arguments to pass to ContextAgent constructor
         
     Returns:
         ContextAgent: Configured ContextAgent for comprehensive context generation following PRD principles
     """
     if llm is None:
-        llm = "gpt-5-nano"
+        llm = "gpt-4o-mini"
     
     return ContextAgent(llm=llm, **kwargs)

src/praisonai-agents/praisonaiagents/agent/router_agent.py

@@ -44,7 +44,7 @@ class RouterAgent(Agent):
         # Initialize model router
         self.model_router = model_router or ModelRouter()
         self.routing_strategy = routing_strategy
-        self.fallback_model = fallback_model or os.getenv('OPENAI_MODEL_NAME', 'gpt-5-nano')
+        self.fallback_model = fallback_model or os.getenv('OPENAI_MODEL_NAME', 'gpt-4o-mini')
         
         # Process models configuration
         self.available_models = self._process_models_config(models)

src/praisonai-agents/praisonaiagents/agents/agents.py

@@ -363,7 +363,7 @@ class Agents:
         self.variables = variables if variables else {}
         
         # Check for manager_llm in environment variable if not provided
-        self.manager_llm = manager_llm or os.getenv('OPENAI_MODEL_NAME', 'gpt-5-nano')
+        self.manager_llm = manager_llm or os.getenv('OPENAI_MODEL_NAME', 'gpt-4o-mini')
         
         # Set logger level based on verbose
         if _verbose >= 5:
@@ -1193,30 +1193,143 @@ Context:
             return str(agent[0])
         return None
 
-    def start(self, content=None, return_dict=False, **kwargs):
-        """Start agent execution with optional content and config
+    def start(self, content=None, return_dict=False, output=None, **kwargs):
+        """Start agent execution with verbose output (beginner-friendly).
+        
+        Shows Rich panels with workflow progress when in TTY. Use .run() for
+        silent execution in production/scripts.
         
         Args:
             content: Optional content to add to all tasks' context
-            return_dict: If True, returns the full results dictionary instead of only the final response
+            return_dict: If True, returns the full results dictionary
+            output: Output preset - "silent", "verbose", "normal", etc.
+                    Default in TTY: "verbose" (shows progress)
+                    Default non-TTY: "silent"
             **kwargs: Additional arguments
+            
+        Example:
+            ```python
+            # Interactive - shows Rich panels
+            agents = Agents(agents=[agent1, agent2])
+            result = agents.start()  # Verbose output by default
+            
+            # Force silent mode
+            result = agents.start(output="silent")
+            ```
         """
+        import sys
+        from ..main import PRAISON_COLORS
+        
+        # Determine if we're in an interactive TTY
+        is_tty = sys.stdout.isatty()
+        
+        # Resolve output mode (TTY-aware)
+        if output is None:
+            # Default: verbose in TTY (beginner-friendly), silent otherwise
+            # Note: Don't check self.verbose here - start() is for interactive use
+            show_verbose = is_tty
+        elif output == "silent":
+            show_verbose = False
+        elif output in ("verbose", "debug", "normal"):
+            show_verbose = True
+        else:
+            show_verbose = is_tty
+        
+        # Add content to context if provided
         if content:
-            # Add content to context of all tasks
             for task in self.tasks.values():
                 if isinstance(content, (str, list)):
-                    # If context is empty, initialize it
                     if not task.context:
                         task.context = []
-                    # Add content to context
                     task.context.append(content)
         
-        # Planning Mode: Create plan and todo list before execution
-        if self.planning:
-            self._run_with_planning()
+        # ─────────────────────────────────────────────────────────────
+        # Verbose Mode: Show Rich panels for multi-agent workflow
+        # ─────────────────────────────────────────────────────────────
+        if show_verbose and is_tty:
+            from rich.panel import Panel
+            from rich.text import Text
+            console = Console()
+            import time as time_module
+            
+            # Show workflow overview panel
+            agent_names = " → ".join([a.name for a in self.agents])
+            workflow_info = f"[bold {PRAISON_COLORS['metrics']}]Process:[/] {self.process}\n"
+            workflow_info += f"[bold {PRAISON_COLORS['metrics']}]Agents:[/] {agent_names}"
+            
+            console.print(Panel(
+                workflow_info,
+                title="[bold]Multi-Agent Workflow[/]",
+                border_style=PRAISON_COLORS["agent"],
+                padding=(1, 2)
+            ))
+            console.print()
+            
+            # Execute tasks with verbose output
+            total_agents = len(self.agents)
+            workflow_start_time = time_module.time()
+            
+            for idx, (task_id, task) in enumerate(self.tasks.items(), 1):
+                agent = task.agent
+                agent_name = agent.name if agent else "Unknown"
+                agent_model = getattr(agent, 'llm', 'gpt-4o-mini') if agent else "unknown"
+                
+                # Show agent task panel with model info
+                task_desc = task.description[:100] + "..." if len(task.description) > 100 else task.description
+                panel_content = f"[bold {PRAISON_COLORS['task_text']}]📋 Task:[/] {task_desc}\n"
+                panel_content += f"[dim]🤖 Model: {agent_model}[/dim]"
+                console.print(Panel.fit(
+                    panel_content,
+                    title=f"[bold]Agent [{idx}/{total_agents}]: {agent_name}[/]",
+                    border_style=PRAISON_COLORS["task"]
+                ))
+                
+                # Execute with timing and status
+                start_time = time_module.time()
+                
+                # Show working spinner
+                with console.status(
+                    f"[bold yellow]Working...[/]  {agent_name} generating response...",
+                    spinner="dots",
+                    spinner_style="yellow"
+                ):
+                    # Run the task
+                    if self.planning:
+                        self._run_with_planning()
+                        break  # Planning mode handles all tasks
+                    else:
+                        self.run_task(task_id)
+                
+                elapsed = time_module.time() - start_time
+                
+                # Show response panel - FULL response, no truncation
+                result = self.get_task_result(task_id)
+                if result:
+                    response_text = str(result.raw)
+                    # No truncation - show full response in verbose mode
+                    console.print(Panel(
+                        Text(response_text),
+                        title=f"[bold]Agent [{idx}/{total_agents}] Complete ({elapsed:.1f}s)[/]",
+                        border_style=PRAISON_COLORS["response"],
+                        padding=(1, 2)
+                    ))
+                console.print()
+            
+            # Workflow summary panel
+            total_elapsed = time_module.time() - workflow_start_time
+            console.print(Panel.fit(
+                f"[bold green]Total Time:[/] {total_elapsed:.1f}s\n"
+                f"[bold green]Agents Run:[/] {total_agents}/{total_agents}",
+                title="[bold]✅ Workflow Complete[/]",
+                border_style="green"
+            ))
+            console.print()
         else:
-            # Run tasks as before
-            self.run_all_tasks()
+            # Silent mode: Run tasks without display
+            if self.planning:
+                self._run_with_planning()
+            else:
+                self.run_all_tasks()
         
         # Auto-display token metrics if any agent has metrics=True
         metrics_enabled = any(getattr(agent, 'metrics', False) for agent in self.agents)
@@ -1224,10 +1337,8 @@ Context:
             try:
                 self.display_token_usage()
             except (ImportError, AttributeError) as e:
-                # Token tracking not available or not properly configured
                 logging.debug(f"Could not auto-display token usage: {e}")
             except Exception as e:
-                # Log unexpected errors for debugging
                 logging.debug(f"Unexpected error in token metrics display: {e}")
         
         # Get results
@@ -1238,7 +1349,6 @@ Context:
         
         # By default, return only the final agent's response
         if not return_dict:
-            # Get the last task (assuming sequential processing)
             task_ids = list(self.tasks.keys())
             if task_ids:
                 last_task_id = task_ids[-1]
@@ -1246,12 +1356,21 @@ Context:
                 if last_result:
                     return last_result.raw
                     
-        # Return full results dict if return_dict is True or if no final result was found
         return results
 
     def run(self, content=None, return_dict=False, **kwargs):
-        """Alias for start() method to provide consistent API with Agent class"""
-        return self.start(content=content, return_dict=return_dict, **kwargs)
+        """Run agents silently (production use).
+        
+        Unlike .start() which shows verbose output, .run() executes silently
+        for programmatic/production use.
+        
+        Args:
+            content: Optional content to add to all tasks' context
+            return_dict: If True, returns the full results dictionary
+            **kwargs: Additional arguments
+        """
+        # Always run silently - no verbose output
+        return self.start(content=content, return_dict=return_dict, output="silent", **kwargs)
 
     def set_state(self, key: str, value: Any) -> None:
         """Set a state value"""

src/praisonai-agents/praisonaiagents/agents/autoagents.py

@@ -103,7 +103,7 @@ class AutoAgents(Agents):
         self.self_reflect = self_reflect
         self.max_reflect = max_reflect
         self.min_reflect = min_reflect
-        self.llm = llm or os.getenv('OPENAI_MODEL_NAME', 'gpt-5-nano')
+        self.llm = llm or os.getenv('OPENAI_MODEL_NAME', 'gpt-4o-mini')
         self.function_calling_llm = function_calling_llm
         self.code_execution_mode = code_execution_mode
         self.embedder_config = embedder_config

src/praisonai-agents/praisonaiagents/config/feature_configs.py

@@ -44,12 +44,79 @@ class MemoryBackend(str, Enum):
     MONGODB = "mongodb"
 
 
+class LearnScope(str, Enum):
+    """Scope for learning data visibility.
+    
+    PRIVATE: Learning data is private to this user/agent (default, safest)
+    SHARED: Learning data is shared with all agents
+    """
+    PRIVATE = "private"   # Private to this user/agent
+    SHARED = "shared"     # Shared with all agents
+
+
+@dataclass
+class LearnConfig:
+    """
+    Configuration for continuous learning within memory system.
+    
+    Learning captures patterns, preferences, and insights from agent interactions
+    to improve future responses. All learning data is stored within the memory system.
+    
+    Usage:
+        # Simple enable
+        Agent(memory=MemoryConfig(learn=True))
+        
+        # With specific capabilities
+        Agent(memory=MemoryConfig(
+            learn=LearnConfig(
+                persona=True,      # User preferences
+                insights=True,     # Observations
+                patterns=True,     # Reusable knowledge
+            )
+        ))
+    """
+    # Learning capabilities
+    persona: bool = True       # User preferences and profile
+    insights: bool = True      # Observations and learnings
+    thread: bool = True        # Session/conversation context
+    patterns: bool = False     # Reusable knowledge patterns
+    decisions: bool = False    # Decision logging
+    feedback: bool = False     # Outcome signals
+    improvements: bool = False # Self-improvement proposals
+    
+    # Scope configuration
+    scope: Union[str, LearnScope] = LearnScope.PRIVATE
+    
+    # Storage configuration
+    store_path: Optional[str] = None  # Custom storage path
+    
+    # Maintenance settings
+    auto_consolidate: bool = True     # Auto-consolidate learnings
+    retention_days: Optional[int] = None  # Days to retain (None = forever)
+    
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert to dictionary."""
+        return {
+            "persona": self.persona,
+            "insights": self.insights,
+            "thread": self.thread,
+            "patterns": self.patterns,
+            "decisions": self.decisions,
+            "feedback": self.feedback,
+            "improvements": self.improvements,
+            "scope": self.scope.value if isinstance(self.scope, LearnScope) else self.scope,
+            "store_path": self.store_path,
+            "auto_consolidate": self.auto_consolidate,
+            "retention_days": self.retention_days,
+        }
+
+
 @dataclass
 class MemoryConfig:
     """
     Configuration for agent memory and session management.
     
-    Consolidates: memory, auto_memory, claude_memory, user_id, session_id, db
+    Consolidates: memory, auto_memory, claude_memory, user_id, session_id, db, learn
     
     Usage:
         # Simple enable (uses FileMemory)
@@ -58,12 +125,22 @@ class MemoryConfig:
         # With backend
         Agent(memory=MemoryConfig(backend="redis"))
         
-        # Full config
+        # Full config with learning
         Agent(memory=MemoryConfig(
             backend="sqlite",
             user_id="user123",
             session_id="session456",
             auto_memory=True,
+            learn=True,  # Enable continuous learning
+        ))
+        
+        # With detailed learn config
+        Agent(memory=MemoryConfig(
+            learn=LearnConfig(
+                persona=True,
+                insights=True,
+                patterns=True,
+            )
         ))
     """
     # Backend selection
@@ -85,8 +162,17 @@ class MemoryConfig:
     # Memory configuration dict (for provider-specific settings)
     config: Optional[Dict[str, Any]] = None
     
+    # Continuous learning configuration
+    learn: Optional[Union[bool, LearnConfig]] = None
+    
     def to_dict(self) -> Dict[str, Any]:
         """Convert to dictionary."""
+        learn_dict = None
+        if self.learn is True:
+            learn_dict = LearnConfig().to_dict()
+        elif isinstance(self.learn, LearnConfig):
+            learn_dict = self.learn.to_dict()
+        
         return {
             "backend": self.backend.value if isinstance(self.backend, MemoryBackend) else self.backend,
             "user_id": self.user_id,
@@ -94,6 +180,7 @@ class MemoryConfig:
             "auto_memory": self.auto_memory,
             "claude_memory": self.claude_memory,
             "config": self.config,
+            "learn": learn_dict,
         }
 
 
@@ -391,10 +478,21 @@ class OutputConfig:
     
     Consolidates: verbose, markdown, stream, metrics, reasoning_steps, output_style
     
+    DEFAULT: output="silent" (zero overhead, fastest performance)
+    
     Usage:
-        # Simple preset
+        # Default is silent mode (no output overhead, programmatic use)
+        Agent(instructions="...")  # Uses output="silent"
+        
+        # Actions mode (tool calls + final output trace)
+        Agent(output="actions")
+        
+        # Verbose mode with Rich panels
         Agent(output="verbose")
         
+        # JSON mode for piping
+        Agent(output="json")
+        
         # With config
         Agent(output=OutputConfig(
             verbose=True,
@@ -404,11 +502,11 @@ class OutputConfig:
             reasoning_steps=True,
         ))
     """
-    # Verbosity
-    verbose: bool = True
+    # Verbosity - False by default for silent mode (fastest)
+    verbose: bool = False
     
-    # Formatting
-    markdown: bool = True
+    # Formatting - False by default for silent mode
+    markdown: bool = False
     
     # Streaming
     stream: bool = False
@@ -422,6 +520,14 @@ class OutputConfig:
     # Output style (custom styling)
     style: Optional[Any] = None
     
+    # Actions trace mode - shows tool calls, agent lifecycle, final output
+    # When True, registers callbacks and outputs to stderr
+    # DEFAULT: False for zero overhead (silent mode)
+    actions_trace: bool = False
+    
+    # JSON output mode - emit JSONL events for piping
+    json_output: bool = False
+    
     def to_dict(self) -> Dict[str, Any]:
         """Convert to dictionary."""
         return {
@@ -430,6 +536,8 @@ class OutputConfig:
             "stream": self.stream,
             "metrics": self.metrics,
             "reasoning_steps": self.reasoning_steps,
+            "actions_trace": self.actions_trace,
+            "json_output": self.json_output,
         }
 
 
@@ -857,9 +965,259 @@ SkillsParam = Union[List[str], SkillsConfig]
 AutonomyParam = Union[bool, Dict[str, Any], AutonomyConfig]
 
 
+# =============================================================================
+# PRECEDENCE LADDER RESOLVERS
+# =============================================================================
+# Precedence: Instance > Config > Array > Dict > String > Bool > Default
+#
+# These resolvers normalize various input types into the canonical Config form.
+# This allows users to pass simplified inputs while maintaining type safety.
+
+
+def resolve_memory(value: MemoryParam) -> Optional[MemoryConfig]:
+    """
+    Resolve memory= parameter following precedence ladder.
+    
+    Precedence: Instance > Config > Dict > String > Bool > Default
+    
+    Args:
+        value: Memory parameter in any supported form
+        
+    Returns:
+        MemoryConfig if enabled, None if disabled
+        
+    Examples:
+        >>> resolve_memory(None)  # Default: disabled
+        None
+        >>> resolve_memory(False)  # Explicit disable
+        None
+        >>> resolve_memory(True)  # Enable with defaults
+        MemoryConfig(backend='file')
+        >>> resolve_memory("redis")  # String shorthand
+        MemoryConfig(backend='redis')
+        >>> resolve_memory({"backend": "sqlite", "user_id": "alice"})  # Dict
+        MemoryConfig(backend='sqlite', user_id='alice')
+        >>> resolve_memory(MemoryConfig(backend="postgres"))  # Config passthrough
+        MemoryConfig(backend='postgres')
+    """
+    # Default: disabled
+    if value is None:
+        return None
+    
+    # Bool: False = disabled, True = defaults
+    if value is False:
+        return None
+    if value is True:
+        return MemoryConfig()
+    
+    # String: backend shorthand
+    if isinstance(value, str):
+        try:
+            backend = MemoryBackend(value.lower())
+        except ValueError:
+            backend = value  # Allow custom backend strings
+        return MemoryConfig(backend=backend)
+    
+    # Dict: expand to config
+    if isinstance(value, dict):
+        # Handle backend enum conversion
+        backend = value.get("backend", MemoryBackend.FILE)
+        if isinstance(backend, str):
+            try:
+                backend = MemoryBackend(backend.lower())
+            except ValueError:
+                pass  # Keep as string for custom backends
+        return MemoryConfig(
+            backend=backend,
+            user_id=value.get("user_id"),
+            session_id=value.get("session_id"),
+            auto_memory=value.get("auto_memory", False),
+            claude_memory=value.get("claude_memory", False),
+            db=value.get("db"),
+            config=value.get("config"),
+        )
+    
+    # Config: passthrough
+    if isinstance(value, MemoryConfig):
+        return value
+    
+    # Instance (MemoryManager): return as-is (caller handles)
+    # This is the highest precedence - user provided a pre-configured instance
+    return value
+
+
+def resolve_knowledge(value: KnowledgeParam) -> Optional[KnowledgeConfig]:
+    """
+    Resolve knowledge= parameter following precedence ladder.
+    
+    Precedence: Instance > Config > Array > Dict > String > Bool > Default
+    
+    Args:
+        value: Knowledge parameter in any supported form
+        
+    Returns:
+        KnowledgeConfig if enabled, None if disabled
+    """
+    # Default: disabled
+    if value is None:
+        return None
+    
+    # Bool: False = disabled, True = defaults
+    if value is False:
+        return None
+    if value is True:
+        return KnowledgeConfig()
+    
+    # String: single source
+    if isinstance(value, str):
+        return KnowledgeConfig(sources=[value])
+    
+    # Array: list of sources
+    if isinstance(value, list):
+        return KnowledgeConfig(sources=value)
+    
+    # Dict: expand to config
+    if isinstance(value, dict):
+        return KnowledgeConfig(
+            sources=value.get("sources", []),
+            embedder=value.get("embedder", "openai"),
+            embedder_config=value.get("embedder_config"),
+            chunking_strategy=value.get("chunking_strategy", ChunkingStrategy.SEMANTIC),
+            chunk_size=value.get("chunk_size", 1000),
+            chunk_overlap=value.get("chunk_overlap", 200),
+            retrieval_k=value.get("retrieval_k", 5),
+            retrieval_threshold=value.get("retrieval_threshold", 0.0),
+            rerank=value.get("rerank", False),
+            rerank_model=value.get("rerank_model"),
+            auto_retrieve=value.get("auto_retrieve", True),
+        )
+    
+    # Config: passthrough
+    if isinstance(value, KnowledgeConfig):
+        return value
+    
+    # Instance: return as-is
+    return value
+
+
+def resolve_planning(value: PlanningParam) -> Optional[PlanningConfig]:
+    """Resolve planning= parameter following precedence ladder."""
+    if value is None or value is False:
+        return None
+    if value is True:
+        return PlanningConfig()
+    if isinstance(value, dict):
+        return PlanningConfig(**value)
+    if isinstance(value, PlanningConfig):
+        return value
+    return value
+
+
+def resolve_reflection(value: ReflectionParam) -> Optional[ReflectionConfig]:
+    """Resolve reflection= parameter following precedence ladder."""
+    if value is None or value is False:
+        return None
+    if value is True:
+        return ReflectionConfig()
+    if isinstance(value, dict):
+        return ReflectionConfig(**value)
+    if isinstance(value, ReflectionConfig):
+        return value
+    return value
+
+
+def resolve_guardrails(value: GuardrailParam) -> Optional[GuardrailConfig]:
+    """Resolve guardrails= parameter following precedence ladder."""
+    if value is None or value is False:
+        return None
+    if value is True:
+        return GuardrailConfig()
+    if callable(value):
+        return GuardrailConfig(validator=value)
+    if isinstance(value, dict):
+        return GuardrailConfig(**value)
+    if isinstance(value, GuardrailConfig):
+        return value
+    return value
+
+
+def resolve_web(value: WebParam) -> Optional[WebConfig]:
+    """Resolve web= parameter following precedence ladder."""
+    if value is None or value is False:
+        return None
+    if value is True:
+        return WebConfig()
+    if isinstance(value, dict):
+        return WebConfig(**value)
+    if isinstance(value, WebConfig):
+        return value
+    return value
+
+
+def resolve_output(value: OutputParam) -> Optional[OutputConfig]:
+    """Resolve output= parameter following precedence ladder."""
+    if value is None:
+        return None
+    if isinstance(value, str):
+        try:
+            preset = OutputPreset(value.lower())
+            return OutputConfig.from_preset(preset)
+        except ValueError:
+            return None
+    if isinstance(value, dict):
+        return OutputConfig(**value)
+    if isinstance(value, OutputConfig):
+        return value
+    return value
+
+
+def resolve_execution(value: ExecutionParam) -> Optional[ExecutionConfig]:
+    """Resolve execution= parameter following precedence ladder."""
+    if value is None:
+        return None
+    if isinstance(value, str):
+        try:
+            preset = ExecutionPreset(value.lower())
+            return ExecutionConfig.from_preset(preset)
+        except ValueError:
+            return None
+    if isinstance(value, dict):
+        return ExecutionConfig(**value)
+    if isinstance(value, ExecutionConfig):
+        return value
+    return value
+
+
+def resolve_caching(value: CachingParam) -> Optional[CachingConfig]:
+    """Resolve caching= parameter following precedence ladder."""
+    if value is None or value is False:
+        return None
+    if value is True:
+        return CachingConfig()
+    if isinstance(value, dict):
+        return CachingConfig(**value)
+    if isinstance(value, CachingConfig):
+        return value
+    return value
+
+
+def resolve_autonomy(value: AutonomyParam) -> Optional[AutonomyConfig]:
+    """Resolve autonomy= parameter following precedence ladder."""
+    if value is None or value is False:
+        return None
+    if value is True:
+        return AutonomyConfig()
+    if isinstance(value, dict):
+        return AutonomyConfig(**value)
+    if isinstance(value, AutonomyConfig):
+        return value
+    return value
+
+
 __all__ = [
     # Enums
     "MemoryBackend",
+    "LearnScope",
     "ChunkingStrategy",
     "GuardrailAction",
     "WebSearchProvider",
@@ -867,6 +1225,7 @@ __all__ = [
     "ExecutionPreset",
     "AutonomyLevel",
     # Config classes (Agent)
+    "LearnConfig",
     "MemoryConfig",
     "KnowledgeConfig",
     "PlanningConfig",
@@ -900,4 +1259,15 @@ __all__ = [
     "HooksParam",
     "SkillsParam",
     "AutonomyParam",
+    # Precedence ladder resolvers
+    "resolve_memory",
+    "resolve_knowledge",
+    "resolve_planning",
+    "resolve_reflection",
+    "resolve_guardrails",
+    "resolve_web",
+    "resolve_output",
+    "resolve_execution",
+    "resolve_caching",
+    "resolve_autonomy",
 ]

src/praisonai-agents/praisonaiagents/config/presets.py

@@ -41,6 +41,37 @@ MEMORY_URL_SCHEMES: Dict[str, str] = {
 # =============================================================================
 
 OUTPUT_PRESETS: Dict[str, Dict[str, Any]] = {
+    # Silent preset - DEFAULT for SDK
+    # Zero overhead, no callbacks, no I/O on hot path
+    # Fastest performance for programmatic use
+    "silent": {
+        "verbose": False,
+        "markdown": False,
+        "stream": False,
+        "metrics": False,
+        "reasoning_steps": False,
+        "actions_trace": False,
+    },
+    # Actions preset - opt-in observability
+    # Shows action trace (tool calls, agent lifecycle) + final output
+    # Registers callbacks, outputs to stderr
+    "actions": {
+        "verbose": False,
+        "markdown": False,
+        "stream": False,
+        "metrics": False,
+        "reasoning_steps": False,
+        "actions_trace": True,  # Special flag for action trace mode
+    },
+    # Plain preset - final output only, no action trace
+    "plain": {
+        "verbose": False,
+        "markdown": False,
+        "stream": False,
+        "metrics": False,
+        "reasoning_steps": False,
+        "actions_trace": False,
+    },
     "minimal": {
         "verbose": False,
         "markdown": False,
@@ -69,13 +100,6 @@ OUTPUT_PRESETS: Dict[str, Dict[str, Any]] = {
         "metrics": True,
         "reasoning_steps": True,
     },
-    "silent": {
-        "verbose": False,
-        "markdown": False,
-        "stream": False,
-        "metrics": False,
-        "reasoning_steps": False,
-    },
     # Streaming preset - enables streaming by default
     "stream": {
         "verbose": True,
@@ -84,8 +108,23 @@ OUTPUT_PRESETS: Dict[str, Dict[str, Any]] = {
         "metrics": False,
         "reasoning_steps": False,
     },
+    # JSON preset - JSONL output for piping
+    "json": {
+        "verbose": False,
+        "markdown": False,
+        "stream": False,
+        "metrics": False,
+        "reasoning_steps": False,
+        "actions_trace": True,
+        "json_output": True,  # Output as JSONL
+    },
 }
 
+# Default output mode - can be overridden by PRAISONAI_OUTPUT env var
+# "silent" = zero overhead, fastest performance (DEFAULT)
+# "actions" = tool call trace + final output (opt-in observability)
+DEFAULT_OUTPUT_MODE = "silent"
+
 
 # =============================================================================
 # Execution Presets

src/praisonai-agents/praisonaiagents/context/__init__.py

@@ -157,6 +157,16 @@ __all__ = [
     "FastContextResult",
     "FileMatch",
     "LineRange",
+    # Artifacts (Dynamic Context Discovery)
+    "ArtifactRef",
+    "ArtifactMetadata",
+    "GrepMatch",
+    "QueueConfig",
+    "ArtifactStoreProtocol",
+    "HistoryStoreProtocol",
+    "TerminalLoggerProtocol",
+    "compute_checksum",
+    "generate_summary",
 ]
 
 
@@ -252,4 +262,11 @@ def __getattr__(name: str):
         from praisonaiagents.context.fast.result import LineRange
         return LineRange
     
+    # Artifacts (Dynamic Context Discovery)
+    if name in ("ArtifactRef", "ArtifactMetadata", "GrepMatch", "QueueConfig",
+                "ArtifactStoreProtocol", "HistoryStoreProtocol", "TerminalLoggerProtocol",
+                "compute_checksum", "generate_summary"):
+        from . import artifacts
+        return getattr(artifacts, name)
+    
     raise AttributeError(f"module {__name__!r} has no attribute {name!r}")

src/praisonai-agents/praisonaiagents/context/artifacts.py

@@ -0,0 +1,580 @@
+"""
+Artifact Types and Protocols for Dynamic Context Discovery.
+
+This module provides lightweight types and protocols for artifact-based
+context management. Implementations live in the wrapper (praisonai).
+
+Zero Performance Impact:
+- Pure dataclasses and protocols
+- No heavy imports
+- No I/O at import time
+
+Usage:
+    from praisonaiagents.context.artifacts import ArtifactRef, ArtifactStoreProtocol
+    
+    # Create a reference to a stored artifact
+    ref = ArtifactRef(
+        path="/path/to/artifact.json",
+        summary="API response with 50,000 records",
+        size_bytes=1_200_000,
+        mime_type="application/json",
+    )
+    
+    # Implement the protocol in wrapper
+    class FileSystemArtifactStore:
+        def store(self, content, metadata): ...
+        def load(self, ref): ...
+"""
+
+import time
+import hashlib
+from dataclasses import dataclass, field
+from typing import Any, Dict, List, Optional, Protocol, Union, runtime_checkable
+
+
+@dataclass
+class ArtifactRef:
+    """
+    Reference to a stored artifact.
+    
+    Used to replace large tool outputs in context with a lightweight pointer.
+    The agent can then use tail/grep/load operations to access content on demand.
+    
+    Attributes:
+        path: Absolute path to the artifact file
+        summary: Brief description or first ~200 chars of content
+        size_bytes: Original content size in bytes
+        mime_type: Content type (e.g., "application/json", "text/plain")
+        checksum: SHA256 hash for integrity verification
+        created_at: Unix timestamp when artifact was created
+        artifact_id: Unique identifier for the artifact
+        agent_id: ID of the agent that created this artifact
+        run_id: ID of the run/session this artifact belongs to
+        tool_name: Name of the tool that produced this output (if applicable)
+        turn_id: Conversation turn number when artifact was created
+    """
+    path: str
+    summary: str
+    size_bytes: int
+    mime_type: str = "application/octet-stream"
+    checksum: str = ""
+    created_at: float = field(default_factory=time.time)
+    artifact_id: str = ""
+    agent_id: str = ""
+    run_id: str = ""
+    tool_name: Optional[str] = None
+    turn_id: int = 0
+    
+    def to_inline(self) -> str:
+        """
+        Generate inline representation for context.
+        
+        Returns a compact string that can be included in the conversation
+        context, allowing the agent to reference the artifact.
+        """
+        size_str = self._format_size(self.size_bytes)
+        tool_info = f" from {self.tool_name}" if self.tool_name else ""
+        return f"[Artifact{tool_info}: {self.path} ({size_str}) - {self.summary}]"
+    
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert to dictionary for serialization."""
+        return {
+            "path": self.path,
+            "summary": self.summary,
+            "size_bytes": self.size_bytes,
+            "mime_type": self.mime_type,
+            "checksum": self.checksum,
+            "created_at": self.created_at,
+            "artifact_id": self.artifact_id,
+            "agent_id": self.agent_id,
+            "run_id": self.run_id,
+            "tool_name": self.tool_name,
+            "turn_id": self.turn_id,
+        }
+    
+    @classmethod
+    def from_dict(cls, data: Dict[str, Any]) -> "ArtifactRef":
+        """Create from dictionary."""
+        return cls(
+            path=data.get("path", ""),
+            summary=data.get("summary", ""),
+            size_bytes=data.get("size_bytes", 0),
+            mime_type=data.get("mime_type", "application/octet-stream"),
+            checksum=data.get("checksum", ""),
+            created_at=data.get("created_at", time.time()),
+            artifact_id=data.get("artifact_id", ""),
+            agent_id=data.get("agent_id", ""),
+            run_id=data.get("run_id", ""),
+            tool_name=data.get("tool_name"),
+            turn_id=data.get("turn_id", 0),
+        )
+    
+    @staticmethod
+    def _format_size(size_bytes: int) -> str:
+        """Format size in human-readable form."""
+        if size_bytes < 1024:
+            return f"{size_bytes} B"
+        elif size_bytes < 1024 * 1024:
+            return f"{size_bytes / 1024:.1f} KB"
+        elif size_bytes < 1024 * 1024 * 1024:
+            return f"{size_bytes / (1024 * 1024):.1f} MB"
+        else:
+            return f"{size_bytes / (1024 * 1024 * 1024):.1f} GB"
+
+
+@dataclass
+class ArtifactMetadata:
+    """
+    Metadata for artifact creation and tracking.
+    
+    Passed to ArtifactStore.store() to provide context about the artifact.
+    """
+    agent_id: str = ""
+    run_id: str = ""
+    tool_name: Optional[str] = None
+    turn_id: int = 0
+    content_type: str = "auto"  # auto, json, text, binary
+    tags: List[str] = field(default_factory=list)
+    extra: Dict[str, Any] = field(default_factory=dict)
+    
+    def to_dict(self) -> Dict[str, Any]:
+        return {
+            "agent_id": self.agent_id,
+            "run_id": self.run_id,
+            "tool_name": self.tool_name,
+            "turn_id": self.turn_id,
+            "content_type": self.content_type,
+            "tags": self.tags,
+            "extra": self.extra,
+        }
+
+
+@dataclass
+class GrepMatch:
+    """A single match from grep operation."""
+    line_number: int
+    line_content: str
+    context_before: List[str] = field(default_factory=list)
+    context_after: List[str] = field(default_factory=list)
+    
+    def to_dict(self) -> Dict[str, Any]:
+        return {
+            "line_number": self.line_number,
+            "line_content": self.line_content,
+            "context_before": self.context_before,
+            "context_after": self.context_after,
+        }
+
+
+@runtime_checkable
+class ArtifactStoreProtocol(Protocol):
+    """
+    Protocol for artifact storage backends.
+    
+    Implementations should handle:
+    - Storing content to persistent storage
+    - Loading content back
+    - Efficient tail/grep operations for large files
+    - Checksum verification
+    - Secret redaction (if enabled)
+    
+    Example implementation in wrapper:
+        class FileSystemArtifactStore:
+            def store(self, content, metadata):
+                # Write to ~/.praison/runs/{run_id}/artifacts/
+                ...
+    """
+    
+    def store(
+        self,
+        content: Any,
+        metadata: ArtifactMetadata,
+    ) -> ArtifactRef:
+        """
+        Store content as an artifact.
+        
+        Args:
+            content: The content to store (will be serialized)
+            metadata: Metadata about the artifact
+            
+        Returns:
+            ArtifactRef pointing to the stored artifact
+        """
+        ...
+    
+    def load(self, ref: ArtifactRef) -> Any:
+        """
+        Load full content from an artifact.
+        
+        Args:
+            ref: Reference to the artifact
+            
+        Returns:
+            The deserialized content
+        """
+        ...
+    
+    def tail(self, ref: ArtifactRef, lines: int = 50) -> str:
+        """
+        Get the last N lines of an artifact.
+        
+        Args:
+            ref: Reference to the artifact
+            lines: Number of lines to return
+            
+        Returns:
+            String containing the last N lines
+        """
+        ...
+    
+    def head(self, ref: ArtifactRef, lines: int = 50) -> str:
+        """
+        Get the first N lines of an artifact.
+        
+        Args:
+            ref: Reference to the artifact
+            lines: Number of lines to return
+            
+        Returns:
+            String containing the first N lines
+        """
+        ...
+    
+    def grep(
+        self,
+        ref: ArtifactRef,
+        pattern: str,
+        context_lines: int = 2,
+        max_matches: int = 50,
+    ) -> List[GrepMatch]:
+        """
+        Search for pattern in artifact content.
+        
+        Args:
+            ref: Reference to the artifact
+            pattern: Regex pattern to search for
+            context_lines: Number of context lines before/after match
+            max_matches: Maximum number of matches to return
+            
+        Returns:
+            List of GrepMatch objects
+        """
+        ...
+    
+    def chunk(
+        self,
+        ref: ArtifactRef,
+        start_line: int = 1,
+        end_line: Optional[int] = None,
+    ) -> str:
+        """
+        Get a chunk of lines from an artifact.
+        
+        Args:
+            ref: Reference to the artifact
+            start_line: Starting line number (1-indexed)
+            end_line: Ending line number (inclusive), None for end of file
+            
+        Returns:
+            String containing the requested lines
+        """
+        ...
+    
+    def delete(self, ref: ArtifactRef) -> bool:
+        """
+        Delete an artifact.
+        
+        Args:
+            ref: Reference to the artifact
+            
+        Returns:
+            True if deleted successfully
+        """
+        ...
+    
+    def list_artifacts(
+        self,
+        run_id: Optional[str] = None,
+        agent_id: Optional[str] = None,
+        tool_name: Optional[str] = None,
+    ) -> List[ArtifactRef]:
+        """
+        List artifacts matching filters.
+        
+        Args:
+            run_id: Filter by run ID
+            agent_id: Filter by agent ID
+            tool_name: Filter by tool name
+            
+        Returns:
+            List of matching ArtifactRef objects
+        """
+        ...
+
+
+@runtime_checkable
+class HistoryStoreProtocol(Protocol):
+    """
+    Protocol for conversation history storage.
+    
+    Enables loss recovery during summarization by persisting
+    full conversation history to artifacts.
+    """
+    
+    def append(
+        self,
+        message: Dict[str, Any],
+        agent_id: str,
+        run_id: str,
+    ) -> None:
+        """
+        Append a message to history.
+        
+        Args:
+            message: The message dict (role, content, etc.)
+            agent_id: ID of the agent
+            run_id: ID of the run/session
+        """
+        ...
+    
+    def get_ref(self, agent_id: str, run_id: str) -> Optional[ArtifactRef]:
+        """
+        Get artifact reference for history file.
+        
+        Args:
+            agent_id: ID of the agent
+            run_id: ID of the run/session
+            
+        Returns:
+            ArtifactRef if history exists, None otherwise
+        """
+        ...
+    
+    def search(
+        self,
+        query: str,
+        agent_id: str,
+        run_id: str,
+        max_results: int = 20,
+    ) -> List[Dict[str, Any]]:
+        """
+        Search history for matching messages.
+        
+        Args:
+            query: Search query (supports regex)
+            agent_id: ID of the agent
+            run_id: ID of the run/session
+            max_results: Maximum results to return
+            
+        Returns:
+            List of matching message dicts
+        """
+        ...
+    
+    def get_messages(
+        self,
+        agent_id: str,
+        run_id: str,
+        start_turn: int = 0,
+        end_turn: Optional[int] = None,
+    ) -> List[Dict[str, Any]]:
+        """
+        Get messages from history.
+        
+        Args:
+            agent_id: ID of the agent
+            run_id: ID of the run/session
+            start_turn: Starting turn number
+            end_turn: Ending turn number (None for all)
+            
+        Returns:
+            List of message dicts
+        """
+        ...
+
+
+@runtime_checkable
+class TerminalLoggerProtocol(Protocol):
+    """
+    Protocol for terminal session logging.
+    
+    Captures shell command outputs to artifacts for later retrieval.
+    """
+    
+    def log_command(
+        self,
+        command: str,
+        output: str,
+        exit_code: int,
+        agent_id: str,
+        run_id: str,
+    ) -> ArtifactRef:
+        """
+        Log a command execution.
+        
+        Args:
+            command: The command that was executed
+            output: Combined stdout/stderr output
+            exit_code: Command exit code
+            agent_id: ID of the agent
+            run_id: ID of the run/session
+            
+        Returns:
+            ArtifactRef to the logged output
+        """
+        ...
+    
+    def get_session_ref(self, agent_id: str, run_id: str) -> Optional[ArtifactRef]:
+        """
+        Get artifact reference for terminal session log.
+        
+        Args:
+            agent_id: ID of the agent
+            run_id: ID of the run/session
+            
+        Returns:
+            ArtifactRef if session exists, None otherwise
+        """
+        ...
+    
+    def tail_session(
+        self,
+        agent_id: str,
+        run_id: str,
+        lines: int = 100,
+    ) -> str:
+        """
+        Get last N lines from terminal session.
+        
+        Args:
+            agent_id: ID of the agent
+            run_id: ID of the run/session
+            lines: Number of lines to return
+            
+        Returns:
+            String containing the last N lines
+        """
+        ...
+    
+    def grep_session(
+        self,
+        pattern: str,
+        agent_id: str,
+        run_id: str,
+        max_matches: int = 50,
+    ) -> List[GrepMatch]:
+        """
+        Search terminal session for pattern.
+        
+        Args:
+            pattern: Regex pattern to search for
+            agent_id: ID of the agent
+            run_id: ID of the run/session
+            max_matches: Maximum matches to return
+            
+        Returns:
+            List of GrepMatch objects
+        """
+        ...
+
+
+@dataclass
+class QueueConfig:
+    """
+    Configuration for tool output queuing.
+    
+    Controls when and how large tool outputs are queued to artifacts
+    instead of being included inline in the context.
+    """
+    enabled: bool = True
+    inline_max_bytes: int = 32 * 1024  # 32KB default
+    redact_secrets: bool = True
+    summary_max_chars: int = 200
+    checksum_algorithm: str = "sha256"
+    
+    # Secret patterns to redact (regex)
+    secret_patterns: List[str] = field(default_factory=lambda: [
+        r'(?i)(api[_-]?key|apikey)["\']?\s*[:=]\s*["\']?[\w\-]+',
+        r'(?i)(secret|password|passwd|pwd)["\']?\s*[:=]\s*["\']?[\w\-]+',
+        r'(?i)(token|bearer)["\']?\s*[:=]\s*["\']?[\w\-]+',
+        r'(?i)(auth|authorization)["\']?\s*[:=]\s*["\']?[\w\-]+',
+        r'sk-[a-zA-Z0-9]{20,}',  # OpenAI keys
+        r'ghp_[a-zA-Z0-9]{36}',  # GitHub tokens
+        r'gho_[a-zA-Z0-9]{36}',  # GitHub OAuth
+        r'xox[baprs]-[a-zA-Z0-9\-]+',  # Slack tokens
+    ])
+    
+    def to_dict(self) -> Dict[str, Any]:
+        return {
+            "enabled": self.enabled,
+            "inline_max_bytes": self.inline_max_bytes,
+            "redact_secrets": self.redact_secrets,
+            "summary_max_chars": self.summary_max_chars,
+            "checksum_algorithm": self.checksum_algorithm,
+            "secret_patterns": self.secret_patterns,
+        }
+    
+    @classmethod
+    def from_dict(cls, data: Dict[str, Any]) -> "QueueConfig":
+        return cls(
+            enabled=data.get("enabled", True),
+            inline_max_bytes=data.get("inline_max_bytes", 32 * 1024),
+            redact_secrets=data.get("redact_secrets", True),
+            summary_max_chars=data.get("summary_max_chars", 200),
+            checksum_algorithm=data.get("checksum_algorithm", "sha256"),
+            secret_patterns=data.get("secret_patterns", cls().secret_patterns),
+        )
+
+
+
+
+
+def compute_checksum(content: Union[str, bytes], algorithm: str = "sha256") -> str:
+    """
+    Compute checksum of content.
+    
+    Args:
+        content: String or bytes to hash
+        algorithm: Hash algorithm (sha256, md5, etc.)
+        
+    Returns:
+        Hex digest of the hash
+    """
+    if isinstance(content, str):
+        content = content.encode("utf-8")
+    
+    hasher = hashlib.new(algorithm)
+    hasher.update(content)
+    return hasher.hexdigest()
+
+
+def generate_summary(content: Any, max_chars: int = 200) -> str:
+    """
+    Generate a brief summary of content.
+    
+    Args:
+        content: Content to summarize
+        max_chars: Maximum characters in summary
+        
+    Returns:
+        Brief summary string
+    """
+    if isinstance(content, (dict, list)):
+        # For structured data, describe the structure
+        if isinstance(content, dict):
+            keys = list(content.keys())[:5]
+            key_str = ", ".join(str(k) for k in keys)
+            if len(content) > 5:
+                key_str += f", ... ({len(content)} total keys)"
+            return f"Dict with keys: {key_str}"[:max_chars]
+        else:
+            if len(content) > 0:
+                first_type = type(content[0]).__name__
+                return f"List of {len(content)} {first_type} items"[:max_chars]
+            return "Empty list"
+    
+    # For strings, take first N chars
+    text = str(content)
+    if len(text) <= max_chars:
+        return text
+    return text[:max_chars - 3] + "..."

src/praisonai-agents/praisonaiagents/llm/llm.py

@@ -68,8 +68,8 @@ class LLM:
     MODEL_WINDOWS = {
         # OpenAI
         "gpt-4": 6144,                    # 8,192 actual
-        "gpt-5-nano": 96000,                  # 128,000 actual
-        "gpt-5-nano": 96000,            # 128,000 actual
+        "gpt-4o-mini": 96000,                  # 128,000 actual
+        "gpt-4o-mini": 96000,            # 128,000 actual
         "gpt-4-turbo": 96000,            # 128,000 actual
         "o1-preview": 96000,             # 128,000 actual
         "o1-mini": 96000,                # 128,000 actual
@@ -1430,6 +1430,8 @@ Now provide your final answer using this result."""
         task_id: Optional[str] = None,
         execute_tool_fn: Optional[Callable] = None,
         stream: bool = True,
+        stream_callback: Optional[Callable] = None,
+        emit_events: bool = False,
         **kwargs
     ) -> str:
         """Enhanced get_response with all OpenAI-like features"""
@@ -1489,6 +1491,19 @@ Now provide your final answer using this result."""
             # Disable litellm debug messages
             litellm.set_verbose = False
             
+            # Setup StreamEvent emission if enabled
+            _emit = emit_events and stream_callback is not None
+            if _emit:
+                from ..streaming.events import StreamEvent, StreamEventType
+                _first_token_emitted = False
+                _last_content_time = None
+                _request_start_perf = time.perf_counter()
+                stream_callback(StreamEvent(
+                    type=StreamEventType.REQUEST_START,
+                    timestamp=_request_start_perf,
+                    metadata={"model": self.model, "provider": "litellm"}
+                ))
+            
             # Format tools if provided
             formatted_tools = self._format_tools_for_litellm(tools)
             
@@ -1525,6 +1540,7 @@ Now provide your final answer using this result."""
             max_iterations = 10  # Prevent infinite loops
             iteration_count = 0
             final_response_text = ""
+            response_text = ""  # Initialize to prevent UnboundLocalError on API errors
             stored_reasoning_content = None  # Store reasoning content from tool execution
             accumulated_tool_results = []  # Store all tool results across iterations
 

src/praisonai-agents/praisonaiagents/llm/model_router.py

@@ -51,7 +51,7 @@ class ModelRouter:
     DEFAULT_MODELS = [
         # Lightweight/cheap models for simple tasks
         ModelProfile(
-            name="gpt-5-nano",
+            name="gpt-4o-mini",
             provider="openai",
             complexity_range=(TaskComplexity.SIMPLE, TaskComplexity.MODERATE),
             cost_per_1k_tokens=0.00075,  # Average of $0.00015 input, $0.0006 output
@@ -80,7 +80,7 @@ class ModelRouter:
         
         # Mid-tier models for moderate complexity
         ModelProfile(
-            name="gpt-5-nano",
+            name="gpt-4o-mini",
             provider="openai",
             complexity_range=(TaskComplexity.MODERATE, TaskComplexity.COMPLEX),
             cost_per_1k_tokens=0.0075,  # Average of $0.0025 input, $0.01 output
@@ -145,7 +145,7 @@ class ModelRouter:
             preferred_providers: List of preferred providers in order
         """
         self.models = models or self.DEFAULT_MODELS
-        self.default_model = default_model or os.getenv('OPENAI_MODEL_NAME', 'gpt-5-nano')
+        self.default_model = default_model or os.getenv('OPENAI_MODEL_NAME', 'gpt-4o-mini')
         self.cost_threshold = cost_threshold
         self.preferred_providers = preferred_providers or []
         

src/praisonai-agents/praisonaiagents/llm/openai_client.py

@@ -582,6 +582,8 @@ class OpenAIClient:
         console: Optional[Console] = None,
         display_fn: Optional[Callable] = None,
         reasoning_steps: bool = False,
+        stream_callback: Optional[Callable] = None,
+        emit_events: bool = False,
         **kwargs
     ) -> Optional[ChatCompletion]:
         """
@@ -596,11 +598,18 @@ class OpenAIClient:
             console: Console for output
             display_fn: Display function for live updates
             reasoning_steps: Whether to show reasoning steps
+            stream_callback: Optional callback for StreamEvent emission
+            emit_events: Whether to emit StreamEvents (requires stream_callback)
             **kwargs: Additional parameters for the API
             
         Returns:
             ChatCompletion object or None if error
         """
+        # Lazy import StreamEvent types only when needed
+        _emit = emit_events and stream_callback is not None
+        if _emit:
+            from ..streaming.events import StreamEvent, StreamEventType
+        
         try:
             # Default start time and console if not provided
             if start_time is None:
@@ -608,6 +617,15 @@ class OpenAIClient:
             if console is None:
                 console = self.console
             
+            # Emit REQUEST_START event
+            request_start_perf = time.perf_counter()
+            if _emit:
+                stream_callback(StreamEvent(
+                    type=StreamEventType.REQUEST_START,
+                    timestamp=request_start_perf,
+                    metadata={"model": model, "message_count": len(messages)}
+                ))
+            
             # Create the response stream
             response_stream = self.sync_client.chat.completions.create(
                 model=model,
@@ -615,29 +633,81 @@ class OpenAIClient:
                 temperature=temperature,
                 tools=tools if tools else None,
                 stream=True,
+                stream_options={"include_usage": True} if _emit else None,
                 **kwargs
             )
             
+            # Emit HEADERS_RECEIVED event (stream object created means headers received)
+            if _emit:
+                stream_callback(StreamEvent(
+                    type=StreamEventType.HEADERS_RECEIVED,
+                    timestamp=time.perf_counter()
+                ))
+            
             full_response_text = ""
             reasoning_content = ""
             chunks = []
+            first_token_emitted = False
+            last_content_time = None
+            ttft_logged = False
             
             # If display function provided, use Live display
             if display_fn:
                 with Live(
                     display_fn("", start_time),
                     console=console,
-                    refresh_per_second=4,
+                    refresh_per_second=10,  # Increased for more responsive streaming
                     transient=True,
                     vertical_overflow="ellipsis",
                     auto_refresh=True
                 ) as live:
                     for chunk in response_stream:
                         chunks.append(chunk)
-                        if chunk.choices[0].delta.content:
-                            full_response_text += chunk.choices[0].delta.content
+                        
+                        # Check for content delta
+                        if chunk.choices and chunk.choices[0].delta.content:
+                            content = chunk.choices[0].delta.content
+                            full_response_text += content
+                            last_content_time = time.perf_counter()
+                            
+                            # Log TTFT on first token
+                            if not ttft_logged:
+                                ttft_ms = (last_content_time - request_start_perf) * 1000
+                                logging.debug(f"TTFT (Time To First Token): {ttft_ms:.0f}ms")
+                                ttft_logged = True
+                            
+                            # Emit FIRST_TOKEN on first content
+                            if _emit and not first_token_emitted:
+                                stream_callback(StreamEvent(
+                                    type=StreamEventType.FIRST_TOKEN,
+                                    timestamp=last_content_time,
+                                    content=content
+                                ))
+                                first_token_emitted = True
+                            elif _emit:
+                                # Emit DELTA_TEXT for subsequent tokens
+                                stream_callback(StreamEvent(
+                                    type=StreamEventType.DELTA_TEXT,
+                                    timestamp=last_content_time,
+                                    content=content
+                                ))
+                            
                             live.update(display_fn(full_response_text, start_time))
                         
+                        # Handle tool calls streaming
+                        if _emit and chunk.choices and hasattr(chunk.choices[0].delta, 'tool_calls') and chunk.choices[0].delta.tool_calls:
+                            for tc in chunk.choices[0].delta.tool_calls:
+                                stream_callback(StreamEvent(
+                                    type=StreamEventType.DELTA_TOOL_CALL,
+                                    timestamp=time.perf_counter(),
+                                    tool_call={
+                                        "index": tc.index,
+                                        "id": getattr(tc, 'id', None),
+                                        "name": getattr(tc.function, 'name', None) if tc.function else None,
+                                        "arguments": getattr(tc.function, 'arguments', None) if tc.function else None
+                                    }
+                                ))
+                        
                         # Update live display with reasoning content if enabled
                         if reasoning_steps and hasattr(chunk.choices[0].delta, "reasoning_content"):
                             rc = chunk.choices[0].delta.reasoning_content
@@ -648,9 +718,61 @@ class OpenAIClient:
                 # Clear the last generating display with a blank line
                 console.print()
             else:
-                # Just collect chunks without display
+                # Just collect chunks without display, but still emit events
                 for chunk in response_stream:
                     chunks.append(chunk)
+                    
+                    if chunk.choices and chunk.choices[0].delta.content:
+                        content = chunk.choices[0].delta.content
+                        last_content_time = time.perf_counter()
+                        
+                        # Log TTFT on first token
+                        if not ttft_logged:
+                            ttft_ms = (last_content_time - request_start_perf) * 1000
+                            logging.debug(f"TTFT (Time To First Token): {ttft_ms:.0f}ms")
+                            ttft_logged = True
+                        
+                        if _emit and not first_token_emitted:
+                            stream_callback(StreamEvent(
+                                type=StreamEventType.FIRST_TOKEN,
+                                timestamp=last_content_time,
+                                content=content
+                            ))
+                            first_token_emitted = True
+                        elif _emit:
+                            # Emit DELTA_TEXT for subsequent tokens
+                            stream_callback(StreamEvent(
+                                type=StreamEventType.DELTA_TEXT,
+                                timestamp=last_content_time,
+                                content=content
+                            ))
+                    
+                    # Handle tool calls streaming
+                    if _emit and chunk.choices and hasattr(chunk.choices[0].delta, 'tool_calls') and chunk.choices[0].delta.tool_calls:
+                        for tc in chunk.choices[0].delta.tool_calls:
+                            stream_callback(StreamEvent(
+                                type=StreamEventType.DELTA_TOOL_CALL,
+                                timestamp=time.perf_counter(),
+                                tool_call={
+                                    "index": tc.index,
+                                    "id": getattr(tc, 'id', None),
+                                    "name": getattr(tc.function, 'name', None) if tc.function else None,
+                                    "arguments": getattr(tc.function, 'arguments', None) if tc.function else None
+                                }
+                            ))
+            
+            # Emit LAST_TOKEN and STREAM_END events
+            if _emit:
+                if last_content_time:
+                    stream_callback(StreamEvent(
+                        type=StreamEventType.LAST_TOKEN,
+                        timestamp=last_content_time
+                    ))
+                stream_callback(StreamEvent(
+                    type=StreamEventType.STREAM_END,
+                    timestamp=time.perf_counter(),
+                    metadata={"chunk_count": len(chunks)}
+                ))
             
             final_response = process_stream_chunks(chunks)
             return final_response
@@ -669,6 +791,8 @@ class OpenAIClient:
         console: Optional[Console] = None,
         display_fn: Optional[Callable] = None,
         reasoning_steps: bool = False,
+        stream_callback: Optional[Callable] = None,
+        emit_events: bool = False,
         **kwargs
     ) -> Optional[ChatCompletion]:
         """
@@ -683,11 +807,18 @@ class OpenAIClient:
             console: Console for output
             display_fn: Display function for live updates
             reasoning_steps: Whether to show reasoning steps
+            stream_callback: Optional callback for StreamEvent emission (can be sync or async)
+            emit_events: Whether to emit StreamEvents (requires stream_callback)
             **kwargs: Additional parameters for the API
             
         Returns:
             ChatCompletion object or None if error
         """
+        # Lazy import StreamEvent types only when needed
+        _emit = emit_events and stream_callback is not None
+        if _emit:
+            from ..streaming.events import StreamEvent, StreamEventType
+        
         try:
             # Default start time and console if not provided
             if start_time is None:
@@ -695,6 +826,19 @@ class OpenAIClient:
             if console is None:
                 console = self.console
             
+            # Emit REQUEST_START event
+            request_start_perf = time.perf_counter()
+            if _emit:
+                event = StreamEvent(
+                    type=StreamEventType.REQUEST_START,
+                    timestamp=request_start_perf,
+                    metadata={"model": model, "message_count": len(messages)}
+                )
+                if asyncio.iscoroutinefunction(stream_callback):
+                    await stream_callback(event)
+                else:
+                    stream_callback(event)
+            
             # Create the response stream
             response_stream = await self.async_client.chat.completions.create(
                 model=model,
@@ -702,29 +846,81 @@ class OpenAIClient:
                 temperature=temperature,
                 tools=tools if tools else None,
                 stream=True,
+                stream_options={"include_usage": True} if _emit else None,
                 **kwargs
             )
             
+            # Emit HEADERS_RECEIVED event
+            if _emit:
+                event = StreamEvent(
+                    type=StreamEventType.HEADERS_RECEIVED,
+                    timestamp=time.perf_counter()
+                )
+                if asyncio.iscoroutinefunction(stream_callback):
+                    await stream_callback(event)
+                else:
+                    stream_callback(event)
+            
             full_response_text = ""
             reasoning_content = ""
             chunks = []
+            first_token_emitted = False
+            last_content_time = None
+            
+            async def emit_event(event: 'StreamEvent'):
+                if asyncio.iscoroutinefunction(stream_callback):
+                    await stream_callback(event)
+                else:
+                    stream_callback(event)
             
             # If display function provided, use Live display
             if display_fn:
                 with Live(
                     display_fn("", start_time),
                     console=console,
-                    refresh_per_second=4,
+                    refresh_per_second=10,
                     transient=True,
                     vertical_overflow="ellipsis",
                     auto_refresh=True
                 ) as live:
                     async for chunk in response_stream:
                         chunks.append(chunk)
-                        if chunk.choices[0].delta.content:
-                            full_response_text += chunk.choices[0].delta.content
+                        
+                        if chunk.choices and chunk.choices[0].delta.content:
+                            content = chunk.choices[0].delta.content
+                            full_response_text += content
+                            last_content_time = time.perf_counter()
+                            
+                            if _emit and not first_token_emitted:
+                                await emit_event(StreamEvent(
+                                    type=StreamEventType.FIRST_TOKEN,
+                                    timestamp=last_content_time,
+                                    content=content
+                                ))
+                                first_token_emitted = True
+                            elif _emit:
+                                await emit_event(StreamEvent(
+                                    type=StreamEventType.DELTA_TEXT,
+                                    timestamp=last_content_time,
+                                    content=content
+                                ))
+                            
                             live.update(display_fn(full_response_text, start_time))
                         
+                        # Handle tool calls streaming
+                        if _emit and chunk.choices and hasattr(chunk.choices[0].delta, 'tool_calls') and chunk.choices[0].delta.tool_calls:
+                            for tc in chunk.choices[0].delta.tool_calls:
+                                await emit_event(StreamEvent(
+                                    type=StreamEventType.DELTA_TOOL_CALL,
+                                    timestamp=time.perf_counter(),
+                                    tool_call={
+                                        "index": tc.index,
+                                        "id": getattr(tc, 'id', None),
+                                        "name": getattr(tc.function, 'name', None) if tc.function else None,
+                                        "arguments": getattr(tc.function, 'arguments', None) if tc.function else None
+                                    }
+                                ))
+                        
                         # Update live display with reasoning content if enabled
                         if reasoning_steps and hasattr(chunk.choices[0].delta, "reasoning_content"):
                             rc = chunk.choices[0].delta.reasoning_content
@@ -738,6 +934,50 @@ class OpenAIClient:
                 # Just collect chunks without display
                 async for chunk in response_stream:
                     chunks.append(chunk)
+                    
+                    if _emit and chunk.choices and chunk.choices[0].delta.content:
+                        content = chunk.choices[0].delta.content
+                        last_content_time = time.perf_counter()
+                        
+                        if not first_token_emitted:
+                            await emit_event(StreamEvent(
+                                type=StreamEventType.FIRST_TOKEN,
+                                timestamp=last_content_time,
+                                content=content
+                            ))
+                            first_token_emitted = True
+                        else:
+                            await emit_event(StreamEvent(
+                                type=StreamEventType.DELTA_TEXT,
+                                timestamp=last_content_time,
+                                content=content
+                            ))
+                    
+                    if _emit and chunk.choices and hasattr(chunk.choices[0].delta, 'tool_calls') and chunk.choices[0].delta.tool_calls:
+                        for tc in chunk.choices[0].delta.tool_calls:
+                            await emit_event(StreamEvent(
+                                type=StreamEventType.DELTA_TOOL_CALL,
+                                timestamp=time.perf_counter(),
+                                tool_call={
+                                    "index": tc.index,
+                                    "id": getattr(tc, 'id', None),
+                                    "name": getattr(tc.function, 'name', None) if tc.function else None,
+                                    "arguments": getattr(tc.function, 'arguments', None) if tc.function else None
+                                }
+                            ))
+            
+            # Emit LAST_TOKEN and STREAM_END events
+            if _emit:
+                if last_content_time:
+                    await emit_event(StreamEvent(
+                        type=StreamEventType.LAST_TOKEN,
+                        timestamp=last_content_time
+                    ))
+                await emit_event(StreamEvent(
+                    type=StreamEventType.STREAM_END,
+                    timestamp=time.perf_counter(),
+                    metadata={"chunk_count": len(chunks)}
+                ))
             
             final_response = process_stream_chunks(chunks)
             return final_response
@@ -749,7 +989,7 @@ class OpenAIClient:
     def create_completion(
         self,
         messages: List[Dict[str, Any]],
-        model: str = "gpt-5-nano",
+        model: str = "gpt-4o-mini",
         temperature: float = 1.0,
         stream: bool = False,
         tools: Optional[List[Dict[str, Any]]] = None,
@@ -794,7 +1034,7 @@ class OpenAIClient:
     async def acreate_completion(
         self,
         messages: List[Dict[str, Any]],
-        model: str = "gpt-5-nano",
+        model: str = "gpt-4o-mini",
         temperature: float = 1.0,
         stream: bool = False,
         tools: Optional[List[Dict[str, Any]]] = None,
@@ -839,7 +1079,7 @@ class OpenAIClient:
     def chat_completion_with_tools(
         self,
         messages: List[Dict[str, Any]],
-        model: str = "gpt-5-nano",
+        model: str = "gpt-4o-mini",
         temperature: float = 1.0,
         tools: Optional[List[Any]] = None,
         execute_tool_fn: Optional[Callable] = None,
@@ -904,7 +1144,9 @@ class OpenAIClient:
                 if display_fn and console:
                     # When verbose (display_fn provided), use streaming for better UX
                     try:
-                        with Live(display_fn("", start_time), console=console, refresh_per_second=4, transient=True) as live:
+                        request_start_perf = time.perf_counter()
+                        ttft_logged = False
+                        with Live(display_fn("", start_time), console=console, refresh_per_second=10, transient=True) as live:
                             # Use streaming when display_fn is provided for progressive display
                             response_stream = self.create_completion(
                                 messages=messages,
@@ -922,6 +1164,11 @@ class OpenAIClient:
                             for chunk in response_stream:
                                 chunks.append(chunk)
                                 if chunk.choices[0].delta.content:
+                                    # Log TTFT on first token
+                                    if not ttft_logged:
+                                        ttft_ms = (time.perf_counter() - request_start_perf) * 1000
+                                        logging.debug(f"TTFT (Time To First Token): {ttft_ms:.0f}ms")
+                                        ttft_logged = True
                                     full_response_text += chunk.choices[0].delta.content
                                     live.update(display_fn(full_response_text, start_time))
                             
@@ -1020,7 +1267,7 @@ class OpenAIClient:
     async def achat_completion_with_tools(
         self,
         messages: List[Dict[str, Any]],
-        model: str = "gpt-5-nano",
+        model: str = "gpt-4o-mini",
         temperature: float = 1.0,
         tools: Optional[List[Any]] = None,
         execute_tool_fn: Optional[Callable] = None,
@@ -1204,7 +1451,7 @@ class OpenAIClient:
     def chat_completion_with_tools_stream(
         self,
         messages: List[Dict[str, Any]],
-        model: str = "gpt-5-nano",
+        model: str = "gpt-4o-mini",
         temperature: float = 1.0,
         tools: Optional[List[Any]] = None,
         execute_tool_fn: Optional[Callable] = None,
@@ -1356,7 +1603,7 @@ class OpenAIClient:
         self,
         messages: List[Dict[str, Any]],
         response_format: BaseModel,
-        model: str = "gpt-5-nano",
+        model: str = "gpt-4o-mini",
         temperature: float = 1.0,
         **kwargs
     ) -> Any:
@@ -1390,7 +1637,7 @@ class OpenAIClient:
         self,
         messages: List[Dict[str, Any]],
         response_format: BaseModel,
-        model: str = "gpt-5-nano",
+        model: str = "gpt-4o-mini",
         temperature: float = 1.0,
         **kwargs
     ) -> Any:

src/praisonai-agents/praisonaiagents/main.py

@@ -30,6 +30,53 @@ async_display_callbacks = {}
 # Global approval callback registry
 approval_callback = None
 
+# ─────────────────────────────────────────────────────────────────────────────
+# PraisonAI Unique Color Palette: "Elegant Intelligence"
+# Creates a visual narrative flow: Agent → Task → Working → Response
+# ─────────────────────────────────────────────────────────────────────────────
+PRAISON_COLORS = {
+    # Agent identity - grounded, stable
+    "agent": "#86A789",       # Soft Sage Green
+    "agent_text": "#D2E3C8",  # Light sage for text
+    
+    # Task/Question - input, attention-grabbing
+    "task": "#FF9B9B",        # Warm Coral
+    "task_text": "#FFE5E5",   # Light coral for text
+    
+    # Working/Processing - action, energy
+    "working": "#FFB347",     # Amber
+    "working_text": "#FFF3E0", # Light amber for text
+    
+    # Response/Output - completion, calm
+    "response": "#4A90D9",    # Ocean Blue
+    "response_text": "#E3F2FD", # Light blue for text
+    
+    # Tool calls - special action
+    "tool": "#9B7EDE",        # Violet Accent
+    "tool_text": "#EDE7F6",   # Light violet for text
+    
+    # Reasoning - thought process
+    "reasoning": "#78909C",   # Blue Gray
+    "reasoning_text": "#ECEFF1", # Light gray for text
+    
+    # Error/Warning - alert
+    "error": "#E57373",       # Alert Red
+    "error_text": "#FFEBEE",  # Light red for text
+    
+    # Metrics - meta information
+    "metrics": "#B4B4B3",     # Cool Gray
+    "metrics_text": "#FAFAFA", # Near white for text
+}
+
+# Status animation frames for "Working" indicator
+WORKING_FRAMES = ["●○○", "○●○", "○○●", "○●○"]
+WORKING_PHASES = [
+    "Analyzing query...",
+    "Processing context...",
+    "Generating response...",
+    "Finalizing output...",
+]
+
 # At the top of the file, add display_callbacks to __all__
 __all__ = [
     'error_logs',
@@ -39,7 +86,19 @@ __all__ = [
     'async_display_callbacks',
     'execute_callback',
     'approval_callback',
-    # ... other exports
+    # Color palette and animation constants
+    'PRAISON_COLORS',
+    'WORKING_FRAMES',
+    'WORKING_PHASES',
+    # Display functions
+    'display_interaction',
+    'display_instruction',
+    'display_tool_call',
+    'display_error',
+    'display_generating',
+    'display_reasoning_steps',
+    'display_working_status',
+    'display_self_reflection',
 ]
 
 def register_display_callback(display_type: str, callback_fn, is_async: bool = False):
@@ -149,10 +208,14 @@ def _clean_display_content(content: str, max_length: int = 20000) -> str:
     
     return content.strip()
 
-def display_interaction(message, response, markdown=True, generation_time=None, console=None, agent_name=None, agent_role=None, agent_tools=None, task_name=None, task_description=None, task_id=None):
+def display_interaction(message, response, markdown=True, generation_time=None, console=None, agent_name=None, agent_role=None, agent_tools=None, task_name=None, task_description=None, task_id=None, metrics=None):
     """Synchronous version of display_interaction.
     
-    Displays the task/message and response in clean panels with response time in title.
+    Displays the task/message and response in clean panels with semantic colors
+    and optional metrics footer. Uses PraisonAI's unique color palette.
+    
+    Args:
+        metrics: Optional dict with token_in, token_out, cost, model for footer display
     """
     if console is None:
         console = Console()
@@ -180,20 +243,38 @@ def display_interaction(message, response, markdown=True, generation_time=None,
         agent_tools=agent_tools,
         task_name=task_name,
         task_description=task_description,
-        task_id=task_id
+        task_id=task_id,
+        metrics=metrics
     )
     
-    # Build response title with time (Agno-style: "Response (1.2s)")
+    # Build response title with time
     response_title = "Response"
     if generation_time:
         response_title = f"Response ({generation_time:.1f}s)"
+    
+    # Build response content with optional metrics footer
+    response_content = response
+    if metrics and isinstance(metrics, dict):
+        # Add dashed separator and compact metrics line
+        tokens_in = metrics.get('tokens_in', 0)
+        tokens_out = metrics.get('tokens_out', 0)
+        cost = metrics.get('cost', 0)
+        model = metrics.get('model', '')
+        
+        metrics_line = f"\n\n─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─\n📊 {tokens_in} tokens in · {tokens_out} out"
+        if cost > 0:
+            metrics_line += f" · ${cost:.4f}"
+        if model:
+            metrics_line += f" · {model}"
+        response_content = response + metrics_line
 
+    # Use semantic colors: Task (coral), Response (ocean blue)
     if markdown:
-        console.print(Panel.fit(Markdown(message), title="Task", border_style="cyan"))
-        console.print(Panel.fit(Markdown(response), title=response_title, border_style="cyan"))
+        console.print(Panel.fit(Markdown(message), title="Task", border_style=PRAISON_COLORS["task"]))
+        console.print(Panel.fit(Markdown(response_content), title=response_title, border_style=PRAISON_COLORS["response"]))
     else:
-        console.print(Panel.fit(Text(message, style="bold green"), title="Task", border_style="cyan"))
-        console.print(Panel.fit(Text(response, style="bold blue"), title=response_title, border_style="cyan"))
+        console.print(Panel.fit(Text(message, style=f"bold {PRAISON_COLORS['task_text']}"), title="Task", border_style=PRAISON_COLORS["task"]))
+        console.print(Panel.fit(Text(response_content, style=f"bold {PRAISON_COLORS['response_text']}"), title=response_title, border_style=PRAISON_COLORS["response"]))
 
 def display_self_reflection(message: str, console=None):
     if not message or not message.strip():
@@ -231,8 +312,11 @@ def display_instruction(message: str, console=None, agent_name: str = None, agen
     if logging.getLogger().getEffectiveLevel() == logging.DEBUG:
         console.print(Panel.fit(Text(message, style="bold blue"), title="Instruction", border_style="cyan"))
 
-def display_tool_call(message: str, console=None, tool_name: str = None, tool_input: dict = None, tool_output: str = None):
-    """Display tool call information in a consolidated panel.
+def display_tool_call(message: str, console=None, tool_name: str = None, tool_input: dict = None, tool_output: str = None, elapsed_time: float = None, success: bool = True):
+    """Display tool call information in PraisonAI's unique timeline format.
+    
+    Uses ▸ prefix, inline timing [X.Xs], and status icons ✓/✗ for a clean,
+    scannable tool activity display.
     
     Args:
         message: The tool call message (legacy format)
@@ -240,6 +324,8 @@ def display_tool_call(message: str, console=None, tool_name: str = None, tool_in
         tool_name: Name of the tool being called
         tool_input: Input arguments to the tool
         tool_output: Output from the tool (if available)
+        elapsed_time: Time taken for tool execution in seconds
+        success: Whether the tool call succeeded
     """
     logging.debug(f"display_tool_call called with message: {repr(message)}")
     if not message or not message.strip():
@@ -253,17 +339,45 @@ def display_tool_call(message: str, console=None, tool_name: str = None, tool_in
     
     # Only print panel if console is provided (verbose mode)
     if console is not None:
-        # Build consolidated tool call display (Agno-style)
-        if tool_name and tool_input is not None:
-            # Format: tool_name(arg1=val1, arg2=val2)
-            args_str = ", ".join(f"{k}={v}" for k, v in tool_input.items()) if tool_input else ""
-            tool_display = f"• {tool_name}({args_str})"
+        # Build unique PraisonAI timeline format
+        if tool_name:
+            # Format: ▸ tool_name(args) [X.Xs] ✓
+            args_str = ""
+            if tool_input:
+                # Truncate long values for display
+                args_parts = []
+                for k, v in tool_input.items():
+                    v_str = str(v)
+                    if len(v_str) > 50:
+                        v_str = v_str[:47] + "..."
+                    args_parts.append(f"{k}={repr(v_str) if isinstance(v, str) else v_str}")
+                args_str = ", ".join(args_parts)
+            
+            # Build the timeline entry
+            status_icon = "✓" if success else "✗"
+            time_str = f"[{elapsed_time:.1f}s]" if elapsed_time else ""
+            
+            tool_display = f"▸ {tool_name}({args_str}) {time_str} {status_icon}"
+            
+            # Add output on next line with tree connector
             if tool_output:
-                tool_display += f"\n  → {tool_output}"
-            console.print(Panel.fit(Text(tool_display, style="bold cyan"), title="Tool Call", border_style="green"))
+                output_str = str(tool_output)
+                if len(output_str) > 100:
+                    output_str = output_str[:97] + "..."
+                tool_display += f"\n  └─ {output_str}"
+            
+            console.print(Panel.fit(
+                Text(tool_display, style=f"bold {PRAISON_COLORS['tool_text']}"), 
+                title="Tool Activity", 
+                border_style=PRAISON_COLORS["tool"]
+            ))
         else:
-            # Legacy format
-            console.print(Panel.fit(Text(message, style="bold cyan"), title="Tool Call", border_style="green"))
+            # Legacy format with new colors
+            console.print(Panel.fit(
+                Text(message, style=f"bold {PRAISON_COLORS['tool_text']}"), 
+                title="Tool Activity", 
+                border_style=PRAISON_COLORS["tool"]
+            ))
 
 def display_error(message: str, console=None):
     if not message or not message.strip():
@@ -275,7 +389,12 @@ def display_error(message: str, console=None):
     # Execute synchronous callbacks
     execute_sync_callback('error', message=message)
     
-    console.print(Panel.fit(Text(message, style="bold red"), title="Error", border_style="red"))
+    # Use semantic error color
+    console.print(Panel.fit(
+        Text(message, style=f"bold {PRAISON_COLORS['error_text']}"), 
+        title="⚠ Error", 
+        border_style=PRAISON_COLORS["error"]
+    ))
     error_logs.append(message)
 
 def display_generating(content: str = "", start_time: Optional[float] = None):
@@ -293,7 +412,73 @@ def display_generating(content: str = "", start_time: Optional[float] = None):
     # Execute synchronous callbacks
     execute_sync_callback('generating', content=content, elapsed_time=elapsed_str.strip() if elapsed_str else None)
     
-    return Panel(Markdown(content), title=f"Generating...{elapsed_str}", border_style="green")
+    # Use semantic response color
+    return Panel(Markdown(content), title=f"Generating...{elapsed_str}", border_style=PRAISON_COLORS["response"])
+
+def display_reasoning_steps(steps: List[str], console=None):
+    """Display reasoning steps with unique numbered circles.
+    
+    Uses ①②③ numbered circles for a distinctive, scannable format
+    that shows the agent's thought process.
+    
+    Args:
+        steps: List of reasoning step descriptions
+        console: Rich console for output
+    """
+    if not steps:
+        return
+    if console is None:
+        console = Console()
+    
+    # Circle number mapping (supports up to 20 steps)
+    circle_numbers = "①②③④⑤⑥⑦⑧⑨⑩⑪⑫⑬⑭⑮⑯⑰⑱⑲⑳"
+    
+    # Build reasoning display with numbered circles
+    reasoning_lines = []
+    for i, step in enumerate(steps):
+        circle = circle_numbers[i] if i < len(circle_numbers) else f"({i+1})"
+        step_text = _clean_display_content(str(step))
+        if len(step_text) > 80:
+            step_text = step_text[:77] + "..."
+        reasoning_lines.append(f"{circle} {step_text}")
+    
+    reasoning_display = "\n".join(reasoning_lines)
+    
+    console.print(Panel.fit(
+        Text(reasoning_display, style=f"italic {PRAISON_COLORS['reasoning_text']}"),
+        title="Reasoning",
+        border_style=PRAISON_COLORS["reasoning"]
+    ))
+
+def display_working_status(phase: int = 0, status_text: str = None, console=None):
+    """Display animated working status with pulsing dots.
+    
+    Shows a unique "Working ●○○" indicator with phase-specific status.
+    This is PraisonAI's distinctive approach to showing processing status.
+    
+    Args:
+        phase: Current animation phase (0-3)
+        status_text: Optional status description
+        console: Rich console for output
+    
+    Returns:
+        Panel object for use with Rich.Live
+    """
+    if console is None:
+        console = Console()
+    
+    # Get current frame and phase text
+    frame = WORKING_FRAMES[phase % len(WORKING_FRAMES)]
+    phase_text = status_text or WORKING_PHASES[phase % len(WORKING_PHASES)]
+    
+    # Build working status display
+    working_display = f"Working {frame}  {phase_text}"
+    
+    return Panel.fit(
+        Text(working_display, style=f"bold {PRAISON_COLORS['working_text']}"),
+        title="Status",
+        border_style=PRAISON_COLORS["working"]
+    )
 
 # Async versions with 'a' prefix
 async def adisplay_interaction(message, response, markdown=True, generation_time=None, console=None, agent_name=None, agent_role=None, agent_tools=None, task_name=None, task_description=None, task_id=None):

src/praisonai-agents/praisonaiagents/mcp/mcp.py

@@ -162,7 +162,7 @@ class MCP:
         # Method 1: Using command and args separately
         agent = Agent(
             instructions="You are a helpful assistant...",
-            llm="gpt-5-nano",
+            llm="gpt-4o-mini",
             tools=MCP(
                 command="/path/to/python",
                 args=["/path/to/app.py"]
@@ -172,14 +172,14 @@ class MCP:
         # Method 2: Using a single command string
         agent = Agent(
             instructions="You are a helpful assistant...",
-            llm="gpt-5-nano",
+            llm="gpt-4o-mini",
             tools=MCP("/path/to/python /path/to/app.py")
         )
         
         # Method 3: Using an SSE endpoint
         agent = Agent(
             instructions="You are a helpful assistant...",
-            llm="gpt-5-nano",
+            llm="gpt-4o-mini",
             tools=MCP("http://localhost:8080/sse")
         )
         
@@ -548,7 +548,7 @@ class MCP:
         """Convert the MCP tool to an OpenAI-compatible tool definition.
         
         This method is specifically invoked by the Agent class when using
-        provider/model format (e.g., "openai/gpt-5-nano").
+        provider/model format (e.g., "openai/gpt-4o-mini").
         
         Returns:
             dict or list: OpenAI-compatible tool definition(s)

src/praisonai-agents/praisonaiagents/memory/__init__.py

@@ -110,6 +110,31 @@ def __getattr__(name):
     if name == "create_hooks_manager":
         from .hooks import create_hooks_manager
         return create_hooks_manager
+    # Learn module
+    if name == "LearnManager":
+        from .learn import LearnManager
+        return LearnManager
+    if name == "PersonaStore":
+        from .learn import PersonaStore
+        return PersonaStore
+    if name == "InsightStore":
+        from .learn import InsightStore
+        return InsightStore
+    if name == "ThreadStore":
+        from .learn import ThreadStore
+        return ThreadStore
+    if name == "PatternStore":
+        from .learn import PatternStore
+        return PatternStore
+    if name == "DecisionStore":
+        from .learn import DecisionStore
+        return DecisionStore
+    if name == "FeedbackStore":
+        from .learn import FeedbackStore
+        return FeedbackStore
+    if name == "ImprovementStore":
+        from .learn import ImprovementStore
+        return ImprovementStore
     # DocsManager and MCPConfigManager are already imported at module level
     raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
 
@@ -153,4 +178,13 @@ __all__ = [
     # MCP config management
     "MCPConfigManager",
     "MCPConfig",
+    # Learn module
+    "LearnManager",
+    "PersonaStore",
+    "InsightStore",
+    "ThreadStore",
+    "PatternStore",
+    "DecisionStore",
+    "FeedbackStore",
+    "ImprovementStore",
 ] 

src/praisonai-agents/praisonaiagents/memory/learn/__init__.py

@@ -0,0 +1,61 @@
+"""
+Learn Module - Continuous Learning within Memory System.
+
+Provides learning capabilities for agents to capture patterns, preferences,
+and insights from interactions to improve future responses.
+
+Usage:
+    from praisonaiagents import Agent, MemoryConfig, LearnConfig
+    
+    # Simple enable
+    agent = Agent(memory=MemoryConfig(learn=True))
+    
+    # With specific capabilities
+    agent = Agent(memory=MemoryConfig(
+        learn=LearnConfig(
+            persona=True,
+            insights=True,
+            patterns=True,
+        )
+    ))
+"""
+
+__all__ = [
+    "LearnManager",
+    "PersonaStore",
+    "InsightStore",
+    "ThreadStore",
+    "PatternStore",
+    "DecisionStore",
+    "FeedbackStore",
+    "ImprovementStore",
+]
+
+
+def __getattr__(name: str):
+    """Lazy loading for learn module components."""
+    if name == "LearnManager":
+        from .manager import LearnManager
+        return LearnManager
+    if name == "PersonaStore":
+        from .stores import PersonaStore
+        return PersonaStore
+    if name == "InsightStore":
+        from .stores import InsightStore
+        return InsightStore
+    if name == "ThreadStore":
+        from .stores import ThreadStore
+        return ThreadStore
+    if name == "PatternStore":
+        from .stores import PatternStore
+        return PatternStore
+    if name == "DecisionStore":
+        from .stores import DecisionStore
+        return DecisionStore
+    if name == "FeedbackStore":
+        from .stores import FeedbackStore
+        return FeedbackStore
+    if name == "ImprovementStore":
+        from .stores import ImprovementStore
+        return ImprovementStore
+    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")

src/praisonai-agents/praisonaiagents/memory/learn/manager.py

@@ -0,0 +1,241 @@
+"""
+LearnManager - Central manager for continuous learning capabilities.
+
+Coordinates all learning stores and provides a unified interface for
+capturing and retrieving learned information.
+"""
+
+from typing import Any, Dict, List, Optional
+
+from ...config.feature_configs import LearnConfig, LearnScope
+from .stores import (
+    PersonaStore,
+    InsightStore,
+    ThreadStore,
+    PatternStore,
+    DecisionStore,
+    FeedbackStore,
+    ImprovementStore,
+    LearnEntry,
+)
+
+
+class LearnManager:
+    """
+    Central manager for continuous learning within the memory system.
+    
+    Coordinates all learning stores based on LearnConfig settings.
+    
+    Usage:
+        config = LearnConfig(persona=True, insights=True)
+        manager = LearnManager(config, user_id="alice")
+        
+        # Capture learning
+        manager.capture_persona("User prefers concise responses")
+        manager.capture_insight("User works in data science")
+        
+        # Retrieve learnings
+        context = manager.get_learning_context()
+    """
+    
+    def __init__(
+        self,
+        config: Optional[LearnConfig] = None,
+        user_id: Optional[str] = None,
+        store_path: Optional[str] = None,
+    ):
+        self.config = config or LearnConfig()
+        self.user_id = user_id or "default"
+        self.store_path = store_path or self.config.store_path
+        
+        scope = self.config.scope
+        if isinstance(scope, LearnScope):
+            scope = scope.value
+        
+        self._stores: Dict[str, Any] = {}
+        
+        if self.config.persona:
+            self._stores["persona"] = PersonaStore(
+                store_path=self._get_store_path("persona"),
+                user_id=self.user_id,
+                scope=scope,
+            )
+        
+        if self.config.insights:
+            self._stores["insights"] = InsightStore(
+                store_path=self._get_store_path("insights"),
+                user_id=self.user_id,
+                scope=scope,
+            )
+        
+        if self.config.thread:
+            self._stores["threads"] = ThreadStore(
+                store_path=self._get_store_path("threads"),
+                user_id=self.user_id,
+                scope=scope,
+            )
+        
+        if self.config.patterns:
+            self._stores["patterns"] = PatternStore(
+                store_path=self._get_store_path("patterns"),
+                user_id=self.user_id,
+                scope=scope,
+            )
+        
+        if self.config.decisions:
+            self._stores["decisions"] = DecisionStore(
+                store_path=self._get_store_path("decisions"),
+                user_id=self.user_id,
+                scope=scope,
+            )
+        
+        if self.config.feedback:
+            self._stores["feedback"] = FeedbackStore(
+                store_path=self._get_store_path("feedback"),
+                user_id=self.user_id,
+                scope=scope,
+            )
+        
+        if self.config.improvements:
+            self._stores["improvements"] = ImprovementStore(
+                store_path=self._get_store_path("improvements"),
+                user_id=self.user_id,
+                scope=scope,
+            )
+    
+    def _get_store_path(self, store_name: str) -> Optional[str]:
+        """Get path for a specific store."""
+        if self.store_path:
+            return f"{self.store_path}/{store_name}.json"
+        return None
+    
+    def capture_persona(self, content: str, category: str = "general") -> Optional[LearnEntry]:
+        """Capture user preference or profile information."""
+        if "persona" in self._stores:
+            return self._stores["persona"].add_preference(content, category)
+        return None
+    
+    def capture_insight(self, content: str, source: str = "interaction") -> Optional[LearnEntry]:
+        """Capture an observation or learning."""
+        if "insights" in self._stores:
+            return self._stores["insights"].add_insight(content, source)
+        return None
+    
+    def capture_thread(self, summary: str, thread_id: str) -> Optional[LearnEntry]:
+        """Capture thread/session context."""
+        if "threads" in self._stores:
+            return self._stores["threads"].add_thread_summary(summary, thread_id)
+        return None
+    
+    def capture_pattern(self, pattern: str, pattern_type: str = "general") -> Optional[LearnEntry]:
+        """Capture a reusable knowledge pattern."""
+        if "patterns" in self._stores:
+            return self._stores["patterns"].add_pattern(pattern, pattern_type)
+        return None
+    
+    def capture_decision(
+        self,
+        decision: str,
+        reasoning: str = "",
+        outcome: str = "",
+    ) -> Optional[LearnEntry]:
+        """Capture a decision record."""
+        if "decisions" in self._stores:
+            return self._stores["decisions"].add_decision(decision, reasoning, outcome)
+        return None
+    
+    def capture_feedback(
+        self,
+        feedback: str,
+        rating: Optional[int] = None,
+        source: str = "user",
+    ) -> Optional[LearnEntry]:
+        """Capture feedback."""
+        if "feedback" in self._stores:
+            return self._stores["feedback"].add_feedback(feedback, rating, source)
+        return None
+    
+    def capture_improvement(
+        self,
+        proposal: str,
+        area: str = "general",
+        priority: str = "medium",
+    ) -> Optional[LearnEntry]:
+        """Capture an improvement proposal."""
+        if "improvements" in self._stores:
+            return self._stores["improvements"].add_improvement(proposal, area, priority)
+        return None
+    
+    def get_learning_context(self, limit_per_store: int = 5) -> Dict[str, List[Dict[str, Any]]]:
+        """
+        Get learning context from all enabled stores.
+        
+        Returns a dictionary with entries from each store, suitable for
+        injection into agent context.
+        """
+        context = {}
+        for store_name, store in self._stores.items():
+            entries = store.list_all(limit=limit_per_store)
+            context[store_name] = [e.to_dict() for e in entries]
+        return context
+    
+    def search(self, query: str, limit: int = 10) -> Dict[str, List[Dict[str, Any]]]:
+        """Search across all enabled stores."""
+        results = {}
+        for store_name, store in self._stores.items():
+            entries = store.search(query, limit=limit)
+            if entries:
+                results[store_name] = [e.to_dict() for e in entries]
+        return results
+    
+    def get_persona_context(self, limit: int = 10) -> List[str]:
+        """Get persona entries as context strings."""
+        if "persona" not in self._stores:
+            return []
+        entries = self._stores["persona"].list_all(limit=limit)
+        return [e.content for e in entries]
+    
+    def get_insights_context(self, limit: int = 10) -> List[str]:
+        """Get insight entries as context strings."""
+        if "insights" not in self._stores:
+            return []
+        entries = self._stores["insights"].list_all(limit=limit)
+        return [e.content for e in entries]
+    
+    def clear_all(self) -> Dict[str, int]:
+        """Clear all stores."""
+        cleared = {}
+        for store_name, store in self._stores.items():
+            cleared[store_name] = store.clear()
+        return cleared
+    
+    def get_stats(self) -> Dict[str, int]:
+        """Get entry counts for all stores."""
+        return {
+            store_name: len(store._entries)
+            for store_name, store in self._stores.items()
+        }
+    
+    def to_system_prompt_context(self) -> str:
+        """
+        Generate context suitable for injection into system prompt.
+        
+        Returns a formatted string with relevant learnings.
+        """
+        parts = []
+        
+        if "persona" in self._stores:
+            personas = self.get_persona_context(limit=5)
+            if personas:
+                parts.append("User Preferences:")
+                for p in personas:
+                    parts.append(f"- {p}")
+        
+        if "insights" in self._stores:
+            insights = self.get_insights_context(limit=5)
+            if insights:
+                parts.append("\nLearned Insights:")
+                for i in insights:
+                    parts.append(f"- {i}")
+        
+        return "\n".join(parts) if parts else ""

src/praisonai-agents/praisonaiagents/memory/learn/stores.py

@@ -0,0 +1,279 @@
+"""
+Learning Stores - Storage backends for different learning capabilities.
+
+Each store handles a specific type of learning data:
+- PersonaStore: User preferences and profile
+- InsightStore: Observations and learnings
+- ThreadStore: Session/conversation context
+- PatternStore: Reusable knowledge patterns
+- DecisionStore: Decision logging
+- FeedbackStore: Outcome signals
+- ImprovementStore: Self-improvement proposals
+"""
+
+import json
+import os
+from abc import ABC, abstractmethod
+from dataclasses import dataclass, field
+from datetime import datetime
+from pathlib import Path
+from typing import Any, Dict, List, Optional
+
+
+@dataclass
+class LearnEntry:
+    """Base entry for all learning stores."""
+    id: str
+    content: str
+    metadata: Dict[str, Any] = field(default_factory=dict)
+    created_at: str = field(default_factory=lambda: datetime.utcnow().isoformat())
+    updated_at: str = field(default_factory=lambda: datetime.utcnow().isoformat())
+    
+    def to_dict(self) -> Dict[str, Any]:
+        return {
+            "id": self.id,
+            "content": self.content,
+            "metadata": self.metadata,
+            "created_at": self.created_at,
+            "updated_at": self.updated_at,
+        }
+    
+    @classmethod
+    def from_dict(cls, data: Dict[str, Any]) -> "LearnEntry":
+        return cls(
+            id=data["id"],
+            content=data["content"],
+            metadata=data.get("metadata", {}),
+            created_at=data.get("created_at", datetime.utcnow().isoformat()),
+            updated_at=data.get("updated_at", datetime.utcnow().isoformat()),
+        )
+
+
+class BaseStore(ABC):
+    """Abstract base class for learning stores."""
+    
+    def __init__(
+        self,
+        store_path: Optional[str] = None,
+        user_id: Optional[str] = None,
+        scope: str = "private",
+    ):
+        self.user_id = user_id or "default"
+        self.scope = scope
+        self.store_path = store_path or self._default_path()
+        self._entries: Dict[str, LearnEntry] = {}
+        self._load()
+    
+    @property
+    @abstractmethod
+    def store_name(self) -> str:
+        """Name of the store (used for file naming)."""
+        pass
+    
+    def _default_path(self) -> str:
+        """Default storage path."""
+        base = Path.home() / ".praison" / "learn" / self.scope / self.user_id
+        base.mkdir(parents=True, exist_ok=True)
+        return str(base / f"{self.store_name}.json")
+    
+    def _load(self) -> None:
+        """Load entries from storage."""
+        if os.path.exists(self.store_path):
+            try:
+                with open(self.store_path, "r") as f:
+                    data = json.load(f)
+                    self._entries = {
+                        k: LearnEntry.from_dict(v) for k, v in data.items()
+                    }
+            except (json.JSONDecodeError, IOError):
+                self._entries = {}
+    
+    def _save(self) -> None:
+        """Save entries to storage."""
+        os.makedirs(os.path.dirname(self.store_path), exist_ok=True)
+        with open(self.store_path, "w") as f:
+            json.dump(
+                {k: v.to_dict() for k, v in self._entries.items()},
+                f,
+                indent=2,
+            )
+    
+    def add(self, content: str, metadata: Optional[Dict[str, Any]] = None) -> LearnEntry:
+        """Add a new entry."""
+        entry_id = f"{self.store_name}_{len(self._entries)}_{datetime.utcnow().timestamp()}"
+        entry = LearnEntry(
+            id=entry_id,
+            content=content,
+            metadata=metadata or {},
+        )
+        self._entries[entry_id] = entry
+        self._save()
+        return entry
+    
+    def get(self, entry_id: str) -> Optional[LearnEntry]:
+        """Get entry by ID."""
+        return self._entries.get(entry_id)
+    
+    def search(self, query: str, limit: int = 10) -> List[LearnEntry]:
+        """Simple text search (can be overridden for semantic search)."""
+        query_lower = query.lower()
+        results = [
+            entry for entry in self._entries.values()
+            if query_lower in entry.content.lower()
+        ]
+        return results[:limit]
+    
+    def list_all(self, limit: int = 100) -> List[LearnEntry]:
+        """List all entries."""
+        entries = list(self._entries.values())
+        entries.sort(key=lambda x: x.updated_at, reverse=True)
+        return entries[:limit]
+    
+    def update(self, entry_id: str, content: str, metadata: Optional[Dict[str, Any]] = None) -> Optional[LearnEntry]:
+        """Update an existing entry."""
+        if entry_id not in self._entries:
+            return None
+        entry = self._entries[entry_id]
+        entry.content = content
+        if metadata:
+            entry.metadata.update(metadata)
+        entry.updated_at = datetime.utcnow().isoformat()
+        self._save()
+        return entry
+    
+    def delete(self, entry_id: str) -> bool:
+        """Delete an entry."""
+        if entry_id in self._entries:
+            del self._entries[entry_id]
+            self._save()
+            return True
+        return False
+    
+    def clear(self) -> int:
+        """Clear all entries."""
+        count = len(self._entries)
+        self._entries = {}
+        self._save()
+        return count
+
+
+class PersonaStore(BaseStore):
+    """Store for user preferences and profile information."""
+    
+    @property
+    def store_name(self) -> str:
+        return "persona"
+    
+    def add_preference(self, preference: str, category: str = "general") -> LearnEntry:
+        """Add a user preference."""
+        return self.add(preference, {"category": category, "type": "preference"})
+    
+    def add_profile(self, profile_data: str, aspect: str = "general") -> LearnEntry:
+        """Add profile information."""
+        return self.add(profile_data, {"aspect": aspect, "type": "profile"})
+
+
+class InsightStore(BaseStore):
+    """Store for observations and learnings."""
+    
+    @property
+    def store_name(self) -> str:
+        return "insights"
+    
+    def add_insight(self, insight: str, source: str = "interaction") -> LearnEntry:
+        """Add an insight."""
+        return self.add(insight, {"source": source, "type": "insight"})
+    
+    def add_observation(self, observation: str, context: str = "") -> LearnEntry:
+        """Add an observation."""
+        return self.add(observation, {"context": context, "type": "observation"})
+
+
+class ThreadStore(BaseStore):
+    """Store for session/conversation context."""
+    
+    @property
+    def store_name(self) -> str:
+        return "threads"
+    
+    def add_thread_summary(self, summary: str, thread_id: str) -> LearnEntry:
+        """Add a thread summary."""
+        return self.add(summary, {"thread_id": thread_id, "type": "summary"})
+    
+    def add_context(self, context: str, thread_id: str) -> LearnEntry:
+        """Add thread context."""
+        return self.add(context, {"thread_id": thread_id, "type": "context"})
+
+
+class PatternStore(BaseStore):
+    """Store for reusable knowledge patterns."""
+    
+    @property
+    def store_name(self) -> str:
+        return "patterns"
+    
+    def add_pattern(self, pattern: str, pattern_type: str = "general") -> LearnEntry:
+        """Add a knowledge pattern."""
+        return self.add(pattern, {"pattern_type": pattern_type, "type": "pattern"})
+
+
+class DecisionStore(BaseStore):
+    """Store for decision logging."""
+    
+    @property
+    def store_name(self) -> str:
+        return "decisions"
+    
+    def add_decision(
+        self,
+        decision: str,
+        reasoning: str = "",
+        outcome: str = "",
+    ) -> LearnEntry:
+        """Add a decision record."""
+        return self.add(decision, {
+            "reasoning": reasoning,
+            "outcome": outcome,
+            "type": "decision",
+        })
+
+
+class FeedbackStore(BaseStore):
+    """Store for outcome signals and feedback."""
+    
+    @property
+    def store_name(self) -> str:
+        return "feedback"
+    
+    def add_feedback(
+        self,
+        feedback: str,
+        rating: Optional[int] = None,
+        source: str = "user",
+    ) -> LearnEntry:
+        """Add feedback."""
+        metadata = {"source": source, "type": "feedback"}
+        if rating is not None:
+            metadata["rating"] = rating
+        return self.add(feedback, metadata)
+
+
+class ImprovementStore(BaseStore):
+    """Store for self-improvement proposals."""
+    
+    @property
+    def store_name(self) -> str:
+        return "improvements"
+    
+    def add_improvement(
+        self,
+        proposal: str,
+        area: str = "general",
+        priority: str = "medium",
+    ) -> LearnEntry:
+        """Add an improvement proposal."""
+        return self.add(proposal, {
+            "area": area,
+            "priority": priority,
+            "type": "improvement",
+        })

src/praisonai-agents/praisonaiagents/memory/memory.py

@@ -109,7 +109,7 @@ class Memory:
         },
         "llm": {
           "provider": "openai",
-          "config": {"model": "gpt-5-nano", "api_key": "..."}
+          "config": {"model": "gpt-4o-mini", "api_key": "..."}
         },
         "embedder": {
           "provider": "openai",
@@ -146,6 +146,8 @@ class Memory:
         self.use_rag = (self.provider.lower() == "rag") and CHROMADB_AVAILABLE and self.cfg.get("use_embedding", False)
         self.use_mongodb = (self.provider.lower() == "mongodb") and PYMONGO_AVAILABLE
         self.graph_enabled = False  # Initialize graph support flag
+        self._learn_manager = None  # Lazy-loaded LearnManager
+        self._learn_config = self.cfg.get("learn", None)  # Learn configuration
         
         # Extract embedding model from config
         self.embedder_config = self.cfg.get("embedder", {})
@@ -1442,7 +1444,7 @@ class Memory:
                 import litellm
                 
                 # Convert model name if it's in litellm format
-                model_name = llm or "gpt-5-nano"
+                model_name = llm or "gpt-4o-mini"
                 
                 response = litellm.completion(
                     model=model_name,
@@ -1459,7 +1461,7 @@ class Memory:
                 client = OpenAI()
                 
                 response = client.chat.completions.create(
-                    model=llm or "gpt-5-nano",
+                    model=llm or "gpt-4o-mini",
                     messages=[{
                         "role": "user", 
                         "content": custom_prompt or default_prompt
@@ -1599,3 +1601,50 @@ class Memory:
         except Exception as e:
             self._log_verbose(f"Error getting all memories: {e}", logging.ERROR)
             return []
+
+    # -------------------------------------------------------------------------
+    #                          Learn Integration
+    # -------------------------------------------------------------------------
+    @property
+    def learn(self):
+        """
+        Get the LearnManager for continuous learning capabilities.
+        
+        Returns None if learn is not enabled in config.
+        
+        Usage:
+            memory = Memory({"learn": True})
+            memory.learn.capture_persona("User prefers concise responses")
+            memory.learn.capture_insight("User works in data science")
+        """
+        if self._learn_manager is not None:
+            return self._learn_manager
+        
+        if self._learn_config is None or self._learn_config is False:
+            return None
+        
+        from .learn import LearnManager
+        from ..config.feature_configs import LearnConfig
+        
+        if self._learn_config is True:
+            config = LearnConfig()
+        elif isinstance(self._learn_config, dict):
+            config = LearnConfig(**self._learn_config)
+        elif isinstance(self._learn_config, LearnConfig):
+            config = self._learn_config
+        else:
+            return None
+        
+        user_id = self.cfg.get("user_id", "default")
+        self._learn_manager = LearnManager(config=config, user_id=user_id)
+        return self._learn_manager
+    
+    def get_learn_context(self) -> str:
+        """
+        Get learning context suitable for injection into system prompt.
+        
+        Returns empty string if learn is not enabled.
+        """
+        if self.learn is None:
+            return ""
+        return self.learn.to_system_prompt_context()

src/praisonai-agents/praisonaiagents/profiling/__init__.py

@@ -0,0 +1,479 @@
+"""
+PraisonAI Agents Core Profiling Module
+
+Lightweight, protocol-based profiling for the core SDK.
+Zero overhead when disabled. Opt-in only.
+
+Features:
+- Pre-LLM latency breakdown
+- LLM client initialization timing
+- Tool resolution timing
+- Message serialization timing
+- Streaming TTFT tracking
+- Import timing
+- Async-safe
+
+Usage:
+    from praisonaiagents.profiling import Profiler, profile_block
+    
+    # Enable profiling
+    Profiler.enable()
+    
+    # Profile a block
+    with profile_block("my_operation"):
+        do_something()
+    
+    # Get report
+    Profiler.report()
+"""
+
+import time
+import os
+import threading
+from dataclasses import dataclass, field, asdict
+from typing import Dict, List, Optional, Any, Callable
+from contextlib import contextmanager
+import functools
+import json
+
+
+# ============================================================================
+# Data Classes
+# ============================================================================
+
+@dataclass
+class TimingRecord:
+    """Record of a single timing measurement."""
+    name: str
+    duration_ms: float
+    category: str = "block"
+    parent: str = ""
+    timestamp: float = field(default_factory=time.time)
+    metadata: Dict[str, Any] = field(default_factory=dict)
+
+
+@dataclass
+class StreamingRecord:
+    """Record of streaming operation."""
+    name: str
+    ttft_ms: float  # Time to first token
+    total_ms: float
+    chunk_count: int = 0
+    timestamp: float = field(default_factory=time.time)
+
+
+@dataclass
+class PreLLMBreakdown:
+    """Breakdown of pre-LLM latency phases."""
+    agent_init_ms: float = 0.0
+    llm_client_init_ms: float = 0.0
+    tool_resolution_ms: float = 0.0
+    message_build_ms: float = 0.0
+    context_management_ms: float = 0.0
+    knowledge_retrieval_ms: float = 0.0
+    guardrail_check_ms: float = 0.0
+    total_pre_llm_ms: float = 0.0
+    
+    def to_dict(self) -> Dict[str, float]:
+        return asdict(self)
+
+
+# ============================================================================
+# Streaming Tracker
+# ============================================================================
+
+class StreamingTracker:
+    """Track streaming operations for TTFT metrics."""
+    
+    def __init__(self, name: str):
+        self.name = name
+        self._start_time: Optional[float] = None
+        self._first_token_time: Optional[float] = None
+        self._end_time: Optional[float] = None
+        self._chunk_count: int = 0
+    
+    def start(self) -> None:
+        self._start_time = time.perf_counter()
+    
+    def first_token(self) -> None:
+        if self._first_token_time is None:
+            self._first_token_time = time.perf_counter()
+    
+    def chunk(self) -> None:
+        self._chunk_count += 1
+    
+    def end(self) -> None:
+        self._end_time = time.perf_counter()
+        if self._start_time is not None:
+            ttft_ms = 0.0
+            if self._first_token_time is not None:
+                ttft_ms = (self._first_token_time - self._start_time) * 1000
+            total_ms = (self._end_time - self._start_time) * 1000
+            Profiler.record_streaming(
+                name=self.name,
+                ttft_ms=ttft_ms,
+                total_ms=total_ms,
+                chunk_count=self._chunk_count
+            )
+    
+    @property
+    def ttft_ms(self) -> float:
+        if self._start_time and self._first_token_time:
+            return (self._first_token_time - self._start_time) * 1000
+        return 0.0
+    
+    @property
+    def elapsed_ms(self) -> float:
+        if self._start_time:
+            end = self._end_time or time.perf_counter()
+            return (end - self._start_time) * 1000
+        return 0.0
+
+
+# ============================================================================
+# Core Profiler
+# ============================================================================
+
+class Profiler:
+    """
+    Lightweight profiler for PraisonAI Agents core SDK.
+    
+    Thread-safe singleton with zero overhead when disabled.
+    """
+    
+    _lock = threading.Lock()
+    _enabled: bool = False
+    _timings: List[TimingRecord] = []
+    _streaming: List[StreamingRecord] = []
+    _pre_llm_breakdown: Optional[PreLLMBreakdown] = None
+    _current_phases: Dict[str, float] = {}  # Track in-progress phases
+    _phase_stack: List[str] = []  # Track nested phases
+    
+    @classmethod
+    def enable(cls) -> None:
+        """Enable profiling."""
+        cls._enabled = True
+    
+    @classmethod
+    def disable(cls) -> None:
+        """Disable profiling."""
+        cls._enabled = False
+    
+    @classmethod
+    def is_enabled(cls) -> bool:
+        """Check if profiling is enabled."""
+        return cls._enabled or os.environ.get('PRAISONAI_PROFILE', '').lower() in ('1', 'true', 'yes')
+    
+    @classmethod
+    def clear(cls) -> None:
+        """Clear all profiling data."""
+        with cls._lock:
+            cls._timings.clear()
+            cls._streaming.clear()
+            cls._pre_llm_breakdown = None
+            cls._current_phases.clear()
+            cls._phase_stack.clear()
+    
+    @classmethod
+    def record_timing(cls, name: str, duration_ms: float, category: str = "block",
+                      parent: str = "", metadata: Optional[Dict[str, Any]] = None) -> None:
+        """Record a timing measurement."""
+        if not cls.is_enabled():
+            return
+        with cls._lock:
+            cls._timings.append(TimingRecord(
+                name=name,
+                duration_ms=duration_ms,
+                category=category,
+                parent=parent,
+                metadata=metadata or {}
+            ))
+    
+    @classmethod
+    def record_streaming(cls, name: str, ttft_ms: float, total_ms: float,
+                         chunk_count: int = 0) -> None:
+        """Record streaming metrics."""
+        if not cls.is_enabled():
+            return
+        with cls._lock:
+            cls._streaming.append(StreamingRecord(
+                name=name,
+                ttft_ms=ttft_ms,
+                total_ms=total_ms,
+                chunk_count=chunk_count
+            ))
+    
+    @classmethod
+    def start_phase(cls, phase_name: str) -> None:
+        """Start timing a phase (for pre-LLM breakdown)."""
+        if not cls.is_enabled():
+            return
+        with cls._lock:
+            cls._current_phases[phase_name] = time.perf_counter()
+            cls._phase_stack.append(phase_name)
+    
+    @classmethod
+    def end_phase(cls, phase_name: str) -> float:
+        """End timing a phase and return duration in ms."""
+        if not cls.is_enabled():
+            return 0.0
+        with cls._lock:
+            start = cls._current_phases.pop(phase_name, None)
+            if phase_name in cls._phase_stack:
+                cls._phase_stack.remove(phase_name)
+            if start is None:
+                return 0.0
+            duration_ms = (time.perf_counter() - start) * 1000
+            # Record as timing
+            parent = cls._phase_stack[-1] if cls._phase_stack else ""
+            cls._timings.append(TimingRecord(
+                name=phase_name,
+                duration_ms=duration_ms,
+                category="phase",
+                parent=parent
+            ))
+            return duration_ms
+    
+    @classmethod
+    @contextmanager
+    def phase(cls, name: str, category: str = "phase"):
+        """Context manager for timing a phase."""
+        if not cls.is_enabled():
+            yield
+            return
+        
+        start = time.perf_counter()
+        parent = ""
+        with cls._lock:
+            parent = cls._phase_stack[-1] if cls._phase_stack else ""
+            cls._phase_stack.append(name)
+        try:
+            yield
+        finally:
+            duration_ms = (time.perf_counter() - start) * 1000
+            with cls._lock:
+                if name in cls._phase_stack:
+                    cls._phase_stack.remove(name)
+                cls._timings.append(TimingRecord(
+                    name=name,
+                    duration_ms=duration_ms,
+                    category=category,
+                    parent=parent
+                ))
+    
+    @classmethod
+    @contextmanager
+    def streaming(cls, name: str):
+        """Context manager for streaming profiling."""
+        tracker = StreamingTracker(name)
+        tracker.start()
+        try:
+            yield tracker
+        finally:
+            tracker.end()
+    
+    @classmethod
+    def set_pre_llm_breakdown(cls, breakdown: PreLLMBreakdown) -> None:
+        """Set the pre-LLM breakdown data."""
+        if not cls.is_enabled():
+            return
+        with cls._lock:
+            cls._pre_llm_breakdown = breakdown
+    
+    @classmethod
+    def get_timings(cls, category: Optional[str] = None) -> List[TimingRecord]:
+        """Get timing records."""
+        with cls._lock:
+            if category:
+                return [t for t in cls._timings if t.category == category]
+            return cls._timings.copy()
+    
+    @classmethod
+    def get_streaming_records(cls) -> List[StreamingRecord]:
+        """Get streaming records."""
+        with cls._lock:
+            return cls._streaming.copy()
+    
+    @classmethod
+    def get_pre_llm_breakdown(cls) -> Optional[PreLLMBreakdown]:
+        """Get pre-LLM breakdown."""
+        with cls._lock:
+            return cls._pre_llm_breakdown
+    
+    @classmethod
+    def get_summary(cls) -> Dict[str, Any]:
+        """Get profiling summary."""
+        with cls._lock:
+            total_time = sum(t.duration_ms for t in cls._timings)
+            
+            # Group by category
+            by_category: Dict[str, float] = {}
+            for t in cls._timings:
+                by_category[t.category] = by_category.get(t.category, 0) + t.duration_ms
+            
+            # Top slowest
+            slowest = sorted(cls._timings, key=lambda x: x.duration_ms, reverse=True)[:15]
+            
+            # Pre-LLM breakdown
+            pre_llm = cls._pre_llm_breakdown.to_dict() if cls._pre_llm_breakdown else {}
+            
+            # Streaming stats
+            streaming_stats = []
+            for s in cls._streaming:
+                streaming_stats.append({
+                    'name': s.name,
+                    'ttft_ms': s.ttft_ms,
+                    'total_ms': s.total_ms,
+                    'chunk_count': s.chunk_count
+                })
+            
+            return {
+                'total_time_ms': total_time,
+                'timing_count': len(cls._timings),
+                'by_category': by_category,
+                'slowest_operations': [(s.name, s.duration_ms, s.category) for s in slowest],
+                'pre_llm_breakdown': pre_llm,
+                'streaming': streaming_stats
+            }
+    
+    @classmethod
+    def report(cls, output: str = "console") -> str:
+        """Generate and output profiling report."""
+        summary = cls.get_summary()
+        
+        lines = [
+            "=" * 70,
+            "PraisonAI Agents Core Profiling Report",
+            "=" * 70,
+            "",
+            f"Total Time: {summary['total_time_ms']:.2f}ms",
+            f"Timing Records: {summary['timing_count']}",
+            "",
+        ]
+        
+        # Pre-LLM Breakdown (key for debugging the 7s delay)
+        if summary['pre_llm_breakdown']:
+            lines.append("## Pre-LLM Latency Breakdown")
+            lines.append("-" * 50)
+            for phase, ms in summary['pre_llm_breakdown'].items():
+                if ms > 0:
+                    lines.append(f"  {phase:<30}: {ms:>10.2f} ms")
+            lines.append("")
+        
+        # By Category
+        lines.append("## By Category")
+        lines.append("-" * 50)
+        for cat, time_ms in summary['by_category'].items():
+            lines.append(f"  {cat}: {time_ms:.2f}ms")
+        lines.append("")
+        
+        # Slowest Operations
+        lines.append("## Slowest Operations")
+        lines.append("-" * 50)
+        for name, time_ms, cat in summary['slowest_operations']:
+            lines.append(f"  {name:<40}: {time_ms:>10.2f}ms [{cat}]")
+        lines.append("")
+        
+        # Streaming Stats
+        if summary['streaming']:
+            lines.append("## Streaming Metrics")
+            lines.append("-" * 50)
+            for s in summary['streaming']:
+                lines.append(f"  {s['name']}: TTFT={s['ttft_ms']:.2f}ms, Total={s['total_ms']:.2f}ms, Chunks={s['chunk_count']}")
+            lines.append("")
+        
+        lines.append("=" * 70)
+        
+        report_text = "\n".join(lines)
+        
+        if output == "console":
+            print(report_text)
+        
+        return report_text
+    
+    @classmethod
+    def export_json(cls) -> str:
+        """Export profiling data as JSON."""
+        summary = cls.get_summary()
+        with cls._lock:
+            data = {
+                'summary': summary,
+                'timings': [asdict(t) for t in cls._timings],
+                'streaming': [asdict(s) for s in cls._streaming],
+            }
+            return json.dumps(data, indent=2, default=str)
+
+
+# ============================================================================
+# Decorators
+# ============================================================================
+
+def profile_func(func: Optional[Callable] = None, *, category: str = "function"):
+    """Decorator to profile a function."""
+    def decorator(fn: Callable) -> Callable:
+        @functools.wraps(fn)
+        def wrapper(*args, **kwargs):
+            if not Profiler.is_enabled():
+                return fn(*args, **kwargs)
+            start = time.perf_counter()
+            try:
+                return fn(*args, **kwargs)
+            finally:
+                duration_ms = (time.perf_counter() - start) * 1000
+                Profiler.record_timing(fn.__name__, duration_ms, category)
+        return wrapper
+    
+    if func is not None:
+        return decorator(func)
+    return decorator
+
+
+def profile_async(func: Optional[Callable] = None, *, category: str = "async"):
+    """Decorator to profile an async function."""
+    def decorator(fn: Callable) -> Callable:
+        @functools.wraps(fn)
+        async def wrapper(*args, **kwargs):
+            if not Profiler.is_enabled():
+                return await fn(*args, **kwargs)
+            start = time.perf_counter()
+            try:
+                return await fn(*args, **kwargs)
+            finally:
+                duration_ms = (time.perf_counter() - start) * 1000
+                Profiler.record_timing(fn.__name__, duration_ms, category)
+        return wrapper
+    
+    if func is not None:
+        return decorator(func)
+    return decorator
+
+
+@contextmanager
+def profile_block(name: str, category: str = "block"):
+    """Context manager for profiling a block of code."""
+    if not Profiler.is_enabled():
+        yield
+        return
+    
+    start = time.perf_counter()
+    try:
+        yield
+    finally:
+        duration_ms = (time.perf_counter() - start) * 1000
+        Profiler.record_timing(name, duration_ms, category)
+
+
+# ============================================================================
+# Exports
+# ============================================================================
+
+__all__ = [
+    'Profiler',
+    'StreamingTracker',
+    'TimingRecord',
+    'StreamingRecord',
+    'PreLLMBreakdown',
+    'profile_func',
+    'profile_async',
+    'profile_block',
+]

src/praisonai-agents/praisonaiagents/streaming/__init__.py

@@ -0,0 +1,75 @@
+"""
+PraisonAI Streaming Module.
+
+Provides the StreamEvent protocol for pass-through streaming from LLM providers.
+
+This module is lazily loaded to ensure zero performance impact when streaming
+features are not used.
+
+Usage:
+    from praisonaiagents.streaming import (
+        StreamEvent,
+        StreamEventType,
+        StreamMetrics,
+        StreamEventEmitter,
+        create_text_printer_callback,
+    )
+"""
+
+__all__ = [
+    "StreamEvent",
+    "StreamEventType", 
+    "StreamMetrics",
+    "StreamEventEmitter",
+    "StreamCallback",
+    "AsyncStreamCallback",
+    "create_text_printer_callback",
+    "create_metrics_callback",
+    "StreamLogger",
+    "create_logging_callback",
+]
+
+
+def __getattr__(name: str):
+    """Lazy loading of streaming components."""
+    if name in __all__:
+        # Events module exports
+        if name in (
+            "StreamEvent", "StreamEventType", "StreamMetrics", 
+            "StreamEventEmitter", "StreamCallback", "AsyncStreamCallback",
+            "create_text_printer_callback", "create_metrics_callback"
+        ):
+            from .events import (
+                StreamEvent,
+                StreamEventType,
+                StreamMetrics,
+                StreamEventEmitter,
+                StreamCallback,
+                AsyncStreamCallback,
+                create_text_printer_callback,
+                create_metrics_callback,
+            )
+            
+            _events_exports = {
+                "StreamEvent": StreamEvent,
+                "StreamEventType": StreamEventType,
+                "StreamMetrics": StreamMetrics,
+                "StreamEventEmitter": StreamEventEmitter,
+                "StreamCallback": StreamCallback,
+                "AsyncStreamCallback": AsyncStreamCallback,
+                "create_text_printer_callback": create_text_printer_callback,
+                "create_metrics_callback": create_metrics_callback,
+            }
+            return _events_exports.get(name)
+        
+        # Logging module exports
+        if name in ("StreamLogger", "create_logging_callback"):
+            from .logging import StreamLogger, create_logging_callback
+            
+            _logging_exports = {
+                "StreamLogger": StreamLogger,
+                "create_logging_callback": create_logging_callback,
+            }
+            return _logging_exports.get(name)
+    
+    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")

src/praisonai-agents/praisonaiagents/streaming/events.py

@@ -0,0 +1,300 @@
+"""
+StreamEvent protocol for PraisonAI streaming.
+
+This module defines the minimal streaming event protocol that both OpenAI SDK
+and LiteLLM streaming paths map their native chunks/events into.
+
+Design principles:
+- Zero overhead when not used (lazy imports, no global state)
+- Multi-agent safe (per-agent callbacks, no shared mutable state)
+- Async-safe (separate sync/async callback interfaces)
+- Provider-agnostic (same events for OpenAI, LiteLLM, etc.)
+"""
+
+from dataclasses import dataclass, field
+from enum import Enum
+from typing import Any, Dict, List, Optional, Protocol
+import time
+
+
+class StreamEventType(Enum):
+    """Types of streaming events emitted during LLM response streaming."""
+    
+    REQUEST_START = "request_start"      # Before API call is made
+    HEADERS_RECEIVED = "headers_received"  # When HTTP 200 headers arrive
+    FIRST_TOKEN = "first_token"          # First content delta received (TTFT marker)
+    DELTA_TEXT = "delta_text"            # Text content delta
+    DELTA_TOOL_CALL = "delta_tool_call"  # Tool call delta (function name, args chunk)
+    TOOL_CALL_END = "tool_call_end"      # Tool call complete
+    LAST_TOKEN = "last_token"            # Final content delta
+    STREAM_END = "stream_end"            # Stream completed successfully
+    ERROR = "error"                      # Error during streaming
+
+
+@dataclass
+class StreamEvent:
+    """
+    A single streaming event emitted during LLM response streaming.
+    
+    Attributes:
+        type: The type of event (see StreamEventType)
+        timestamp: High-precision timestamp (time.perf_counter())
+        content: Text content for DELTA_TEXT events
+        tool_call: Tool call data for DELTA_TOOL_CALL events
+        metadata: Additional event-specific metadata
+        error: Error message for ERROR events
+    """
+    type: StreamEventType
+    timestamp: float = field(default_factory=time.perf_counter)
+    content: Optional[str] = None
+    tool_call: Optional[Dict[str, Any]] = None
+    metadata: Optional[Dict[str, Any]] = None
+    error: Optional[str] = None
+    
+    def __repr__(self) -> str:
+        parts = [f"StreamEvent({self.type.value}"]
+        if self.content:
+            preview = self.content[:20] + "..." if len(self.content) > 20 else self.content
+            parts.append(f", content={preview!r}")
+        if self.tool_call:
+            parts.append(f", tool_call={self.tool_call.get('name', 'unknown')}")
+        if self.error:
+            parts.append(f", error={self.error!r}")
+        parts.append(")")
+        return "".join(parts)
+
+
+class StreamCallback(Protocol):
+    """Protocol for synchronous stream event callbacks."""
+    
+    def __call__(self, event: StreamEvent) -> None:
+        """Called when a stream event is emitted."""
+        ...
+
+
+class AsyncStreamCallback(Protocol):
+    """Protocol for asynchronous stream event callbacks."""
+    
+    async def __call__(self, event: StreamEvent) -> None:
+        """Called when a stream event is emitted (async)."""
+        ...
+
+
+@dataclass
+class StreamMetrics:
+    """
+    Timing metrics for a streaming response.
+    
+    Tracks key timestamps to measure:
+    - TTFT (Time To First Token): How long until first token arrives
+    - Stream Duration: How long the token streaming takes
+    - Total Time: End-to-end request time
+    
+    All times are in seconds, measured with time.perf_counter() for precision.
+    """
+    request_start: float = 0.0
+    headers_received: float = 0.0
+    first_token: float = 0.0
+    last_token: float = 0.0
+    stream_end: float = 0.0
+    token_count: int = 0
+    
+    @property
+    def ttft(self) -> float:
+        """Time To First Token in seconds (first_token - request_start)."""
+        if self.first_token and self.request_start:
+            return self.first_token - self.request_start
+        return 0.0
+    
+    @property
+    def stream_duration(self) -> float:
+        """Duration of token streaming in seconds (last_token - first_token)."""
+        if self.last_token and self.first_token:
+            return self.last_token - self.first_token
+        return 0.0
+    
+    @property
+    def total_time(self) -> float:
+        """Total request time in seconds (stream_end - request_start)."""
+        if self.stream_end and self.request_start:
+            return self.stream_end - self.request_start
+        return 0.0
+    
+    @property
+    def tokens_per_second(self) -> float:
+        """Token generation rate (tokens / stream_duration)."""
+        if self.stream_duration > 0 and self.token_count > 0:
+            return self.token_count / self.stream_duration
+        return 0.0
+    
+    def update_from_event(self, event: StreamEvent) -> None:
+        """Update metrics from a StreamEvent."""
+        if event.type == StreamEventType.REQUEST_START:
+            self.request_start = event.timestamp
+        elif event.type == StreamEventType.HEADERS_RECEIVED:
+            self.headers_received = event.timestamp
+        elif event.type == StreamEventType.FIRST_TOKEN:
+            self.first_token = event.timestamp
+            self.token_count = 1
+        elif event.type == StreamEventType.DELTA_TEXT:
+            self.token_count += 1
+        elif event.type == StreamEventType.LAST_TOKEN:
+            self.last_token = event.timestamp
+        elif event.type == StreamEventType.STREAM_END:
+            self.stream_end = event.timestamp
+    
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert metrics to dictionary for logging/display."""
+        return {
+            "ttft_ms": round(self.ttft * 1000, 1),
+            "stream_duration_ms": round(self.stream_duration * 1000, 1),
+            "total_time_ms": round(self.total_time * 1000, 1),
+            "token_count": self.token_count,
+            "tokens_per_second": round(self.tokens_per_second, 1),
+        }
+    
+    def format_summary(self) -> str:
+        """Format metrics as a human-readable summary string."""
+        return (
+            f"TTFT: {self.ttft*1000:.0f}ms | "
+            f"Stream: {self.stream_duration*1000:.0f}ms | "
+            f"Total: {self.total_time*1000:.0f}ms | "
+            f"Tokens: {self.token_count} ({self.tokens_per_second:.1f}/s)"
+        )
+
+
+class StreamEventEmitter:
+    """
+    Manages stream event callbacks for an agent or streaming session.
+    
+    Thread-safe and multi-agent safe - each agent has its own emitter instance.
+    Supports both sync and async callbacks.
+    
+    Usage:
+        emitter = StreamEventEmitter()
+        emitter.add_callback(my_callback)
+        
+        # During streaming:
+        emitter.emit(StreamEvent(type=StreamEventType.DELTA_TEXT, content="Hello"))
+    """
+    
+    def __init__(self) -> None:
+        self._callbacks: List[StreamCallback] = []
+        self._async_callbacks: List[AsyncStreamCallback] = []
+        self._metrics: Optional[StreamMetrics] = None
+        self._collect_metrics: bool = False
+    
+    def add_callback(self, callback: StreamCallback) -> None:
+        """Add a synchronous callback to receive stream events."""
+        self._callbacks.append(callback)
+    
+    def remove_callback(self, callback: StreamCallback) -> None:
+        """Remove a synchronous callback."""
+        if callback in self._callbacks:
+            self._callbacks.remove(callback)
+    
+    def add_async_callback(self, callback: AsyncStreamCallback) -> None:
+        """Add an asynchronous callback to receive stream events."""
+        self._async_callbacks.append(callback)
+    
+    def remove_async_callback(self, callback: AsyncStreamCallback) -> None:
+        """Remove an asynchronous callback."""
+        if callback in self._async_callbacks:
+            self._async_callbacks.remove(callback)
+    
+    def enable_metrics(self) -> None:
+        """Enable metrics collection for this streaming session."""
+        self._collect_metrics = True
+        self._metrics = StreamMetrics()
+    
+    def get_metrics(self) -> Optional[StreamMetrics]:
+        """Get collected metrics (None if metrics not enabled)."""
+        return self._metrics
+    
+    def reset_metrics(self) -> None:
+        """Reset metrics for a new streaming session."""
+        if self._collect_metrics:
+            self._metrics = StreamMetrics()
+    
+    def emit(self, event: StreamEvent) -> None:
+        """
+        Emit a stream event to all registered callbacks.
+        
+        This is the synchronous emission path. For async streaming,
+        use emit_async().
+        """
+        # Update metrics if enabled
+        if self._collect_metrics and self._metrics:
+            self._metrics.update_from_event(event)
+        
+        # Call all sync callbacks
+        for callback in self._callbacks:
+            try:
+                callback(event)
+            except Exception:
+                # Don't let callback errors break streaming
+                pass
+    
+    async def emit_async(self, event: StreamEvent) -> None:
+        """
+        Emit a stream event to all registered callbacks (async version).
+        
+        Calls both sync and async callbacks.
+        """
+        # Update metrics if enabled
+        if self._collect_metrics and self._metrics:
+            self._metrics.update_from_event(event)
+        
+        # Call sync callbacks
+        for callback in self._callbacks:
+            try:
+                callback(event)
+            except Exception:
+                pass
+        
+        # Call async callbacks
+        for callback in self._async_callbacks:
+            try:
+                await callback(event)
+            except Exception:
+                pass
+    
+    @property
+    def has_callbacks(self) -> bool:
+        """Check if any callbacks are registered."""
+        return bool(self._callbacks or self._async_callbacks)
+
+
+def create_text_printer_callback(flush: bool = True) -> StreamCallback:
+    """
+    Create a simple callback that prints text deltas to stdout.
+    
+    Args:
+        flush: Whether to flush stdout after each print (default True)
+    
+    Returns:
+        A StreamCallback that prints DELTA_TEXT content
+    """
+    def printer(event: StreamEvent) -> None:
+        if event.type == StreamEventType.DELTA_TEXT and event.content:
+            print(event.content, end="", flush=flush)
+        elif event.type == StreamEventType.STREAM_END:
+            print()  # Newline at end
+    
+    return printer
+
+
+def create_metrics_callback(metrics: StreamMetrics) -> StreamCallback:
+    """
+    Create a callback that updates a StreamMetrics instance.
+    
+    Args:
+        metrics: The StreamMetrics instance to update
+    
+    Returns:
+        A StreamCallback that updates the metrics
+    """
+    def updater(event: StreamEvent) -> None:
+        metrics.update_from_event(event)
+    
+    return updater

src/praisonai-agents/praisonaiagents/streaming/logging.py

@@ -0,0 +1,207 @@
+"""
+Streaming logging utilities for PraisonAI.
+
+Provides guarded logging for streaming events with clear timestamps
+for request_start, headers_received, first_token, last_token, and stream_end.
+
+Usage:
+    from praisonaiagents.streaming.logging import StreamLogger
+    
+    logger = StreamLogger(verbose=True, metrics=True)
+    logger.log_request_start(model="gpt-4o-mini")
+    # ... streaming ...
+    logger.log_first_token()
+    # ... more tokens ...
+    logger.log_stream_end()
+    print(logger.get_metrics_summary())
+"""
+
+import logging
+import time
+from typing import Optional, Dict, Any
+
+from .events import StreamMetrics, StreamEvent, StreamEventType
+
+
+class StreamLogger:
+    """
+    Guarded logging for streaming events with timing metrics.
+    
+    Only logs when verbose=True or metrics=True to ensure zero overhead
+    when logging is disabled.
+    
+    Attributes:
+        verbose: Whether to log detailed streaming events
+        metrics: Whether to collect and display timing metrics
+    """
+    
+    def __init__(
+        self,
+        verbose: bool = False,
+        metrics: bool = False,
+        logger: Optional[logging.Logger] = None
+    ):
+        self.verbose = verbose
+        self.metrics = metrics
+        self._logger = logger or logging.getLogger(__name__)
+        self._metrics = StreamMetrics() if metrics else None
+        self._enabled = verbose or metrics
+    
+    def log_request_start(
+        self,
+        model: Optional[str] = None,
+        message_count: Optional[int] = None
+    ) -> None:
+        """Log when the streaming request is initiated."""
+        if not self._enabled:
+            return
+        
+        timestamp = time.perf_counter()
+        if self._metrics:
+            self._metrics.request_start = timestamp
+        
+        if self.verbose:
+            extra = f" model={model}" if model else ""
+            extra += f" messages={message_count}" if message_count else ""
+            self._logger.info(f"[STREAM] REQUEST_START{extra}")
+    
+    def log_headers_received(self) -> None:
+        """Log when HTTP response headers are received."""
+        if not self._enabled:
+            return
+        
+        timestamp = time.perf_counter()
+        if self._metrics:
+            self._metrics.headers_received = timestamp
+        
+        if self.verbose:
+            self._logger.info("[STREAM] HEADERS_RECEIVED")
+    
+    def log_first_token(self, content: Optional[str] = None) -> None:
+        """Log when the first content token is received (TTFT marker)."""
+        if not self._enabled:
+            return
+        
+        timestamp = time.perf_counter()
+        if self._metrics:
+            self._metrics.first_token = timestamp
+            self._metrics.token_count = 1
+        
+        if self.verbose:
+            preview = content[:20] + "..." if content and len(content) > 20 else content
+            self._logger.info(f"[STREAM] FIRST_TOKEN content={preview!r}")
+    
+    def log_delta_text(self, content: Optional[str] = None) -> None:
+        """Log a text delta (only when verbose, not for every token)."""
+        if not self._enabled:
+            return
+        
+        if self._metrics:
+            self._metrics.token_count += 1
+        
+        # Don't log every delta to avoid noise - just track metrics
+    
+    def log_last_token(self) -> None:
+        """Log when the last content token is received."""
+        if not self._enabled:
+            return
+        
+        timestamp = time.perf_counter()
+        if self._metrics:
+            self._metrics.last_token = timestamp
+        
+        if self.verbose:
+            self._logger.info("[STREAM] LAST_TOKEN")
+    
+    def log_stream_end(self, chunk_count: Optional[int] = None) -> None:
+        """Log when the stream ends."""
+        if not self._enabled:
+            return
+        
+        timestamp = time.perf_counter()
+        if self._metrics:
+            self._metrics.stream_end = timestamp
+        
+        if self.verbose:
+            extra = f" chunks={chunk_count}" if chunk_count else ""
+            self._logger.info(f"[STREAM] STREAM_END{extra}")
+    
+    def log_error(self, error: str) -> None:
+        """Log a streaming error."""
+        if self.verbose:
+            self._logger.error(f"[STREAM] ERROR: {error}")
+    
+    def get_metrics(self) -> Optional[StreamMetrics]:
+        """Get the collected metrics (None if metrics not enabled)."""
+        return self._metrics
+    
+    def get_metrics_summary(self) -> str:
+        """Get a formatted summary of streaming metrics."""
+        if not self._metrics:
+            return "Metrics not enabled"
+        return self._metrics.format_summary()
+    
+    def get_metrics_dict(self) -> Optional[Dict[str, Any]]:
+        """Get metrics as a dictionary."""
+        if not self._metrics:
+            return None
+        return self._metrics.to_dict()
+    
+    def update_from_event(self, event: StreamEvent) -> None:
+        """Update logger state from a StreamEvent."""
+        if not self._enabled:
+            return
+        
+        if self._metrics:
+            self._metrics.update_from_event(event)
+        
+        if self.verbose:
+            if event.type == StreamEventType.REQUEST_START:
+                self.log_request_start()
+            elif event.type == StreamEventType.HEADERS_RECEIVED:
+                self.log_headers_received()
+            elif event.type == StreamEventType.FIRST_TOKEN:
+                self.log_first_token(event.content)
+            elif event.type == StreamEventType.LAST_TOKEN:
+                self.log_last_token()
+            elif event.type == StreamEventType.STREAM_END:
+                self.log_stream_end()
+            elif event.type == StreamEventType.ERROR:
+                self.log_error(event.error or "Unknown error")
+
+
+def create_logging_callback(
+    verbose: bool = False,
+    metrics: bool = False,
+    logger: Optional[logging.Logger] = None
+) -> tuple:
+    """
+    Create a StreamLogger and a callback function for use with streaming.
+    
+    Args:
+        verbose: Whether to log detailed events
+        metrics: Whether to collect timing metrics
+        logger: Optional custom logger
+    
+    Returns:
+        Tuple of (StreamLogger, callback_function)
+    
+    Usage:
+        stream_logger, callback = create_logging_callback(verbose=True, metrics=True)
+        
+        # Pass callback to streaming method
+        response = client.process_stream_response(
+            ...,
+            stream_callback=callback,
+            emit_events=True
+        )
+        
+        # After streaming, get metrics
+        print(stream_logger.get_metrics_summary())
+    """
+    stream_logger = StreamLogger(verbose=verbose, metrics=metrics, logger=logger)
+    
+    def callback(event: StreamEvent) -> None:
+        stream_logger.update_from_event(event)
+    
+    return stream_logger, callback

src/praisonai-agents/praisonaiagents/task/task.py

@@ -326,7 +326,7 @@ class Task:
                         if hasattr(self.agent.llm_instance, 'model'):
                             llm_model = self.agent.llm_instance.model
                         else:
-                            llm_model = "gpt-5-nano"  # Default fallback
+                            llm_model = "gpt-4o-mini"  # Default fallback
                     elif hasattr(self.agent, 'llm') and self.agent.llm:
                         # For standard model strings
                         llm_model = self.agent.llm

src/praisonai-agents/praisonaiagents/telemetry/integration.py

@@ -26,16 +26,41 @@ _telemetry_executor = None
 _telemetry_queue = None
 _queue_processor_running = False
 _queue_lock = threading.Lock()
+_atexit_registered = False
+
+def _atexit_cleanup():
+    """Cleanup handler registered with atexit to prevent hanging on exit."""
+    global _queue_processor_running, _telemetry_executor, _telemetry_queue
+    
+    # Signal queue processor to stop
+    _queue_processor_running = False
+    
+    # Shutdown executor without waiting (to prevent hang)
+    if _telemetry_executor is not None:
+        try:
+            _telemetry_executor.shutdown(wait=False)
+        except:
+            pass
+        _telemetry_executor = None
+    
+    _telemetry_queue = None
 
 def _get_telemetry_executor():
     """Get or create the shared telemetry thread pool executor."""
-    global _telemetry_executor
+    global _telemetry_executor, _atexit_registered
     if _telemetry_executor is None:
         # Use a small thread pool to avoid resource overhead
         _telemetry_executor = concurrent.futures.ThreadPoolExecutor(
             max_workers=2, 
             thread_name_prefix="telemetry"
         )
+        
+        # Register atexit cleanup to prevent hanging on exit
+        if not _atexit_registered:
+            import atexit
+            atexit.register(_atexit_cleanup)
+            _atexit_registered = True
+            
     return _telemetry_executor
 
 def _get_telemetry_queue():
@@ -48,8 +73,13 @@ def _get_telemetry_queue():
     with _queue_lock:
         if not _queue_processor_running:
             _queue_processor_running = True
-            executor = _get_telemetry_executor()
-            executor.submit(_process_telemetry_queue)
+            # Use a daemon thread directly instead of executor to prevent hang
+            processor_thread = threading.Thread(
+                target=_process_telemetry_queue,
+                name="telemetry-queue-processor",
+                daemon=True  # Daemon thread will be killed on exit
+            )
+            processor_thread.start()
     
     return _telemetry_queue
 
@@ -61,12 +91,22 @@ def _process_telemetry_queue():
     
     try:
         while _queue_processor_running:
+            # Check if interpreter is shutting down
+            try:
+                import sys
+                if hasattr(sys, 'is_finalizing') and sys.is_finalizing():
+                    break
+            except:
+                break
+                
             events = []
             deadline = time.time() + batch_timeout
             
             # Collect events until batch size or timeout
             while len(events) < batch_size and time.time() < deadline:
                 try:
+                    if _telemetry_queue is None:
+                        break
                     event = _telemetry_queue.get(timeout=0.1)
                     events.append(event)
                     _telemetry_queue.task_done()

src/praisonai-agents/praisonaiagents/telemetry/performance_monitor.py

@@ -381,6 +381,10 @@ class PerformanceMonitor:
         Returns:
             List of flow events showing function execution order and timing
         """
+        # If monitoring is disabled, return empty results
+        if self._monitoring_disabled:
+            return []
+            
         with self._lock:
             flow = []
             for event in self._function_flow.copy():
@@ -400,6 +404,10 @@ class PerformanceMonitor:
     
     def get_active_calls(self) -> Dict[str, Any]:
         """Get information about currently executing functions."""
+        # If monitoring is disabled, return empty results
+        if self._monitoring_disabled:
+            return {}
+            
         with self._lock:
             active = {}
             current_time = time.time()
@@ -417,6 +425,10 @@ class PerformanceMonitor:
     
     def get_slowest_functions(self, limit: int = 10) -> List[Dict[str, Any]]:
         """Get the slowest performing functions."""
+        # If monitoring is disabled, return empty results
+        if self._monitoring_disabled:
+            return []
+            
         with self._lock:
             functions = []
             for name, stats in self._function_stats.items():
@@ -435,6 +447,10 @@ class PerformanceMonitor:
     
     def get_slowest_api_calls(self, limit: int = 10) -> List[Dict[str, Any]]:
         """Get the slowest performing API calls."""
+        # If monitoring is disabled, return empty results
+        if self._monitoring_disabled:
+            return []
+            
         with self._lock:
             apis = []
             for name, stats in self._api_calls.items():
@@ -539,6 +555,10 @@ class PerformanceMonitor:
     
     def clear_statistics(self):
         """Clear all performance statistics."""
+        # If monitoring is disabled, nothing to clear
+        if self._monitoring_disabled:
+            return
+            
         with self._lock:
             self._function_stats.clear()
             self._api_calls.clear()

src/praisonai-agents/praisonaiagents/telemetry/telemetry.py

@@ -691,7 +691,7 @@ def force_shutdown_telemetry():
 def enable_telemetry():
     """Programmatically enable telemetry (if not disabled by environment)."""
     global _telemetry_instance
-    if not _TELEMETRY_DISABLED:
+    if not _is_telemetry_disabled():
         if _telemetry_instance:
             _telemetry_instance.enabled = True
         else:

src/praisonai-agents/praisonaiagents/tools/train/data/generatecot.py

@@ -27,7 +27,7 @@ class GenerateCOT:
     def __init__(
         self,
         qa_pairs: Optional[Dict[str, str]] = None,
-        model: str = "gpt-5-nano",
+        model: str = "gpt-4o-mini",
         api_key: Optional[str] = None,
         max_attempts: int = 3,
         verbose: bool = True,

src/praisonai-agents/praisonaiagents/trace/__init__.py

@@ -0,0 +1,68 @@
+"""
+Trace Module for PraisonAI Agents.
+
+Provides lightweight action tracing for the `output="actions"` mode.
+Shows agent lifecycle events, tool calls, and final output without
+verbose internal details.
+
+Zero Performance Impact:
+- All imports are lazy loaded via __getattr__
+- NoOpSink is the default (zero overhead when not used)
+- Disabled emitter has near-zero overhead
+
+Usage:
+    from praisonaiagents import Agent
+    
+    # Simple usage - shows action trace
+    agent = Agent(instructions="...", output="actions")
+    agent.start("Do something")
+    
+    # Advanced - capture to file
+    from praisonaiagents.trace import ActionTraceConfig
+    
+    agent = Agent(
+        instructions="...",
+        output="actions",
+        trace=ActionTraceConfig(
+            sink_type="jsonl",
+            file_path="trace.jsonl",
+        )
+    )
+"""
+
+__all__ = [
+    # Core types
+    "ActionEvent",
+    "ActionEventType",
+    "ActionTraceConfig",
+    # Sinks
+    "TraceSink",
+    "NoOpSink",
+    "ListSink",
+    # Emitter
+    "TraceEmitter",
+    # Redaction
+    "redact_dict",
+    "REDACT_KEYS",
+]
+
+
+def __getattr__(name: str):
+    """Lazy load module components to avoid import overhead."""
+    if name in ("ActionEvent", "ActionEventType", "ActionTraceConfig"):
+        from .protocol import ActionEvent, ActionEventType, ActionTraceConfig
+        return locals()[name]
+    
+    if name in ("TraceSink", "NoOpSink", "ListSink"):
+        from .protocol import TraceSink, NoOpSink, ListSink
+        return locals()[name]
+    
+    if name == "TraceEmitter":
+        from .protocol import TraceEmitter
+        return TraceEmitter
+    
+    if name in ("redact_dict", "REDACT_KEYS"):
+        from .redact import redact_dict, REDACT_KEYS
+        return locals()[name]
+    
+    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")

src/praisonai-agents/praisonaiagents/trace/protocol.py

@@ -0,0 +1,398 @@
+"""
+Action Trace Protocol for PraisonAI Agents.
+
+Lightweight protocol for tracing agent actions. Uses only stdlib
+to ensure zero dependencies and minimal overhead.
+
+Schema Version: 1.0
+"""
+
+import json
+import time
+from dataclasses import dataclass, field
+from enum import Enum
+from typing import Any, Dict, List, Optional, Protocol, runtime_checkable
+
+
+# Schema version for forward compatibility
+SCHEMA_VERSION = "1.0"
+
+
+class ActionEventType(str, Enum):
+    """Types of action events."""
+    AGENT_START = "agent_start"
+    AGENT_END = "agent_end"
+    TOOL_START = "tool_start"
+    TOOL_END = "tool_end"
+    ERROR = "error"
+    OUTPUT = "output"
+
+
+@dataclass
+class ActionEvent:
+    """
+    Lightweight action trace event.
+    
+    Represents a single action in the agent execution trace.
+    Designed for minimal memory footprint and fast serialization.
+    """
+    event_type: str
+    timestamp: float
+    schema_version: str = SCHEMA_VERSION
+    agent_id: Optional[str] = None
+    agent_name: Optional[str] = None
+    tool_name: Optional[str] = None
+    tool_args: Optional[Dict[str, Any]] = None
+    tool_result_summary: Optional[str] = None
+    duration_ms: Optional[float] = None
+    status: Optional[str] = None  # ok, error
+    error_message: Optional[str] = None
+    metadata: Dict[str, Any] = field(default_factory=dict)
+    
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert to dictionary, excluding None values."""
+        result = {
+            "schema_version": self.schema_version,
+            "event_type": self.event_type,
+            "timestamp": self.timestamp,
+        }
+        
+        # Add optional fields only if set
+        if self.agent_id is not None:
+            result["agent_id"] = self.agent_id
+        if self.agent_name is not None:
+            result["agent_name"] = self.agent_name
+        if self.tool_name is not None:
+            result["tool_name"] = self.tool_name
+        if self.tool_args is not None:
+            result["tool_args"] = self.tool_args
+        if self.tool_result_summary is not None:
+            result["tool_result_summary"] = self.tool_result_summary
+        if self.duration_ms is not None:
+            result["duration_ms"] = self.duration_ms
+        if self.status is not None:
+            result["status"] = self.status
+        if self.error_message is not None:
+            result["error_message"] = self.error_message
+        if self.metadata:
+            result["metadata"] = self.metadata
+        
+        return result
+    
+    def to_json(self) -> str:
+        """Convert to JSON string."""
+        return json.dumps(self.to_dict(), default=str)
+
+
+@runtime_checkable
+class TraceSink(Protocol):
+    """
+    Protocol for trace event sinks.
+    
+    Implementations receive action events and handle them
+    (e.g., print to stdout, write to file, send to server).
+    """
+    
+    def emit(self, event: ActionEvent) -> None:
+        """Emit an action event."""
+        ...
+    
+    def flush(self) -> None:
+        """Flush any buffered events."""
+        ...
+    
+    def close(self) -> None:
+        """Close the sink and release resources."""
+        ...
+
+
+class NoOpSink:
+    """
+    No-operation sink that discards all events.
+    
+    Used as the default sink when tracing is disabled.
+    Has near-zero overhead.
+    """
+    
+    __slots__ = ()
+    
+    def emit(self, event: ActionEvent) -> None:
+        """Discard the event."""
+        pass
+    
+    def flush(self) -> None:
+        """No-op."""
+        pass
+    
+    def close(self) -> None:
+        """No-op."""
+        pass
+
+
+class ListSink:
+    """
+    Sink that collects events in a list.
+    
+    Useful for testing and programmatic access to trace events.
+    """
+    
+    __slots__ = ("_events",)
+    
+    def __init__(self):
+        self._events: List[ActionEvent] = []
+    
+    def emit(self, event: ActionEvent) -> None:
+        """Add event to the list."""
+        self._events.append(event)
+    
+    def flush(self) -> None:
+        """No-op for list sink."""
+        pass
+    
+    def close(self) -> None:
+        """No-op for list sink."""
+        pass
+    
+    def get_events(self) -> List[ActionEvent]:
+        """Get all collected events."""
+        return self._events.copy()
+    
+    def clear(self) -> None:
+        """Clear all collected events."""
+        self._events.clear()
+
+
+@dataclass
+class ActionTraceConfig:
+    """
+    Configuration for action tracing.
+    
+    Usage:
+        # Default (disabled)
+        config = ActionTraceConfig()
+        
+        # Enable with stdout
+        config = ActionTraceConfig(enabled=True, sink_type="stdout")
+        
+        # Enable with JSONL file
+        config = ActionTraceConfig(
+            enabled=True,
+            sink_type="jsonl",
+            file_path="trace.jsonl",
+        )
+    """
+    enabled: bool = True
+    redact: bool = True
+    compact: bool = False
+    sink_type: str = "noop"  # noop, stdout, jsonl, file
+    file_path: Optional[str] = None
+    
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert to dictionary."""
+        return {
+            "enabled": self.enabled,
+            "redact": self.redact,
+            "compact": self.compact,
+            "sink_type": self.sink_type,
+            "file_path": self.file_path,
+        }
+
+
+class TraceEmitter:
+    """
+    Emitter for action trace events.
+    
+    Provides a high-level API for emitting trace events.
+    Handles redaction and sink management.
+    
+    Usage:
+        sink = ListSink()
+        emitter = TraceEmitter(sink=sink)
+        
+        emitter.agent_start("researcher")
+        emitter.tool_start("search", {"query": "AI"})
+        emitter.tool_end("search", duration_ms=100, status="ok")
+        emitter.agent_end("researcher", duration_ms=500)
+    """
+    
+    __slots__ = ("_sink", "_enabled", "_redact", "_agent_starts")
+    
+    def __init__(
+        self,
+        sink: Optional[TraceSink] = None,
+        enabled: bool = True,
+        redact: bool = True,
+    ):
+        self._sink = sink or NoOpSink()
+        self._enabled = enabled
+        self._redact = redact
+        self._agent_starts: Dict[str, float] = {}  # Track agent start times
+    
+    def _emit(self, event: ActionEvent) -> None:
+        """Internal emit with enabled check."""
+        if not self._enabled:
+            return
+        self._sink.emit(event)
+    
+    def _redact_args(self, args: Optional[Dict[str, Any]]) -> Optional[Dict[str, Any]]:
+        """Redact sensitive data from args if enabled."""
+        if args is None or not self._redact:
+            return args
+        
+        from .redact import redact_dict
+        return redact_dict(args)
+    
+    def agent_start(
+        self,
+        agent_name: str,
+        agent_id: Optional[str] = None,
+        metadata: Optional[Dict[str, Any]] = None,
+    ) -> None:
+        """Emit agent start event."""
+        ts = time.time()
+        self._agent_starts[agent_name] = ts
+        
+        self._emit(ActionEvent(
+            event_type=ActionEventType.AGENT_START.value,
+            timestamp=ts,
+            agent_name=agent_name,
+            agent_id=agent_id,
+            metadata=metadata or {},
+        ))
+    
+    def agent_end(
+        self,
+        agent_name: str,
+        agent_id: Optional[str] = None,
+        duration_ms: Optional[float] = None,
+        status: str = "ok",
+        metadata: Optional[Dict[str, Any]] = None,
+    ) -> None:
+        """Emit agent end event."""
+        ts = time.time()
+        
+        # Calculate duration if not provided
+        if duration_ms is None and agent_name in self._agent_starts:
+            start_ts = self._agent_starts.pop(agent_name, None)
+            if start_ts:
+                duration_ms = (ts - start_ts) * 1000
+        
+        self._emit(ActionEvent(
+            event_type=ActionEventType.AGENT_END.value,
+            timestamp=ts,
+            agent_name=agent_name,
+            agent_id=agent_id,
+            duration_ms=duration_ms,
+            status=status,
+            metadata=metadata or {},
+        ))
+    
+    def tool_start(
+        self,
+        tool_name: str,
+        tool_args: Optional[Dict[str, Any]] = None,
+        agent_name: Optional[str] = None,
+        agent_id: Optional[str] = None,
+        metadata: Optional[Dict[str, Any]] = None,
+    ) -> None:
+        """Emit tool start event."""
+        self._emit(ActionEvent(
+            event_type=ActionEventType.TOOL_START.value,
+            timestamp=time.time(),
+            tool_name=tool_name,
+            tool_args=self._redact_args(tool_args),
+            agent_name=agent_name,
+            agent_id=agent_id,
+            metadata=metadata or {},
+        ))
+    
+    def tool_end(
+        self,
+        tool_name: str,
+        duration_ms: float,
+        status: str = "ok",
+        result_summary: Optional[str] = None,
+        agent_name: Optional[str] = None,
+        agent_id: Optional[str] = None,
+        error_message: Optional[str] = None,
+        metadata: Optional[Dict[str, Any]] = None,
+    ) -> None:
+        """Emit tool end event."""
+        self._emit(ActionEvent(
+            event_type=ActionEventType.TOOL_END.value,
+            timestamp=time.time(),
+            tool_name=tool_name,
+            duration_ms=duration_ms,
+            status=status,
+            tool_result_summary=result_summary,
+            agent_name=agent_name,
+            agent_id=agent_id,
+            error_message=error_message,
+            metadata=metadata or {},
+        ))
+    
+    def error(
+        self,
+        message: str,
+        tool_name: Optional[str] = None,
+        agent_name: Optional[str] = None,
+        agent_id: Optional[str] = None,
+        metadata: Optional[Dict[str, Any]] = None,
+    ) -> None:
+        """Emit error event."""
+        self._emit(ActionEvent(
+            event_type=ActionEventType.ERROR.value,
+            timestamp=time.time(),
+            error_message=message,
+            tool_name=tool_name,
+            agent_name=agent_name,
+            agent_id=agent_id,
+            status="error",
+            metadata=metadata or {},
+        ))
+    
+    def output(
+        self,
+        content: str,
+        agent_name: Optional[str] = None,
+        agent_id: Optional[str] = None,
+        metadata: Optional[Dict[str, Any]] = None,
+    ) -> None:
+        """Emit final output event."""
+        # Truncate long output for summary
+        summary = content[:500] + "..." if len(content) > 500 else content
+        
+        self._emit(ActionEvent(
+            event_type=ActionEventType.OUTPUT.value,
+            timestamp=time.time(),
+            tool_result_summary=summary,
+            agent_name=agent_name,
+            agent_id=agent_id,
+            metadata=metadata or {},
+        ))
+    
+    def flush(self) -> None:
+        """Flush the sink."""
+        self._sink.flush()
+    
+    def close(self) -> None:
+        """Close the sink."""
+        self._sink.close()
+
+
+# Global default emitter (NoOp by default)
+_default_emitter: Optional[TraceEmitter] = None
+
+
+def get_default_emitter() -> TraceEmitter:
+    """Get the default trace emitter."""
+    global _default_emitter
+    if _default_emitter is None:
+        _default_emitter = TraceEmitter(sink=NoOpSink(), enabled=False)
+    return _default_emitter
+
+
+def set_default_emitter(emitter: TraceEmitter) -> None:
+    """Set the default trace emitter."""
+    global _default_emitter
+    _default_emitter = emitter

src/praisonai-agents/praisonaiagents/trace/redact.py

@@ -0,0 +1,149 @@
+"""
+Redaction utilities for action trace.
+
+Provides deterministic redaction of sensitive data in tool arguments
+and results. Uses only stdlib for zero dependencies.
+"""
+
+from typing import Any, Dict, Set
+import re
+
+# Keys that should always be redacted (case-insensitive matching)
+REDACT_KEYS: Set[str] = {
+    # API keys and tokens
+    "api_key",
+    "apikey",
+    "api-key",
+    "secret",
+    "secret_key",
+    "secretkey",
+    "private_key",
+    "privatekey",
+    # Authentication
+    "password",
+    "passwd",
+    "pwd",
+    "token",
+    "bearer",
+    "authorization",
+    "auth",
+    "auth_token",
+    "access_token",
+    "refresh_token",
+    # Session and cookies
+    "cookie",
+    "session",
+    "session_id",
+    "sessionid",
+    "credential",
+    "credentials",
+    # Database
+    "connection_string",
+    "connectionstring",
+    "database_url",
+    "db_url",
+    "dburl",
+    "db_password",
+    "db_pass",
+    # Cloud provider specific
+    "aws_secret",
+    "aws_secret_access_key",
+    "azure_key",
+    "gcp_key",
+    "openai_api_key",
+    "anthropic_api_key",
+}
+
+# Compiled regex for case-insensitive matching
+_REDACT_PATTERN = re.compile(
+    r"(" + "|".join(re.escape(k) for k in REDACT_KEYS) + r")",
+    re.IGNORECASE
+)
+
+REDACTED_VALUE = "[REDACTED]"
+
+
+def _should_redact(key: str) -> bool:
+    """Check if a key should be redacted."""
+    # Normalize key for comparison
+    normalized = key.lower().replace("-", "_").replace(" ", "_")
+    
+    # Check exact match first
+    if normalized in REDACT_KEYS:
+        return True
+    
+    # Check if any redact key is contained in the normalized key
+    for redact_key in REDACT_KEYS:
+        if redact_key in normalized:
+            return True
+    
+    return False
+
+
+def redact_dict(data: Dict[str, Any], enabled: bool = True) -> Dict[str, Any]:
+    """
+    Redact sensitive values from a dictionary.
+    
+    Args:
+        data: Dictionary to redact
+        enabled: If False, returns data unchanged
+        
+    Returns:
+        New dictionary with sensitive values redacted
+    """
+    if not enabled:
+        return data
+    
+    if not data:
+        return {}
+    
+    return _redact_value(data)
+
+
+def _redact_value(value: Any) -> Any:
+    """Recursively redact sensitive values."""
+    if isinstance(value, dict):
+        return {k: _redact_key_value(k, v) for k, v in value.items()}
+    elif isinstance(value, list):
+        return [_redact_value(item) for item in value]
+    else:
+        return value
+
+
+def _redact_key_value(key: str, value: Any) -> Any:
+    """Redact a value if its key is sensitive."""
+    if _should_redact(key):
+        return REDACTED_VALUE
+    
+    # Recursively process nested structures
+    return _redact_value(value)
+
+
+def redact_string(text: str, enabled: bool = True) -> str:
+    """
+    Redact potential secrets from a string.
+    
+    This is a best-effort function that looks for common patterns
+    like "api_key=xxx" or "password: xxx" in strings.
+    
+    Args:
+        text: String to redact
+        enabled: If False, returns text unchanged
+        
+    Returns:
+        String with potential secrets redacted
+    """
+    if not enabled or not text:
+        return text
+    
+    # Pattern for key=value or key: value
+    patterns = [
+        (r'(["\']?)(' + '|'.join(re.escape(k) for k in REDACT_KEYS) + r')(["\']?\s*[:=]\s*["\']?)([^"\'\s,}\]]+)', 
+         r'\1\2\3[REDACTED]'),
+    ]
+    
+    result = text
+    for pattern, replacement in patterns:
+        result = re.sub(pattern, replacement, result, flags=re.IGNORECASE)
+    
+    return result

src/praisonai-agents/profile-streaming-delay.py

@@ -0,0 +1,232 @@
+#!/usr/bin/env python3
+"""
+Detailed profiler to identify the exact source of streaming delay.
+Traces every step from agent.start() to HTTP request.
+"""
+
+import os
+import sys
+import time
+import functools
+
+os.environ['LOGLEVEL'] = 'DEBUG'
+
+# Microsecond precision timestamps
+def ts():
+    return time.perf_counter()
+
+start_time = ts()
+
+def log(msg):
+    elapsed = (ts() - start_time) * 1000
+    print(f"[{elapsed:8.2f}ms] {msg}")
+
+log("Script starting...")
+
+# ============================================================
+# PATCH 1: OpenAI client initialization
+# ============================================================
+import openai
+
+original_openai_init = openai.OpenAI.__init__
+
+@functools.wraps(original_openai_init)
+def traced_openai_init(self, *args, **kwargs):
+    t0 = ts()
+    result = original_openai_init(self, *args, **kwargs)
+    log(f"OpenAI.__init__() took {(ts()-t0)*1000:.2f}ms")
+    return result
+
+openai.OpenAI.__init__ = traced_openai_init
+
+# ============================================================
+# PATCH 2: OpenAI completions.create
+# ============================================================
+original_create = openai.resources.chat.completions.Completions.create
+
+@functools.wraps(original_create)
+def traced_create(self, *args, **kwargs):
+    log(f">>> OpenAI.create() CALLED - stream={kwargs.get('stream', False)}")
+    t0 = ts()
+    result = original_create(self, *args, **kwargs)
+    log(f"<<< OpenAI.create() RETURNED in {(ts()-t0)*1000:.2f}ms")
+    return result
+
+openai.resources.chat.completions.Completions.create = traced_create
+
+# ============================================================
+# PATCH 3: httpx.Client.send (actual HTTP request)
+# ============================================================
+import httpx
+
+original_send = httpx.Client.send
+
+@functools.wraps(original_send)
+def traced_send(self, request, *args, **kwargs):
+    log(f">>> HTTPX.send() {request.method} {request.url}")
+    t0 = ts()
+    result = original_send(self, request, *args, **kwargs)
+    log(f"<<< HTTPX.send() returned in {(ts()-t0)*1000:.2f}ms - status={result.status_code}")
+    return result
+
+httpx.Client.send = traced_send
+
+# ============================================================
+# PATCH 4: Rich Live display
+# ============================================================
+from rich.live import Live
+
+original_live_init = Live.__init__
+original_live_enter = Live.__enter__
+original_live_exit = Live.__exit__
+
+@functools.wraps(original_live_init)
+def traced_live_init(self, *args, **kwargs):
+    log(f">>> Live.__init__() called")
+    t0 = ts()
+    result = original_live_init(self, *args, **kwargs)
+    log(f"<<< Live.__init__() took {(ts()-t0)*1000:.2f}ms")
+    return result
+
+@functools.wraps(original_live_enter)
+def traced_live_enter(self):
+    log(f">>> Live.__enter__() called")
+    t0 = ts()
+    result = original_live_enter(self)
+    log(f"<<< Live.__enter__() took {(ts()-t0)*1000:.2f}ms")
+    return result
+
+@functools.wraps(original_live_exit)
+def traced_live_exit(self, *args):
+    log(f">>> Live.__exit__() called")
+    t0 = ts()
+    result = original_live_exit(self, *args)
+    log(f"<<< Live.__exit__() took {(ts()-t0)*1000:.2f}ms")
+    return result
+
+Live.__init__ = traced_live_init
+Live.__enter__ = traced_live_enter
+Live.__exit__ = traced_live_exit
+
+log("Patches applied, importing praisonaiagents...")
+
+# ============================================================
+# Import and trace Agent
+# ============================================================
+t0 = ts()
+from praisonaiagents import Agent
+log(f"Agent import took {(ts()-t0)*1000:.2f}ms")
+
+# ============================================================
+# PATCH 5: Agent methods
+# ============================================================
+from praisonaiagents.agent.agent import Agent as AgentClass
+
+original_chat_completion = AgentClass._chat_completion
+original_format_tools = AgentClass._format_tools_for_completion
+
+@functools.wraps(original_chat_completion)
+def traced_chat_completion(self, messages, temperature=1.0, tools=None, stream=True, 
+                           reasoning_steps=False, task_name=None, task_description=None, task_id=None):
+    log(f">>> Agent._chat_completion() stream={stream}, _using_custom_llm={self._using_custom_llm}")
+    t0 = ts()
+    result = original_chat_completion(self, messages, temperature, tools, stream, 
+                                      reasoning_steps, task_name, task_description, task_id)
+    log(f"<<< Agent._chat_completion() took {(ts()-t0)*1000:.2f}ms")
+    return result
+
+@functools.wraps(original_format_tools)
+def traced_format_tools(self, tools):
+    log(f">>> Agent._format_tools_for_completion() tools={tools}")
+    t0 = ts()
+    result = original_format_tools(self, tools)
+    log(f"<<< Agent._format_tools_for_completion() took {(ts()-t0)*1000:.2f}ms")
+    return result
+
+AgentClass._chat_completion = traced_chat_completion
+AgentClass._format_tools_for_completion = traced_format_tools
+
+# ============================================================
+# PATCH 6: OpenAIClient methods
+# ============================================================
+from praisonaiagents.llm.openai_client import OpenAIClient
+
+original_chat_with_tools = OpenAIClient.chat_completion_with_tools
+original_process_stream = OpenAIClient.process_stream_response
+original_create_completion = OpenAIClient.create_completion
+original_format_tools_client = OpenAIClient.format_tools
+
+@functools.wraps(original_chat_with_tools)
+def traced_chat_with_tools(self, *args, **kwargs):
+    log(f">>> OpenAIClient.chat_completion_with_tools() stream={kwargs.get('stream', True)}")
+    t0 = ts()
+    result = original_chat_with_tools(self, *args, **kwargs)
+    log(f"<<< OpenAIClient.chat_completion_with_tools() took {(ts()-t0)*1000:.2f}ms")
+    return result
+
+@functools.wraps(original_process_stream)
+def traced_process_stream(self, *args, **kwargs):
+    log(f">>> OpenAIClient.process_stream_response()")
+    t0 = ts()
+    result = original_process_stream(self, *args, **kwargs)
+    log(f"<<< OpenAIClient.process_stream_response() took {(ts()-t0)*1000:.2f}ms")
+    return result
+
+@functools.wraps(original_create_completion)
+def traced_create_completion(self, *args, **kwargs):
+    log(f">>> OpenAIClient.create_completion() stream={kwargs.get('stream', False)}")
+    t0 = ts()
+    result = original_create_completion(self, *args, **kwargs)
+    log(f"<<< OpenAIClient.create_completion() took {(ts()-t0)*1000:.2f}ms")
+    return result
+
+@functools.wraps(original_format_tools_client)
+def traced_format_tools_client(self, tools):
+    log(f">>> OpenAIClient.format_tools() tools count={len(tools) if tools else 0}")
+    t0 = ts()
+    result = original_format_tools_client(self, tools)
+    log(f"<<< OpenAIClient.format_tools() took {(ts()-t0)*1000:.2f}ms")
+    return result
+
+OpenAIClient.chat_completion_with_tools = traced_chat_with_tools
+OpenAIClient.process_stream_response = traced_process_stream
+OpenAIClient.create_completion = traced_create_completion
+OpenAIClient.format_tools = traced_format_tools_client
+
+# ============================================================
+# PATCH 7: sync_client property (lazy init)
+# ============================================================
+original_sync_client_getter = OpenAIClient.sync_client.fget
+
+def traced_sync_client_getter(self):
+    log(f">>> OpenAIClient.sync_client property accessed")
+    t0 = ts()
+    result = original_sync_client_getter(self)
+    log(f"<<< OpenAIClient.sync_client took {(ts()-t0)*1000:.2f}ms")
+    return result
+
+OpenAIClient.sync_client = property(traced_sync_client_getter)
+
+log("All patches applied. Creating agent...")
+
+# ============================================================
+# RUN THE TEST
+# ============================================================
+t0 = ts()
+agent = Agent(
+    instructions="You are a helpful assistant"
+)
+log(f"Agent created in {(ts()-t0)*1000:.2f}ms")
+log(f"Agent.stream={agent.stream}, Agent.verbose={agent.verbose}")
+
+log("=" * 60)
+log("Starting agent.start()...")
+log("=" * 60)
+
+t0 = ts()
+result = agent.start("Say hello in one word")
+log(f"agent.start() completed in {(ts()-t0)*1000:.2f}ms")
+
+log("=" * 60)
+log("DONE")
+log("=" * 60)

src/praisonai-agents/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "praisonaiagents"
-version = "0.11.1"
+version = "0.11.14"
 description = "Praison AI agents for completing complex tasks with Self Reflection Agents"
 readme = "README.md"
 requires-python = ">=3.10"

src/praisonai-agents/streaming-openai-basic.py

@@ -0,0 +1,24 @@
+"""
+Streaming Example with TTFT Metrics
+
+This example demonstrates true pass-through streaming with timing metrics.
+When verbose output is enabled, tokens are displayed as they arrive from the provider.
+
+Key timing metrics:
+- TTFT (Time To First Token): Time from request to first token received
+- Stream Duration: Time from first to last token
+- Total Time: End-to-end request time
+"""
+from praisonaiagents import Agent
+from praisonaiagents.config import OutputConfig
+
+# Create agent with verbose output (shows streaming progress)
+agent = Agent(
+    instructions="You are a helpful assistant",
+    llm="gpt-4o-mini",
+    output=OutputConfig(verbose=True)  # Show generation progress
+)
+
+# Run the agent - verbose mode shows streaming progress via Rich Live display
+result = agent.start("Write a short paragraph about the history of computing")
+print(f"\n--- Response ---\n{result}")

src/praisonai-agents/streaming-openai.py

@@ -1,11 +0,0 @@
-from praisonaiagents import Agent
-
-agent = Agent(
-    instructions="You are a helpful assistant",
-    llm="gpt-4o-mini"
-)
-
-# Use chat method with stream=True for streaming responses
-for chunk in agent.chat("Write a short paragraph about the history of computing", stream=True):
-    print(chunk, end="", flush=True)
-print()  # Final newline

src/praisonai-agents/test_full_agent.py

@@ -0,0 +1,22 @@
+"""Test full agent flow with new verbose UI."""
+from praisonaiagents import Agent
+
+# Create agent - use output preset for verbose mode
+agent = Agent(
+    name="AIAssistant",
+    role="Helpful AI",
+    instructions="You are a helpful assistant. Be concise and give short answers.",
+    output="verbose",  # Use output preset instead of verbose kwarg
+)
+
+if __name__ == "__main__":
+    print("\n" + "=" * 60)
+    print("TESTING FULL AGENT FLOW WITH NEW UI")
+    print("=" * 60 + "\n")
+    
+    # Use start() which should show verbose output
+    result = agent.start("What is the capital of France? One sentence only.")
+    
+    print("\n" + "=" * 60)
+    print(f"FINAL RESULT: {result}")
+    print("=" * 60)

src/praisonai-agents/test_verbose_ui.py

@@ -0,0 +1,112 @@
+"""Test script with forced verbose mode to verify PraisonAI's unique output UI."""
+import sys
+from rich.console import Console
+
+# Force Rich to think we're in a TTY for testing
+console = Console(force_terminal=True)
+
+from praisonaiagents import Agent
+from praisonaiagents.main import (
+    display_interaction, 
+    display_instruction, 
+    display_tool_call,
+    display_reasoning_steps,
+    display_working_status,
+    display_error,
+    PRAISON_COLORS
+)
+
+print("=" * 70)
+print("TESTING PRAISONAI UNIQUE VERBOSE OUTPUT UI")
+print("=" * 70)
+
+# Test 1: Display Agent Info panel
+print("\n[TEST 1] Agent Info Panel (Sage Green border)")
+display_instruction(
+    message="Test instruction",
+    console=console,
+    agent_name="ResearchAgent",
+    agent_role="Data Analyst",
+    agent_tools=["search_web", "calculate", "read_file"]
+)
+
+# Test 2: Display interaction with semantic colors
+print("\n[TEST 2] Task (Coral) + Response (Ocean Blue) panels")
+display_interaction(
+    message="What is 2+2?",
+    response="2+2 equals 4. This is a basic arithmetic operation.",
+    markdown=True,
+    generation_time=1.5,
+    console=console
+)
+
+# Test 3: Display interaction with metrics footer
+print("\n[TEST 3] Response with Metrics Footer")
+display_interaction(
+    message="Explain quantum computing",
+    response="Quantum computing uses qubits instead of classical bits.",
+    markdown=True,
+    generation_time=3.2,
+    console=console,
+    metrics={
+        'tokens_in': 45,
+        'tokens_out': 128,
+        'cost': 0.0023,
+        'model': 'gpt-4o-mini'
+    }
+)
+
+# Test 4: Display tool call with timeline format
+print("\n[TEST 4] Tool Activity Panel (Violet border, timeline format)")
+display_tool_call(
+    message="search_web(query='quantum computing')",
+    console=console,
+    tool_name="search_web",
+    tool_input={"query": "quantum computing"},
+    tool_output="Found 5 relevant articles",
+    elapsed_time=0.8,
+    success=True
+)
+
+# Test 5: Display tool call failure
+print("\n[TEST 5] Tool Activity Panel (Failed call)")
+display_tool_call(
+    message="api_call failed",
+    console=console,
+    tool_name="api_call",
+    tool_input={"endpoint": "/users", "method": "GET"},
+    tool_output="Connection timeout",
+    elapsed_time=5.0,
+    success=False
+)
+
+# Test 6: Display reasoning steps with numbered circles
+print("\n[TEST 6] Reasoning Steps (Numbered circles ①②③)")
+display_reasoning_steps(
+    steps=[
+        "Identified query about arithmetic",
+        "Retrieved basic math rules",
+        "Calculated result: 2+2=4",
+        "Formulated concise response"
+    ],
+    console=console
+)
+
+# Test 7: Display working status (static, shows format)
+print("\n[TEST 7] Working Status (Amber border, pulsing dots)")
+for phase in range(4):
+    panel = display_working_status(phase=phase, console=console)
+    console.print(panel)
+
+# Test 8: Display error
+print("\n[TEST 8] Error Panel (Red border)")
+display_error(
+    message="API rate limit exceeded. Please try again in 60 seconds.",
+    console=console
+)
+
+print("\n" + "=" * 70)
+print("COLORS BEING USED:")
+for name, color in PRAISON_COLORS.items():
+    console.print(f"  {name}: [bold {color}]████████[/] {color}")
+print("=" * 70)

src/praisonai-agents/tests/benchmarks/output_mode_benchmark.py

@@ -0,0 +1,207 @@
+#!/usr/bin/env python3
+"""
+Micro-benchmark for output mode overhead comparison.
+
+Measures:
+1. Import-time cost (cold import)
+2. Agent initialization overhead per output mode
+3. Hot-path overhead (simulated callback execution)
+
+Usage:
+    python tests/benchmarks/output_mode_benchmark.py
+
+All stdlib - no external dependencies required.
+"""
+
+import subprocess
+import sys
+import time
+from statistics import mean, stdev
+
+
+def measure_import_time(mode: str, iterations: int = 5) -> dict:
+    """Measure cold import time for a specific output mode."""
+    times = []
+    
+    # Script that imports and creates an agent with specific output mode
+    script = f'''
+import time
+start = time.perf_counter()
+import os
+os.environ['PRAISONAI_OUTPUT'] = '{mode}'
+from praisonaiagents import Agent
+agent = Agent(name="BenchAgent", instructions="Test", output="{mode}")
+end = time.perf_counter()
+print(f"{{(end - start) * 1000:.3f}}")
+'''
+    
+    for _ in range(iterations):
+        result = subprocess.run(
+            [sys.executable, '-c', script],
+            capture_output=True,
+            text=True,
+            env={**dict(__import__('os').environ), 'PYTHONDONTWRITEBYTECODE': '1'}
+        )
+        if result.returncode == 0:
+            try:
+                times.append(float(result.stdout.strip()))
+            except ValueError:
+                pass
+    
+    if not times:
+        return {"mode": mode, "mean_ms": -1, "stdev_ms": -1, "samples": 0}
+    
+    return {
+        "mode": mode,
+        "mean_ms": mean(times),
+        "stdev_ms": stdev(times) if len(times) > 1 else 0,
+        "samples": len(times),
+    }
+
+
+def measure_callback_overhead(iterations: int = 1000) -> dict:
+    """Measure hot-path overhead of callback execution vs no callbacks."""
+    results = {}
+    
+    # Silent mode - no callbacks
+    silent_times = []
+    for _ in range(iterations):
+        start = time.perf_counter()
+        # Simulate what happens in silent mode: nothing
+        pass
+        end = time.perf_counter()
+        silent_times.append((end - start) * 1_000_000)  # microseconds
+    
+    results["silent"] = {
+        "mean_us": mean(silent_times),
+        "stdev_us": stdev(silent_times) if len(silent_times) > 1 else 0,
+    }
+    
+    # Actions mode - simulate callback overhead
+    actions_times = []
+    
+    # Simulate the work done by actions mode per tool call
+    def simulate_actions_callback():
+        import time as t
+        # Simulate: time.time(), dict creation, string formatting, stderr write
+        _ = t.time()
+        event = {
+            "type": "tool_call",
+            "tool": "test_tool",
+            "args": {"arg1": "value1", "arg2": "value2"},
+            "timestamp": t.time(),
+        }
+        formatted = f"[tool_call] {event['tool']}({event['args']})"
+        # Don't actually write to avoid polluting output
+        _ = formatted.encode()
+    
+    for _ in range(iterations):
+        start = time.perf_counter()
+        simulate_actions_callback()
+        end = time.perf_counter()
+        actions_times.append((end - start) * 1_000_000)  # microseconds
+    
+    results["actions"] = {
+        "mean_us": mean(actions_times),
+        "stdev_us": stdev(actions_times) if len(actions_times) > 1 else 0,
+    }
+    
+    return results
+
+
+def measure_agent_init_overhead() -> dict:
+    """Measure Agent.__init__ overhead for different output modes."""
+    import os
+    
+    # Clear any env var
+    os.environ.pop('PRAISONAI_OUTPUT', None)
+    
+    results = {}
+    iterations = 10
+    
+    for mode in ["silent", "actions", "verbose"]:
+        times = []
+        for _ in range(iterations):
+            # Import fresh each time to avoid caching effects
+            from praisonaiagents import Agent
+            
+            start = time.perf_counter()
+            agent = Agent(
+                name="BenchAgent",
+                instructions="Test agent",
+                output=mode,
+            )
+            end = time.perf_counter()
+            times.append((end - start) * 1000)  # milliseconds
+            del agent
+        
+        results[mode] = {
+            "mean_ms": mean(times),
+            "stdev_ms": stdev(times) if len(times) > 1 else 0,
+        }
+    
+    return results
+
+
+def main():
+    print("=" * 60)
+    print("OUTPUT MODE MICRO-BENCHMARK")
+    print("=" * 60)
+    print()
+    
+    # 1. Import time comparison
+    print("1. IMPORT + INIT TIME (cold start, subprocess)")
+    print("-" * 40)
+    
+    modes = ["silent", "actions", "verbose"]
+    import_results = []
+    
+    for mode in modes:
+        print(f"   Measuring {mode}...", end=" ", flush=True)
+        result = measure_import_time(mode, iterations=3)
+        import_results.append(result)
+        print(f"{result['mean_ms']:.1f}ms (±{result['stdev_ms']:.1f})")
+    
+    print()
+    
+    # 2. Agent init overhead (warm)
+    print("2. AGENT INIT OVERHEAD (warm, same process)")
+    print("-" * 40)
+    
+    init_results = measure_agent_init_overhead()
+    for mode, data in init_results.items():
+        print(f"   {mode:12s}: {data['mean_ms']:.2f}ms (±{data['stdev_ms']:.2f})")
+    
+    print()
+    
+    # 3. Hot-path callback overhead
+    print("3. HOT-PATH CALLBACK OVERHEAD (per call)")
+    print("-" * 40)
+    
+    callback_results = measure_callback_overhead(iterations=10000)
+    for mode, data in callback_results.items():
+        print(f"   {mode:12s}: {data['mean_us']:.3f}µs (±{data['stdev_us']:.3f})")
+    
+    print()
+    
+    # Summary
+    print("=" * 60)
+    print("SUMMARY")
+    print("=" * 60)
+    
+    silent_init = init_results.get("silent", {}).get("mean_ms", 0)
+    actions_init = init_results.get("actions", {}).get("mean_ms", 0)
+    
+    silent_callback = callback_results.get("silent", {}).get("mean_us", 0)
+    actions_callback = callback_results.get("actions", {}).get("mean_us", 0)
+    
+    print(f"   Init overhead (actions vs silent): +{actions_init - silent_init:.2f}ms")
+    print(f"   Per-call overhead (actions vs silent): +{actions_callback - silent_callback:.3f}µs")
+    print()
+    print("   DEFAULT MODE: silent (zero overhead)")
+    print("   OPT-IN: output='actions' for observability")
+    print()
+
+
+if __name__ == "__main__":
+    main()

src/praisonai-agents/tests/unit/agent/test_execution_api.py

@@ -0,0 +1,257 @@
+"""
+Tests for Agent execution API semantics: start(), run(), stream()
+
+TDD tests for the new execution API:
+- start() = interactive, streams by default in TTY
+- run() = production, silent by default
+- stream() = app-friendly iterator
+"""
+
+import pytest
+import sys
+from unittest.mock import patch
+
+
+class TestAgentStartMethod:
+    """Tests for Agent.start() - interactive, streaming by default in TTY"""
+    
+    def test_start_streams_by_default_in_tty(self):
+        """start() should stream by default when stdout is a TTY"""
+        from praisonaiagents import Agent
+        
+        # Create agent without stream attribute set (None)
+        agent = Agent(name="Test", instructions="You are helpful")
+        # Ensure stream attribute is None so TTY detection kicks in
+        agent.stream = None
+        
+        # Mock TTY detection
+        with patch.object(sys.stdout, 'isatty', return_value=True):
+            # Mock chat to avoid actual LLM call
+            with patch.object(agent, '_start_stream') as mock_stream:
+                mock_stream.return_value = iter(["Hello", " World"])
+                with patch.object(agent, 'chat') as mock_chat:
+                    agent.start("Hello")
+                    # Should call _start_stream when TTY
+                    mock_stream.assert_called_once()
+                    mock_chat.assert_not_called()
+    
+    def test_start_does_not_stream_when_not_tty(self):
+        """start() should not stream when stdout is not a TTY"""
+        from praisonaiagents import Agent
+        
+        agent = Agent(name="Test", instructions="You are helpful")
+        
+        # Mock non-TTY
+        with patch.object(sys.stdout, 'isatty', return_value=False):
+            with patch.object(agent, '_start_stream') as mock_stream:
+                with patch.object(agent, 'chat', return_value="Hello World") as mock_chat:
+                    result = agent.start("Hello")
+                    # Should call chat, not _start_stream
+                    mock_chat.assert_called_once()
+                    mock_stream.assert_not_called()
+    
+    def test_start_respects_explicit_stream_true(self):
+        """start(stream=True) should always stream"""
+        from praisonaiagents import Agent
+        
+        agent = Agent(name="Test", instructions="You are helpful")
+        
+        # Even when not TTY, explicit stream=True should stream
+        with patch.object(sys.stdout, 'isatty', return_value=False):
+            with patch.object(agent, '_start_stream') as mock_stream:
+                mock_stream.return_value = iter(["Hello"])
+                result = agent.start("Hello", stream=True)
+                mock_stream.assert_called_once()
+    
+    def test_start_respects_explicit_stream_false(self):
+        """start(stream=False) should never stream"""
+        from praisonaiagents import Agent
+        
+        agent = Agent(name="Test", instructions="You are helpful")
+        
+        # Even when TTY, explicit stream=False should not stream
+        with patch.object(sys.stdout, 'isatty', return_value=True):
+            with patch.object(agent, 'chat', return_value="Hello") as mock_chat:
+                with patch.object(agent, '_start_stream') as mock_stream:
+                    result = agent.start("Hello", stream=False)
+                    mock_chat.assert_called_once()
+                    mock_stream.assert_not_called()
+    
+    def test_start_respects_agent_stream_attribute(self):
+        """start() should respect agent's stream attribute over TTY detection"""
+        from praisonaiagents import Agent
+        
+        # Agent with stream=True should always stream
+        agent = Agent(name="Test", instructions="You are helpful", output="stream")
+        
+        with patch.object(sys.stdout, 'isatty', return_value=False):
+            with patch.object(agent, '_start_stream') as mock_stream:
+                mock_stream.return_value = iter(["Hello"])
+                # Agent's stream attribute should take precedence
+                if agent.stream:
+                    result = agent.start("Hello")
+                    mock_stream.assert_called()
+
+
+class TestAgentRunMethod:
+    """Tests for Agent.run() - production, silent by default"""
+    
+    def test_run_does_not_stream_by_default(self):
+        """run() should not stream by default"""
+        from praisonaiagents import Agent
+        
+        agent = Agent(name="Test", instructions="You are helpful")
+        
+        with patch.object(agent, 'chat', return_value="Hello World") as mock_chat:
+            with patch.object(agent, '_start_stream') as mock_stream:
+                result = agent.run("Hello")
+                # Should call chat with stream=False
+                mock_chat.assert_called_once()
+                call_kwargs = mock_chat.call_args[1]
+                assert call_kwargs.get('stream') == False
+                mock_stream.assert_not_called()
+    
+    def test_run_returns_result_directly(self):
+        """run() should return the result directly, not a generator"""
+        from praisonaiagents import Agent
+        
+        agent = Agent(name="Test", instructions="You are helpful")
+        
+        with patch.object(agent, 'chat', return_value="Hello World") as mock_chat:
+            result = agent.run("Hello")
+            assert result == "Hello World"
+            assert not hasattr(result, '__iter__') or isinstance(result, str)
+    
+    def test_run_ignores_tty_detection(self):
+        """run() should ignore TTY detection and never stream by default"""
+        from praisonaiagents import Agent
+        
+        agent = Agent(name="Test", instructions="You are helpful")
+        
+        # Even in TTY, run() should not stream
+        with patch.object(sys.stdout, 'isatty', return_value=True):
+            with patch.object(agent, 'chat', return_value="Hello") as mock_chat:
+                with patch.object(agent, '_start_stream') as mock_stream:
+                    result = agent.run("Hello")
+                    mock_chat.assert_called_once()
+                    mock_stream.assert_not_called()
+    
+    def test_run_can_be_forced_to_stream(self):
+        """run(stream=True) should stream when explicitly requested"""
+        from praisonaiagents import Agent
+        
+        agent = Agent(name="Test", instructions="You are helpful")
+        
+        with patch.object(agent, 'chat', return_value="Hello") as mock_chat:
+            result = agent.run("Hello", stream=True)
+            # When stream=True is passed, it should be passed to chat
+            call_kwargs = mock_chat.call_args[1]
+            assert call_kwargs.get('stream') == True
+
+
+class TestAgentIterStreamMethod:
+    """Tests for Agent.iter_stream() - app-friendly iterator"""
+    
+    def test_iter_stream_returns_iterator(self):
+        """iter_stream() should return an iterator"""
+        from praisonaiagents import Agent
+        
+        agent = Agent(name="Test", instructions="You are helpful")
+        
+        with patch.object(agent, '_start_stream') as mock_stream:
+            mock_stream.return_value = iter(["Hello", " ", "World"])
+            result = agent.iter_stream("Hello")
+            # Should be an iterator/generator
+            assert hasattr(result, '__iter__')
+            assert hasattr(result, '__next__')
+    
+    def test_iter_stream_yields_chunks(self):
+        """iter_stream() should yield response chunks"""
+        from praisonaiagents import Agent
+        
+        agent = Agent(name="Test", instructions="You are helpful")
+        
+        with patch.object(agent, '_start_stream') as mock_stream:
+            mock_stream.return_value = iter(["Hello", " ", "World"])
+            chunks = list(agent.iter_stream("Hello"))
+            assert chunks == ["Hello", " ", "World"]
+    
+    def test_iter_stream_always_streams(self):
+        """iter_stream() should always stream regardless of TTY"""
+        from praisonaiagents import Agent
+        
+        agent = Agent(name="Test", instructions="You are helpful")
+        
+        # Even when not TTY
+        with patch.object(sys.stdout, 'isatty', return_value=False):
+            with patch.object(agent, '_start_stream') as mock_stream:
+                mock_stream.return_value = iter(["Hello"])
+                result = agent.iter_stream("Hello")
+                # Consume the generator
+                list(result)
+                mock_stream.assert_called_once()
+
+
+class TestAsyncMethods:
+    """Tests for async versions: arun(), astart()"""
+    
+    @pytest.mark.asyncio
+    async def test_arun_does_not_stream_by_default(self):
+        """arun() should not stream by default"""
+        from praisonaiagents import Agent
+        
+        agent = Agent(name="Test", instructions="You are helpful")
+        
+        with patch.object(agent, 'achat', return_value="Hello World") as mock_achat:
+            result = await agent.arun("Hello")
+            mock_achat.assert_called_once()
+            call_kwargs = mock_achat.call_args[1]
+            assert call_kwargs.get('stream') == False
+    
+    @pytest.mark.asyncio
+    async def test_astart_streams_in_tty(self):
+        """astart() should stream when in TTY"""
+        from praisonaiagents import Agent
+        
+        agent = Agent(name="Test", instructions="You are helpful")
+        # Ensure stream attribute is None so TTY detection kicks in
+        agent.stream = None
+        
+        with patch.object(sys.stdout, 'isatty', return_value=True):
+            with patch.object(agent, 'achat', return_value="Hello") as mock_achat:
+                await agent.astart("Hello")
+                call_kwargs = mock_achat.call_args[1]
+                assert call_kwargs.get('stream')
+
+
+class TestMethodSemanticDistinction:
+    """Tests to verify start() and run() have distinct semantics"""
+    
+    def test_start_and_run_are_different(self):
+        """start() and run() should have different default behaviors"""
+        from praisonaiagents import Agent
+        
+        agent = Agent(name="Test", instructions="You are helpful")
+        # Ensure stream attribute is None so TTY detection kicks in
+        agent.stream = None
+        
+        # In TTY context
+        with patch.object(sys.stdout, 'isatty', return_value=True):
+            with patch.object(agent, 'chat', return_value="Hello") as mock_chat:
+                with patch.object(agent, '_start_stream') as mock_stream:
+                    mock_stream.return_value = iter(["Hello"])
+                    
+                    # start() should stream
+                    agent.start("Hello")
+                    start_streamed = mock_stream.called
+                    
+                    mock_stream.reset_mock()
+                    mock_chat.reset_mock()
+                    
+                    # run() should not stream
+                    agent.run("Hello")
+                    run_streamed = mock_stream.called
+                    
+                    # They should behave differently
+                    assert start_streamed
+                    assert not run_streamed

src/praisonai-agents/tests/unit/agent/test_output_verbose.py

@@ -0,0 +1,156 @@
+"""
+Unit tests for Agent output parameter with verbose preset.
+
+Tests the fix for the bug where output="verbose" would return None
+because reasoning_steps=True in the verbose preset was causing the code
+to try to return reasoning_content, but this attribute exists as None
+for non-reasoning models.
+"""
+import pytest
+from unittest.mock import Mock
+
+
+class TestReasoningContentLogic:
+    """Test the reasoning_content check logic directly."""
+
+    def test_condition_with_none_reasoning_content(self):
+        """Verify the fix: condition should be False when reasoning_content is None."""
+        mock_message = Mock()
+        mock_message.reasoning_content = None
+        
+        reasoning_steps = True
+        
+        # Old buggy condition:
+        # if reasoning_steps and hasattr(mock_message, 'reasoning_content'):
+        old_condition = reasoning_steps and hasattr(mock_message, 'reasoning_content')
+        assert old_condition == True, "Old condition would be True (the bug)"
+        
+        # Fixed condition:
+        # if reasoning_steps and hasattr(mock_message, 'reasoning_content') and mock_message.reasoning_content:
+        # The truthiness of the whole expression should be False
+        fixed_condition = reasoning_steps and hasattr(mock_message, 'reasoning_content') and mock_message.reasoning_content
+        assert not fixed_condition, "Fixed condition should be falsy when reasoning_content is None"
+
+    def test_condition_with_empty_string_reasoning_content(self):
+        """Verify the fix: condition should be False when reasoning_content is empty string."""
+        mock_message = Mock()
+        mock_message.reasoning_content = ""
+        
+        reasoning_steps = True
+        
+        # Old condition would True (bug)
+        old_condition = reasoning_steps and hasattr(mock_message, 'reasoning_content')
+        assert old_condition == True
+        
+        # Fixed condition should be falsy
+        fixed_condition = reasoning_steps and hasattr(mock_message, 'reasoning_content') and mock_message.reasoning_content
+        assert not fixed_condition, "Fixed condition should be falsy when reasoning_content is empty string"
+
+    def test_condition_with_actual_reasoning_content(self):
+        """Verify the fix: condition should be True when reasoning_content has content."""
+        mock_message = Mock()
+        mock_message.reasoning_content = "This is my reasoning process..."
+        
+        reasoning_steps = True
+        
+        # Both old and new condition should be truthy
+        old_condition = reasoning_steps and hasattr(mock_message, 'reasoning_content')
+        assert old_condition
+        
+        fixed_condition = reasoning_steps and hasattr(mock_message, 'reasoning_content') and mock_message.reasoning_content
+        assert fixed_condition, "Fixed condition should be truthy when reasoning_content has content"
+
+    def test_condition_when_reasoning_steps_false(self):
+        """Verify that when reasoning_steps is False, condition is always False."""
+        mock_message = Mock()
+        mock_message.reasoning_content = "This is my reasoning process..."
+        
+        reasoning_steps = False
+        
+        fixed_condition = reasoning_steps and hasattr(mock_message, 'reasoning_content') and mock_message.reasoning_content
+        assert not fixed_condition, "Condition should be falsy when reasoning_steps is False"
+
+    def test_if_statement_behavior_with_none(self):
+        """Verify that an if statement using the condition skips when reasoning_content is None."""
+        mock_message = Mock()
+        mock_message.reasoning_content = None
+        
+        reasoning_steps = True
+        execution_path = "not_taken"
+        
+        # This is how the fixed code behaves
+        if reasoning_steps and hasattr(mock_message, 'reasoning_content') and mock_message.reasoning_content:
+            execution_path = "reasoning_path"
+        else:
+            execution_path = "regular_path"
+        
+        assert execution_path == "regular_path", "Should take regular path when reasoning_content is None"
+
+    def test_if_statement_behavior_with_reasoning(self):
+        """Verify that an if statement using the condition enters when reasoning_content has value."""
+        mock_message = Mock()
+        mock_message.reasoning_content = "Reasoning steps here..."
+        
+        reasoning_steps = True
+        execution_path = "not_taken"
+        
+        if reasoning_steps and hasattr(mock_message, 'reasoning_content') and mock_message.reasoning_content:
+            execution_path = "reasoning_path"
+        else:
+            execution_path = "regular_path"
+        
+        assert execution_path == "reasoning_path", "Should take reasoning path when reasoning_content has value"
+
+
+class TestVerbosePresetConfig:
+    """Test that the verbose preset has the expected configuration."""
+
+    def test_verbose_preset_has_reasoning_steps_true(self):
+        """Verify that the verbose preset enables reasoning_steps."""
+        from praisonaiagents.config.presets import OUTPUT_PRESETS
+        
+        verbose_preset = OUTPUT_PRESETS.get("verbose")
+        assert verbose_preset is not None, "verbose preset should exist"
+        assert verbose_preset.get("reasoning_steps") == True, "verbose preset should have reasoning_steps=True"
+
+    def test_verbose_preset_has_verbose_true(self):
+        """Verify that the verbose preset enables verbose output."""
+        from praisonaiagents.config.presets import OUTPUT_PRESETS
+        
+        verbose_preset = OUTPUT_PRESETS.get("verbose")
+        assert verbose_preset.get("verbose") == True, "verbose preset should have verbose=True"
+
+    def test_verbose_preset_has_stream_false(self):
+        """Verify that the verbose preset does not enable streaming by default."""
+        from praisonaiagents.config.presets import OUTPUT_PRESETS
+        
+        verbose_preset = OUTPUT_PRESETS.get("verbose")
+        assert verbose_preset.get("stream") == False, "verbose preset should have stream=False"
+
+
+class TestChatCompletionMessageDataclass:
+    """Test the ChatCompletionMessage dataclass has reasoning_content attribute."""
+
+    def test_chat_completion_message_has_reasoning_content(self):
+        """Verify ChatCompletionMessage has reasoning_content attribute that defaults to None."""
+        from praisonaiagents.llm.openai_client import ChatCompletionMessage
+        
+        # Create without reasoning_content
+        message = ChatCompletionMessage(content="Test", role="assistant")
+        
+        # Should have the attribute
+        assert hasattr(message, 'reasoning_content')
+        # Should default to None
+        assert message.reasoning_content is None
+
+    def test_chat_completion_message_with_reasoning(self):
+        """Verify ChatCompletionMessage can have reasoning_content set."""
+        from praisonaiagents.llm.openai_client import ChatCompletionMessage
+        
+        message = ChatCompletionMessage(
+            content="Test",
+            role="assistant",
+            reasoning_content="This is my reasoning..."
+        )
+        
+        assert message.reasoning_content == "This is my reasoning..."

src/praisonai-agents/tests/unit/agent/test_start_verbose_tty.py

@@ -0,0 +1,182 @@
+"""
+Unit tests for Agent.start() method defaulting to verbose output in TTY.
+
+Tests the behavior that:
+- .start() defaults to verbose output when running in a TTY
+- .start() uses agent's default (silent) when NOT in TTY
+- .run() always uses the agent's default (silent)
+- Explicit output= kwarg overrides TTY detection
+
+This implements Option B: Make .start() default to verbose output in TTY.
+"""
+import pytest
+from unittest.mock import Mock, patch
+import sys
+
+
+class TestStartVerboseInTTY:
+    """Test that .start() defaults to verbose output in TTY."""
+
+    def test_start_sets_verbose_in_tty(self):
+        """Verify .start() sets verbose=True when running in TTY."""
+        # Mock isatty to return True
+        with patch.object(sys.stdout, 'isatty', return_value=True):
+            # Create agent with default (silent) output
+            from praisonaiagents import Agent
+            agent = Agent(instructions="Test agent")
+            
+            # Verify default is silent
+            assert agent.verbose == False, "Default should be silent (verbose=False)"
+            
+            # Track what happens in start
+            original_chat = agent.chat
+            captured_verbose = []
+            
+            def mock_chat(*args, **kwargs):
+                captured_verbose.append(agent.verbose)
+                return "mock response"
+            
+            agent.chat = mock_chat
+            agent._load_history_context = Mock()
+            agent._auto_save_session = Mock()
+            
+            # Call start
+            result = agent.start("Hello")
+            
+            # Verbose should have been True during the call
+            assert len(captured_verbose) == 1, "chat() should have been called"
+            assert captured_verbose[0] == True, ".start() should set verbose=True in TTY"
+            
+            # Verbose should be restored after
+            assert agent.verbose == False, "verbose should be restored after .start()"
+
+    def test_start_respects_explicit_output_kwarg(self):
+        """Verify .start() respects explicit output='silent' even in TTY."""
+        with patch.object(sys.stdout, 'isatty', return_value=True):
+            from praisonaiagents import Agent
+            agent = Agent(instructions="Test agent")
+            
+            captured_verbose = []
+            
+            def mock_chat(*args, **kwargs):
+                captured_verbose.append(agent.verbose)
+                return "mock response"
+            
+            agent.chat = mock_chat
+            agent._load_history_context = Mock()
+            agent._auto_save_session = Mock()
+            
+            # Call start with explicit output='silent'
+            result = agent.start("Hello", output="silent")
+            
+            # Should respect explicit silent
+            assert captured_verbose[0] == False, "explicit output='silent' should be respected"
+
+    def test_start_silent_when_not_tty(self):
+        """Verify .start() stays silent when not in TTY."""
+        with patch.object(sys.stdout, 'isatty', return_value=False):
+            from praisonaiagents import Agent
+            agent = Agent(instructions="Test agent")
+            
+            captured_verbose = []
+            
+            def mock_chat(*args, **kwargs):
+                captured_verbose.append(agent.verbose)
+                return "mock response"
+            
+            agent.chat = mock_chat
+            agent._load_history_context = Mock()
+            agent._auto_save_session = Mock()
+            
+            # Call start in non-TTY
+            result = agent.start("Hello")
+            
+            # Should remain silent
+            assert captured_verbose[0] == False, ".start() should stay silent when not TTY"
+
+    def test_run_always_silent(self):
+        """Verify .run() is always silent regardless of TTY."""
+        with patch.object(sys.stdout, 'isatty', return_value=True):
+            from praisonaiagents import Agent
+            agent = Agent(instructions="Test agent")
+            
+            captured_verbose = []
+            
+            def mock_chat(*args, **kwargs):
+                captured_verbose.append(agent.verbose)
+                return "mock response"
+            
+            agent.chat = mock_chat
+            agent._load_history_context = Mock()
+            agent._auto_save_session = Mock()
+            
+            # Call run (should stay silent even in TTY)
+            result = agent.run("Hello")
+            
+            # Should be silent
+            assert captured_verbose[0] == False, ".run() should always be silent"
+
+
+class TestStartVerbosePresets:
+    """Test the verbose preset values applied by .start()."""
+
+    def test_start_enables_markdown_in_tty(self):
+        """Verify .start() also enables markdown=True in TTY."""
+        with patch.object(sys.stdout, 'isatty', return_value=True):
+            from praisonaiagents import Agent
+            agent = Agent(instructions="Test agent")
+            
+            # Verify default is no markdown
+            assert agent.markdown == False, "Default should be no markdown"
+            
+            captured_markdown = []
+            
+            def mock_chat(*args, **kwargs):
+                captured_markdown.append(agent.markdown)
+                return "mock response"
+            
+            agent.chat = mock_chat
+            agent._load_history_context = Mock()
+            agent._auto_save_session = Mock()
+            
+            # Call start
+            result = agent.start("Hello")
+            
+            # Markdown should have been True during the call
+            assert captured_markdown[0] == True, ".start() should set markdown=True in TTY"
+            
+            # Should be restored after
+            assert agent.markdown == False, "markdown should be restored after .start()"
+
+
+class TestAgentWithExplicitVerbose:
+    """Test that agents created with output='verbose' work correctly."""
+
+    def test_agent_with_verbose_output(self):
+        """Agent created with output='verbose' should have verbose=True."""
+        from praisonaiagents import Agent
+        agent = Agent(instructions="Test", output="verbose")
+        
+        assert agent.verbose == True, "output='verbose' should set verbose=True"
+        assert agent.markdown == True, "output='verbose' should set markdown=True"
+
+    def test_start_with_already_verbose_agent(self):
+        """Start on already-verbose agent should stay verbose."""
+        with patch.object(sys.stdout, 'isatty', return_value=True):
+            from praisonaiagents import Agent
+            agent = Agent(instructions="Test", output="verbose")
+            
+            captured_verbose = []
+            
+            def mock_chat(*args, **kwargs):
+                captured_verbose.append(agent.verbose)
+                return "mock response"
+            
+            agent.chat = mock_chat
+            agent._load_history_context = Mock()
+            agent._auto_save_session = Mock()
+            
+            result = agent.start("Hello")
+            
+            assert captured_verbose[0] == True
+            assert agent.verbose == True, "Already-verbose agent should stay verbose"

src/praisonai-agents/tests/unit/config/test_learn_scope.py

@@ -0,0 +1,104 @@
+"""
+Tests for LearnScope enum - verifying PRIVATE/SHARED values only.
+
+These tests verify that:
+1. LearnScope only has PRIVATE and SHARED values
+2. TEAM and ORG values do NOT exist
+3. LearnConfig defaults to PRIVATE scope
+4. BaseStore defaults to "private" scope
+"""
+
+import pytest
+
+
+class TestLearnScopeValues:
+    """Test LearnScope enum has exactly the right values."""
+    
+    def test_learn_scope_has_private(self):
+        """LearnScope should have PRIVATE value."""
+        from praisonaiagents.config.feature_configs import LearnScope
+        
+        assert hasattr(LearnScope, 'PRIVATE')
+        assert LearnScope.PRIVATE.value == "private"
+    
+    def test_learn_scope_has_shared(self):
+        """LearnScope should have SHARED value."""
+        from praisonaiagents.config.feature_configs import LearnScope
+        
+        assert hasattr(LearnScope, 'SHARED')
+        assert LearnScope.SHARED.value == "shared"
+    
+    def test_learn_scope_does_not_have_team(self):
+        """LearnScope should NOT have TEAM value."""
+        from praisonaiagents.config.feature_configs import LearnScope
+        
+        assert not hasattr(LearnScope, 'TEAM'), "TEAM should be removed from LearnScope"
+    
+    def test_learn_scope_does_not_have_org(self):
+        """LearnScope should NOT have ORG value."""
+        from praisonaiagents.config.feature_configs import LearnScope
+        
+        assert not hasattr(LearnScope, 'ORG'), "ORG should be removed from LearnScope"
+    
+    def test_learn_scope_does_not_have_user(self):
+        """LearnScope should NOT have USER value (replaced by PRIVATE)."""
+        from praisonaiagents.config.feature_configs import LearnScope
+        
+        assert not hasattr(LearnScope, 'USER'), "USER should be replaced by PRIVATE"
+    
+    def test_learn_scope_does_not_have_global(self):
+        """LearnScope should NOT have GLOBAL value (replaced by SHARED)."""
+        from praisonaiagents.config.feature_configs import LearnScope
+        
+        assert not hasattr(LearnScope, 'GLOBAL'), "GLOBAL should be replaced by SHARED"
+    
+    def test_learn_scope_has_exactly_two_values(self):
+        """LearnScope should have exactly 2 values: PRIVATE and SHARED."""
+        from praisonaiagents.config.feature_configs import LearnScope
+        
+        all_values = list(LearnScope)
+        assert len(all_values) == 2, f"Expected 2 values, got {len(all_values)}: {all_values}"
+
+
+class TestLearnConfigDefaults:
+    """Test LearnConfig uses correct default scope."""
+    
+    def test_learn_config_default_scope_is_private(self):
+        """LearnConfig should default to PRIVATE scope."""
+        from praisonaiagents.config.feature_configs import LearnConfig, LearnScope
+        
+        config = LearnConfig()
+        assert config.scope == LearnScope.PRIVATE
+    
+    def test_learn_config_scope_can_be_set_to_shared(self):
+        """LearnConfig scope can be set to SHARED."""
+        from praisonaiagents.config.feature_configs import LearnConfig, LearnScope
+        
+        config = LearnConfig(scope=LearnScope.SHARED)
+        assert config.scope == LearnScope.SHARED
+    
+    def test_learn_config_to_dict_includes_scope(self):
+        """LearnConfig.to_dict() should include scope with correct value."""
+        from praisonaiagents.config.feature_configs import LearnConfig, LearnScope
+        
+        config = LearnConfig()
+        d = config.to_dict()
+        assert d['scope'] == 'private'
+        
+        config2 = LearnConfig(scope=LearnScope.SHARED)
+        d2 = config2.to_dict()
+        assert d2['scope'] == 'shared'
+
+
+class TestBaseStoreDefaults:
+    """Test BaseStore uses correct default scope."""
+    
+    def test_base_store_default_scope_is_private(self):
+        """BaseStore should default to 'private' scope."""
+        import inspect
+        from praisonaiagents.memory.learn.stores import BaseStore
+        
+        sig = inspect.signature(BaseStore.__init__)
+        scope_param = sig.parameters.get('scope')
+        assert scope_param is not None
+        assert scope_param.default == 'private', f"Expected 'private', got '{scope_param.default}'"

src/praisonai-agents/tests/unit/config/test_precedence_ladder.py

@@ -0,0 +1,237 @@
+"""Tests for precedence ladder resolvers."""
+
+import pytest
+from praisonaiagents.config.feature_configs import (
+    MemoryConfig,
+    MemoryBackend,
+    KnowledgeConfig,
+    PlanningConfig,
+    ReflectionConfig,
+    GuardrailConfig,
+    WebConfig,
+    OutputConfig,
+    ExecutionConfig,
+    CachingConfig,
+    AutonomyConfig,
+    resolve_memory,
+    resolve_knowledge,
+    resolve_planning,
+    resolve_reflection,
+    resolve_guardrails,
+    resolve_web,
+    resolve_output,
+    resolve_execution,
+    resolve_caching,
+    resolve_autonomy,
+)
+
+
+class TestResolveMemory:
+    """Test resolve_memory precedence ladder."""
+    
+    def test_default_none(self):
+        """Default (None) returns None."""
+        assert resolve_memory(None) is None
+    
+    def test_bool_false(self):
+        """False returns None (disabled)."""
+        assert resolve_memory(False) is None
+    
+    def test_bool_true(self):
+        """True returns MemoryConfig with defaults."""
+        result = resolve_memory(True)
+        assert isinstance(result, MemoryConfig)
+        assert result.backend == MemoryBackend.FILE
+    
+    def test_string_backend(self):
+        """String returns MemoryConfig with that backend."""
+        result = resolve_memory("redis")
+        assert isinstance(result, MemoryConfig)
+        assert result.backend == MemoryBackend.REDIS
+    
+    def test_string_custom_backend(self):
+        """Custom string backend is preserved."""
+        result = resolve_memory("custom_backend")
+        assert isinstance(result, MemoryConfig)
+        assert result.backend == "custom_backend"
+    
+    def test_dict_expansion(self):
+        """Dict expands to MemoryConfig."""
+        result = resolve_memory({
+            "backend": "sqlite",
+            "user_id": "alice",
+            "auto_memory": True,
+        })
+        assert isinstance(result, MemoryConfig)
+        assert result.backend == MemoryBackend.SQLITE
+        assert result.user_id == "alice"
+        assert result.auto_memory is True
+    
+    def test_config_passthrough(self):
+        """MemoryConfig passes through unchanged."""
+        config = MemoryConfig(backend=MemoryBackend.POSTGRES, user_id="bob")
+        result = resolve_memory(config)
+        assert result is config
+    
+    def test_instance_passthrough(self):
+        """Instance (non-config object) passes through."""
+        class MockMemoryManager:
+            pass
+        instance = MockMemoryManager()
+        result = resolve_memory(instance)
+        assert result is instance
+
+
+class TestResolveKnowledge:
+    """Test resolve_knowledge precedence ladder."""
+    
+    def test_default_none(self):
+        assert resolve_knowledge(None) is None
+    
+    def test_bool_false(self):
+        assert resolve_knowledge(False) is None
+    
+    def test_bool_true(self):
+        result = resolve_knowledge(True)
+        assert isinstance(result, KnowledgeConfig)
+    
+    def test_string_single_source(self):
+        result = resolve_knowledge("docs/")
+        assert isinstance(result, KnowledgeConfig)
+        assert result.sources == ["docs/"]
+    
+    def test_array_sources(self):
+        result = resolve_knowledge(["docs/", "data.pdf"])
+        assert isinstance(result, KnowledgeConfig)
+        assert result.sources == ["docs/", "data.pdf"]
+    
+    def test_dict_expansion(self):
+        result = resolve_knowledge({
+            "sources": ["docs/"],
+            "rerank": True,
+        })
+        assert isinstance(result, KnowledgeConfig)
+        assert result.sources == ["docs/"]
+        assert result.rerank is True
+    
+    def test_config_passthrough(self):
+        config = KnowledgeConfig(sources=["test/"])
+        result = resolve_knowledge(config)
+        assert result is config
+
+
+class TestResolvePlanning:
+    """Test resolve_planning precedence ladder."""
+    
+    def test_default_none(self):
+        assert resolve_planning(None) is None
+    
+    def test_bool_false(self):
+        assert resolve_planning(False) is None
+    
+    def test_bool_true(self):
+        result = resolve_planning(True)
+        assert isinstance(result, PlanningConfig)
+    
+    def test_config_passthrough(self):
+        config = PlanningConfig()
+        result = resolve_planning(config)
+        assert result is config
+
+
+class TestResolveReflection:
+    """Test resolve_reflection precedence ladder."""
+    
+    def test_default_none(self):
+        assert resolve_reflection(None) is None
+    
+    def test_bool_false(self):
+        assert resolve_reflection(False) is None
+    
+    def test_bool_true(self):
+        result = resolve_reflection(True)
+        assert isinstance(result, ReflectionConfig)
+
+
+class TestResolveGuardrails:
+    """Test resolve_guardrails precedence ladder."""
+    
+    def test_default_none(self):
+        assert resolve_guardrails(None) is None
+    
+    def test_bool_false(self):
+        assert resolve_guardrails(False) is None
+    
+    def test_bool_true(self):
+        result = resolve_guardrails(True)
+        assert isinstance(result, GuardrailConfig)
+    
+    def test_callable_validator(self):
+        def my_validator(output):
+            return True, output
+        result = resolve_guardrails(my_validator)
+        assert isinstance(result, GuardrailConfig)
+        assert result.validator is my_validator
+
+
+class TestResolveWeb:
+    """Test resolve_web precedence ladder."""
+    
+    def test_default_none(self):
+        assert resolve_web(None) is None
+    
+    def test_bool_false(self):
+        assert resolve_web(False) is None
+    
+    def test_bool_true(self):
+        result = resolve_web(True)
+        assert isinstance(result, WebConfig)
+
+
+class TestResolveCaching:
+    """Test resolve_caching precedence ladder."""
+    
+    def test_default_none(self):
+        assert resolve_caching(None) is None
+    
+    def test_bool_false(self):
+        assert resolve_caching(False) is None
+    
+    def test_bool_true(self):
+        result = resolve_caching(True)
+        assert isinstance(result, CachingConfig)
+
+
+class TestResolveAutonomy:
+    """Test resolve_autonomy precedence ladder."""
+    
+    def test_default_none(self):
+        assert resolve_autonomy(None) is None
+    
+    def test_bool_false(self):
+        assert resolve_autonomy(False) is None
+    
+    def test_bool_true(self):
+        result = resolve_autonomy(True)
+        assert isinstance(result, AutonomyConfig)
+
+
+class TestNoCortexParam:
+    """Test that cortex parameter does not exist."""
+    
+    def test_agent_has_no_cortex_param(self):
+        """Agent.__init__ should not have cortex parameter."""
+        import inspect
+        from praisonaiagents import Agent
+        
+        sig = inspect.signature(Agent.__init__)
+        param_names = list(sig.parameters.keys())
+        
+        assert "cortex" not in param_names, "cortex param should be removed from Agent"
+    
+    def test_no_cortex_module_in_agents(self):
+        """praisonaiagents should not have cortex module."""
+        import importlib.util
+        
+        spec = importlib.util.find_spec("praisonaiagents.cortex")
+        assert spec is None, "praisonaiagents.cortex module should not exist"

src/praisonai-agents/tests/unit/memory/test_learn.py

@@ -0,0 +1,292 @@
+"""Tests for the learn module - continuous learning within memory system."""
+
+import os
+import tempfile
+import shutil
+from pathlib import Path
+
+from praisonaiagents.config.feature_configs import LearnConfig, LearnScope, MemoryConfig
+from praisonaiagents.memory.learn.stores import (
+    PersonaStore,
+    InsightStore,
+    ThreadStore,
+    PatternStore,
+    DecisionStore,
+    FeedbackStore,
+    ImprovementStore,
+    LearnEntry,
+)
+from praisonaiagents.memory.learn.manager import LearnManager
+
+
+class TestLearnConfig:
+    """Test LearnConfig dataclass."""
+    
+    def test_default_values(self):
+        """Test default LearnConfig values."""
+        config = LearnConfig()
+        assert config.persona is True
+        assert config.insights is True
+        assert config.thread is True
+        assert config.patterns is False
+        assert config.decisions is False
+        assert config.feedback is False
+        assert config.improvements is False
+        assert config.scope == LearnScope.USER
+        assert config.auto_consolidate is True
+        assert config.retention_days is None
+    
+    def test_custom_values(self):
+        """Test LearnConfig with custom values."""
+        config = LearnConfig(
+            persona=False,
+            insights=True,
+            patterns=True,
+            scope=LearnScope.TEAM,
+            retention_days=30,
+        )
+        assert config.persona is False
+        assert config.patterns is True
+        assert config.scope == LearnScope.TEAM
+        assert config.retention_days == 30
+    
+    def test_to_dict(self):
+        """Test LearnConfig to_dict method."""
+        config = LearnConfig(persona=True, insights=True)
+        d = config.to_dict()
+        assert d["persona"] is True
+        assert d["insights"] is True
+        assert d["scope"] == "user"
+
+
+class TestMemoryConfigWithLearn:
+    """Test MemoryConfig with learn field."""
+    
+    def test_learn_none_by_default(self):
+        """Test that learn is None by default."""
+        config = MemoryConfig()
+        assert config.learn is None
+    
+    def test_learn_bool_true(self):
+        """Test learn=True."""
+        config = MemoryConfig(learn=True)
+        assert config.learn is True
+        d = config.to_dict()
+        assert d["learn"] is not None
+        assert d["learn"]["persona"] is True
+    
+    def test_learn_config_instance(self):
+        """Test learn=LearnConfig(...)."""
+        learn_config = LearnConfig(persona=True, patterns=True)
+        config = MemoryConfig(learn=learn_config)
+        assert config.learn is learn_config
+        d = config.to_dict()
+        assert d["learn"]["patterns"] is True
+
+
+class TestLearnEntry:
+    """Test LearnEntry dataclass."""
+    
+    def test_create_entry(self):
+        """Test creating a LearnEntry."""
+        entry = LearnEntry(
+            id="test_1",
+            content="Test content",
+            metadata={"key": "value"},
+        )
+        assert entry.id == "test_1"
+        assert entry.content == "Test content"
+        assert entry.metadata["key"] == "value"
+    
+    def test_to_dict(self):
+        """Test LearnEntry to_dict."""
+        entry = LearnEntry(id="test_1", content="Test")
+        d = entry.to_dict()
+        assert d["id"] == "test_1"
+        assert d["content"] == "Test"
+        assert "created_at" in d
+    
+    def test_from_dict(self):
+        """Test LearnEntry from_dict."""
+        data = {
+            "id": "test_1",
+            "content": "Test content",
+            "metadata": {"key": "value"},
+        }
+        entry = LearnEntry.from_dict(data)
+        assert entry.id == "test_1"
+        assert entry.content == "Test content"
+
+
+class TestPersonaStore:
+    """Test PersonaStore."""
+    
+    def setup_method(self):
+        """Set up test fixtures."""
+        self.temp_dir = tempfile.mkdtemp()
+        self.store_path = os.path.join(self.temp_dir, "persona.json")
+    
+    def teardown_method(self):
+        """Clean up test fixtures."""
+        shutil.rmtree(self.temp_dir, ignore_errors=True)
+    
+    def test_add_preference(self):
+        """Test adding a preference."""
+        store = PersonaStore(store_path=self.store_path, user_id="test_user")
+        entry = store.add_preference("User prefers concise responses", "communication")
+        assert entry is not None
+        assert "concise" in entry.content
+        assert entry.metadata["category"] == "communication"
+    
+    def test_search(self):
+        """Test searching entries."""
+        store = PersonaStore(store_path=self.store_path, user_id="test_user")
+        store.add_preference("User likes Python", "programming")
+        store.add_preference("User prefers dark mode", "ui")
+        
+        results = store.search("Python")
+        assert len(results) == 1
+        assert "Python" in results[0].content
+
+
+class TestInsightStore:
+    """Test InsightStore."""
+    
+    def setup_method(self):
+        """Set up test fixtures."""
+        self.temp_dir = tempfile.mkdtemp()
+        self.store_path = os.path.join(self.temp_dir, "insights.json")
+    
+    def teardown_method(self):
+        """Clean up test fixtures."""
+        shutil.rmtree(self.temp_dir, ignore_errors=True)
+    
+    def test_add_insight(self):
+        """Test adding an insight."""
+        store = InsightStore(store_path=self.store_path, user_id="test_user")
+        entry = store.add_insight("User works in data science", "conversation")
+        assert entry is not None
+        assert "data science" in entry.content
+
+
+class TestLearnManager:
+    """Test LearnManager."""
+    
+    def setup_method(self):
+        """Set up test fixtures."""
+        self.temp_dir = tempfile.mkdtemp()
+    
+    def teardown_method(self):
+        """Clean up test fixtures."""
+        shutil.rmtree(self.temp_dir, ignore_errors=True)
+    
+    def test_create_manager_with_defaults(self):
+        """Test creating LearnManager with default config."""
+        config = LearnConfig()
+        manager = LearnManager(config=config, user_id="test_user", store_path=self.temp_dir)
+        
+        assert "persona" in manager._stores
+        assert "insights" in manager._stores
+        assert "threads" in manager._stores
+    
+    def test_capture_persona(self):
+        """Test capturing persona."""
+        config = LearnConfig(persona=True)
+        manager = LearnManager(config=config, user_id="test_user", store_path=self.temp_dir)
+        
+        entry = manager.capture_persona("User prefers detailed explanations")
+        assert entry is not None
+        assert "detailed" in entry.content
+    
+    def test_capture_insight(self):
+        """Test capturing insight."""
+        config = LearnConfig(insights=True)
+        manager = LearnManager(config=config, user_id="test_user", store_path=self.temp_dir)
+        
+        entry = manager.capture_insight("User is experienced with Python")
+        assert entry is not None
+    
+    def test_get_learning_context(self):
+        """Test getting learning context."""
+        config = LearnConfig(persona=True, insights=True)
+        manager = LearnManager(config=config, user_id="test_user", store_path=self.temp_dir)
+        
+        manager.capture_persona("User likes concise code")
+        manager.capture_insight("User knows Python")
+        
+        context = manager.get_learning_context()
+        assert "persona" in context
+        assert "insights" in context
+        assert len(context["persona"]) == 1
+        assert len(context["insights"]) == 1
+    
+    def test_search(self):
+        """Test searching across stores."""
+        config = LearnConfig(persona=True, insights=True)
+        manager = LearnManager(config=config, user_id="test_user", store_path=self.temp_dir)
+        
+        manager.capture_persona("User prefers Python")
+        manager.capture_insight("User knows Python well")
+        
+        results = manager.search("Python")
+        assert "persona" in results or "insights" in results
+    
+    def test_clear_all(self):
+        """Test clearing all stores."""
+        config = LearnConfig(persona=True, insights=True)
+        manager = LearnManager(config=config, user_id="test_user", store_path=self.temp_dir)
+        
+        manager.capture_persona("Test 1")
+        manager.capture_insight("Test 2")
+        
+        cleared = manager.clear_all()
+        assert cleared["persona"] == 1
+        assert cleared["insights"] == 1
+    
+    def test_get_stats(self):
+        """Test getting stats."""
+        config = LearnConfig(persona=True, insights=True)
+        manager = LearnManager(config=config, user_id="test_user", store_path=self.temp_dir)
+        
+        manager.capture_persona("Test 1")
+        manager.capture_persona("Test 2")
+        manager.capture_insight("Test 3")
+        
+        stats = manager.get_stats()
+        assert stats["persona"] == 2
+        assert stats["insights"] == 1
+    
+    def test_to_system_prompt_context(self):
+        """Test generating system prompt context."""
+        config = LearnConfig(persona=True, insights=True)
+        manager = LearnManager(config=config, user_id="test_user", store_path=self.temp_dir)
+        
+        manager.capture_persona("User prefers concise responses")
+        manager.capture_insight("User is a Python developer")
+        
+        context = manager.to_system_prompt_context()
+        assert "User Preferences" in context
+        assert "concise" in context
+        assert "Learned Insights" in context
+        assert "Python" in context
+
+
+class TestNoCortexParam:
+    """Test that cortex parameter does not exist."""
+    
+    def test_agent_has_no_cortex_param(self):
+        """Agent.__init__ should not have cortex parameter."""
+        import inspect
+        from praisonaiagents import Agent
+        
+        sig = inspect.signature(Agent.__init__)
+        param_names = list(sig.parameters.keys())
+        
+        assert "cortex" not in param_names, "cortex param should be removed from Agent"
+    
+    def test_no_cortex_module_in_agents(self):
+        """praisonaiagents should not have cortex module."""
+        import importlib.util
+        
+        spec = importlib.util.find_spec("praisonaiagents.cortex")
+        assert spec is None, "praisonaiagents.cortex module should not exist"

src/praisonai-agents/tests/unit/streaming/__init__.py

@@ -0,0 +1 @@
+"""Unit tests for streaming module."""

src/praisonai-agents/tests/unit/streaming/test_events.py

@@ -0,0 +1,387 @@
+"""
+Unit tests for streaming events module.
+
+Tests cover:
+- StreamEvent creation and representation
+- StreamEventType enum values
+- StreamMetrics calculations (TTFT, stream duration, total time)
+- StreamEventEmitter callback management
+- create_text_printer_callback and create_metrics_callback helpers
+"""
+
+import time
+from praisonaiagents.streaming.events import (
+    StreamEvent,
+    StreamEventType,
+    StreamMetrics,
+    StreamEventEmitter,
+    create_text_printer_callback,
+    create_metrics_callback,
+)
+
+
+class TestStreamEventType:
+    """Tests for StreamEventType enum."""
+    
+    def test_event_types_exist(self):
+        """All required event types should exist."""
+        assert StreamEventType.REQUEST_START.value == "request_start"
+        assert StreamEventType.HEADERS_RECEIVED.value == "headers_received"
+        assert StreamEventType.FIRST_TOKEN.value == "first_token"
+        assert StreamEventType.DELTA_TEXT.value == "delta_text"
+        assert StreamEventType.DELTA_TOOL_CALL.value == "delta_tool_call"
+        assert StreamEventType.TOOL_CALL_END.value == "tool_call_end"
+        assert StreamEventType.LAST_TOKEN.value == "last_token"
+        assert StreamEventType.STREAM_END.value == "stream_end"
+        assert StreamEventType.ERROR.value == "error"
+
+
+class TestStreamEvent:
+    """Tests for StreamEvent dataclass."""
+    
+    def test_create_basic_event(self):
+        """Create a basic event with type only."""
+        event = StreamEvent(type=StreamEventType.REQUEST_START)
+        assert event.type == StreamEventType.REQUEST_START
+        assert event.timestamp > 0
+        assert event.content is None
+        assert event.tool_call is None
+        assert event.metadata is None
+        assert event.error is None
+    
+    def test_create_event_with_content(self):
+        """Create an event with content."""
+        event = StreamEvent(
+            type=StreamEventType.DELTA_TEXT,
+            content="Hello"
+        )
+        assert event.type == StreamEventType.DELTA_TEXT
+        assert event.content == "Hello"
+    
+    def test_create_event_with_tool_call(self):
+        """Create an event with tool call data."""
+        tool_data = {"name": "calculator", "arguments": '{"x": 1}'}
+        event = StreamEvent(
+            type=StreamEventType.DELTA_TOOL_CALL,
+            tool_call=tool_data
+        )
+        assert event.type == StreamEventType.DELTA_TOOL_CALL
+        assert event.tool_call == tool_data
+    
+    def test_create_event_with_metadata(self):
+        """Create an event with metadata."""
+        event = StreamEvent(
+            type=StreamEventType.REQUEST_START,
+            metadata={"model": "gpt-4o-mini", "message_count": 2}
+        )
+        assert event.metadata["model"] == "gpt-4o-mini"
+        assert event.metadata["message_count"] == 2
+    
+    def test_create_error_event(self):
+        """Create an error event."""
+        event = StreamEvent(
+            type=StreamEventType.ERROR,
+            error="Connection timeout"
+        )
+        assert event.type == StreamEventType.ERROR
+        assert event.error == "Connection timeout"
+    
+    def test_event_repr(self):
+        """Event repr should be informative."""
+        event = StreamEvent(
+            type=StreamEventType.DELTA_TEXT,
+            content="Hello world"
+        )
+        repr_str = repr(event)
+        assert "DELTA_TEXT" in repr_str or "delta_text" in repr_str
+        assert "Hello world" in repr_str
+
+
+class TestStreamMetrics:
+    """Tests for StreamMetrics dataclass."""
+    
+    def test_default_metrics(self):
+        """Default metrics should be zero."""
+        metrics = StreamMetrics()
+        assert metrics.request_start == 0.0
+        assert metrics.headers_received == 0.0
+        assert metrics.first_token == 0.0
+        assert metrics.last_token == 0.0
+        assert metrics.stream_end == 0.0
+        assert metrics.token_count == 0
+    
+    def test_ttft_calculation(self):
+        """TTFT should be first_token - request_start."""
+        metrics = StreamMetrics()
+        metrics.request_start = 100.0
+        metrics.first_token = 100.5
+        assert metrics.ttft == 0.5
+    
+    def test_ttft_zero_when_no_first_token(self):
+        """TTFT should be 0 when first_token not set."""
+        metrics = StreamMetrics()
+        metrics.request_start = 100.0
+        assert metrics.ttft == 0.0
+    
+    def test_stream_duration_calculation(self):
+        """Stream duration should be last_token - first_token."""
+        metrics = StreamMetrics()
+        metrics.first_token = 100.0
+        metrics.last_token = 102.0
+        assert metrics.stream_duration == 2.0
+    
+    def test_total_time_calculation(self):
+        """Total time should be stream_end - request_start."""
+        metrics = StreamMetrics()
+        metrics.request_start = 100.0
+        metrics.stream_end = 103.0
+        assert metrics.total_time == 3.0
+    
+    def test_tokens_per_second(self):
+        """Tokens per second calculation."""
+        metrics = StreamMetrics()
+        metrics.first_token = 100.0
+        metrics.last_token = 102.0  # 2 second duration
+        metrics.token_count = 100
+        assert metrics.tokens_per_second == 50.0
+    
+    def test_tokens_per_second_zero_duration(self):
+        """Tokens per second should be 0 when duration is 0."""
+        metrics = StreamMetrics()
+        assert metrics.tokens_per_second == 0.0
+    
+    def test_update_from_event(self):
+        """Metrics should update from events."""
+        metrics = StreamMetrics()
+        
+        # Request start
+        event1 = StreamEvent(type=StreamEventType.REQUEST_START, timestamp=100.0)
+        metrics.update_from_event(event1)
+        assert metrics.request_start == 100.0
+        
+        # Headers received
+        event2 = StreamEvent(type=StreamEventType.HEADERS_RECEIVED, timestamp=100.2)
+        metrics.update_from_event(event2)
+        assert metrics.headers_received == 100.2
+        
+        # First token
+        event3 = StreamEvent(type=StreamEventType.FIRST_TOKEN, timestamp=100.5)
+        metrics.update_from_event(event3)
+        assert metrics.first_token == 100.5
+        assert metrics.token_count == 1
+        
+        # Delta text
+        event4 = StreamEvent(type=StreamEventType.DELTA_TEXT, timestamp=100.6)
+        metrics.update_from_event(event4)
+        assert metrics.token_count == 2
+        
+        # Last token
+        event5 = StreamEvent(type=StreamEventType.LAST_TOKEN, timestamp=102.0)
+        metrics.update_from_event(event5)
+        assert metrics.last_token == 102.0
+        
+        # Stream end
+        event6 = StreamEvent(type=StreamEventType.STREAM_END, timestamp=102.1)
+        metrics.update_from_event(event6)
+        assert metrics.stream_end == 102.1
+    
+    def test_to_dict(self):
+        """Metrics should convert to dict."""
+        metrics = StreamMetrics()
+        metrics.request_start = 100.0
+        metrics.first_token = 100.5
+        metrics.last_token = 102.0
+        metrics.stream_end = 102.1
+        metrics.token_count = 50
+        
+        d = metrics.to_dict()
+        assert "ttft_ms" in d
+        assert "stream_duration_ms" in d
+        assert "total_time_ms" in d
+        assert "token_count" in d
+        assert "tokens_per_second" in d
+    
+    def test_format_summary(self):
+        """Metrics should format as summary string."""
+        metrics = StreamMetrics()
+        metrics.request_start = 100.0
+        metrics.first_token = 100.5
+        metrics.last_token = 102.0
+        metrics.stream_end = 102.1
+        metrics.token_count = 50
+        
+        summary = metrics.format_summary()
+        assert "TTFT" in summary
+        assert "Stream" in summary
+        assert "Total" in summary
+        assert "Tokens" in summary
+
+
+class TestStreamEventEmitter:
+    """Tests for StreamEventEmitter."""
+    
+    def test_create_emitter(self):
+        """Create an emitter."""
+        emitter = StreamEventEmitter()
+        assert not emitter.has_callbacks
+    
+    def test_add_callback(self):
+        """Add a callback."""
+        emitter = StreamEventEmitter()
+        events = []
+        
+        def callback(event):
+            events.append(event)
+        
+        emitter.add_callback(callback)
+        assert emitter.has_callbacks
+    
+    def test_remove_callback(self):
+        """Remove a callback."""
+        emitter = StreamEventEmitter()
+        events = []
+        
+        def callback(event):
+            events.append(event)
+        
+        emitter.add_callback(callback)
+        emitter.remove_callback(callback)
+        assert not emitter.has_callbacks
+    
+    def test_emit_calls_callbacks(self):
+        """Emit should call all callbacks."""
+        emitter = StreamEventEmitter()
+        events1 = []
+        events2 = []
+        
+        emitter.add_callback(lambda e: events1.append(e))
+        emitter.add_callback(lambda e: events2.append(e))
+        
+        event = StreamEvent(type=StreamEventType.DELTA_TEXT, content="Hello")
+        emitter.emit(event)
+        
+        assert len(events1) == 1
+        assert len(events2) == 1
+        assert events1[0].content == "Hello"
+    
+    def test_emit_with_metrics(self):
+        """Emit should update metrics when enabled."""
+        emitter = StreamEventEmitter()
+        emitter.enable_metrics()
+        
+        emitter.emit(StreamEvent(type=StreamEventType.REQUEST_START, timestamp=100.0))
+        emitter.emit(StreamEvent(type=StreamEventType.FIRST_TOKEN, timestamp=100.5))
+        emitter.emit(StreamEvent(type=StreamEventType.STREAM_END, timestamp=102.0))
+        
+        metrics = emitter.get_metrics()
+        assert metrics is not None
+        assert metrics.request_start == 100.0
+        assert metrics.first_token == 100.5
+        assert metrics.stream_end == 102.0
+    
+    def test_callback_error_does_not_break_emission(self):
+        """Callback errors should not break other callbacks."""
+        emitter = StreamEventEmitter()
+        events = []
+        
+        def bad_callback(event):
+            raise ValueError("Test error")
+        
+        def good_callback(event):
+            events.append(event)
+        
+        emitter.add_callback(bad_callback)
+        emitter.add_callback(good_callback)
+        
+        event = StreamEvent(type=StreamEventType.DELTA_TEXT)
+        emitter.emit(event)  # Should not raise
+        
+        assert len(events) == 1
+    
+    def test_reset_metrics(self):
+        """Reset metrics should create new metrics."""
+        emitter = StreamEventEmitter()
+        emitter.enable_metrics()
+        
+        emitter.emit(StreamEvent(type=StreamEventType.REQUEST_START, timestamp=100.0))
+        emitter.reset_metrics()
+        
+        metrics = emitter.get_metrics()
+        assert metrics.request_start == 0.0
+
+
+class TestHelperFunctions:
+    """Tests for helper functions."""
+    
+    def test_create_text_printer_callback(self, capsys):
+        """Text printer callback should print content."""
+        callback = create_text_printer_callback(flush=True)
+        
+        callback(StreamEvent(type=StreamEventType.DELTA_TEXT, content="Hello"))
+        callback(StreamEvent(type=StreamEventType.DELTA_TEXT, content=" World"))
+        callback(StreamEvent(type=StreamEventType.STREAM_END))
+        
+        captured = capsys.readouterr()
+        assert "Hello World" in captured.out
+    
+    def test_create_metrics_callback(self):
+        """Metrics callback should update metrics."""
+        metrics = StreamMetrics()
+        callback = create_metrics_callback(metrics)
+        
+        callback(StreamEvent(type=StreamEventType.REQUEST_START, timestamp=100.0))
+        callback(StreamEvent(type=StreamEventType.FIRST_TOKEN, timestamp=100.5))
+        
+        assert metrics.request_start == 100.0
+        assert metrics.first_token == 100.5
+
+
+class TestStreamingEventOrdering:
+    """Tests for event ordering and pass-through behavior."""
+    
+    def test_events_emitted_immediately(self):
+        """Events should be emitted immediately, not buffered."""
+        emitter = StreamEventEmitter()
+        timestamps = []
+        
+        def callback(event):
+            timestamps.append((event.type, time.perf_counter()))
+        
+        emitter.add_callback(callback)
+        
+        # Simulate streaming with small delays
+        emitter.emit(StreamEvent(type=StreamEventType.REQUEST_START))
+        time.sleep(0.01)
+        emitter.emit(StreamEvent(type=StreamEventType.FIRST_TOKEN))
+        time.sleep(0.01)
+        emitter.emit(StreamEvent(type=StreamEventType.DELTA_TEXT))
+        
+        # Verify events were captured in order
+        assert len(timestamps) == 3
+        assert timestamps[0][0] == StreamEventType.REQUEST_START
+        assert timestamps[1][0] == StreamEventType.FIRST_TOKEN
+        assert timestamps[2][0] == StreamEventType.DELTA_TEXT
+        
+        # Verify timestamps are increasing (no buffering)
+        assert timestamps[0][1] < timestamps[1][1] < timestamps[2][1]
+    
+    def test_no_buffering_before_first_token(self):
+        """Events should not be buffered before first token."""
+        emitter = StreamEventEmitter()
+        emitter.enable_metrics()
+        
+        # Emit request start
+        t1 = time.perf_counter()
+        emitter.emit(StreamEvent(type=StreamEventType.REQUEST_START, timestamp=t1))
+        
+        # Small delay (simulating network)
+        time.sleep(0.05)
+        
+        # Emit first token
+        t2 = time.perf_counter()
+        emitter.emit(StreamEvent(type=StreamEventType.FIRST_TOKEN, timestamp=t2))
+        
+        metrics = emitter.get_metrics()
+        
+        # TTFT should reflect the actual delay
+        assert metrics.ttft >= 0.04  # Allow some tolerance

src/praisonai-agents/tests/unit/test_profiling.py

@@ -0,0 +1,371 @@
+"""
+Unit tests for the core SDK profiling module.
+
+Tests cover:
+- Profiler enable/disable
+- Timing recording
+- Streaming tracking
+- Phase tracking
+- Report generation
+- Zero overhead when disabled
+"""
+
+import pytest
+import time
+from unittest.mock import patch
+
+
+class TestProfilerBasics:
+    """Test basic Profiler functionality."""
+    
+    def test_profiler_disabled_by_default(self):
+        """Profiler should be disabled by default."""
+        from praisonaiagents.profiling import Profiler
+        Profiler.disable()  # Ensure clean state
+        assert not Profiler.is_enabled()
+    
+    def test_profiler_enable_disable(self):
+        """Test enable/disable functionality."""
+        from praisonaiagents.profiling import Profiler
+        
+        Profiler.disable()
+        assert not Profiler.is_enabled()
+        
+        Profiler.enable()
+        assert Profiler.is_enabled()
+        
+        Profiler.disable()
+        assert not Profiler.is_enabled()
+    
+    def test_profiler_env_var_enable(self):
+        """Test enabling via environment variable."""
+        from praisonaiagents.profiling import Profiler
+        
+        Profiler.disable()
+        
+        with patch.dict('os.environ', {'PRAISONAI_PROFILE': '1'}):
+            assert Profiler.is_enabled()
+        
+        with patch.dict('os.environ', {'PRAISONAI_PROFILE': 'true'}):
+            assert Profiler.is_enabled()
+        
+        with patch.dict('os.environ', {'PRAISONAI_PROFILE': 'yes'}):
+            assert Profiler.is_enabled()
+    
+    def test_profiler_clear(self):
+        """Test clearing profiling data."""
+        from praisonaiagents.profiling import Profiler
+        
+        Profiler.enable()
+        Profiler.record_timing("test", 100.0)
+        assert len(Profiler.get_timings()) > 0
+        
+        Profiler.clear()
+        assert len(Profiler.get_timings()) == 0
+        Profiler.disable()
+
+
+class TestTimingRecording:
+    """Test timing recording functionality."""
+    
+    def test_record_timing_when_enabled(self):
+        """Timing should be recorded when profiler is enabled."""
+        from praisonaiagents.profiling import Profiler
+        
+        Profiler.enable()
+        Profiler.clear()
+        
+        Profiler.record_timing("test_op", 123.45, category="test")
+        
+        timings = Profiler.get_timings()
+        assert len(timings) == 1
+        assert timings[0].name == "test_op"
+        assert timings[0].duration_ms == 123.45
+        assert timings[0].category == "test"
+        
+        Profiler.disable()
+    
+    def test_record_timing_when_disabled(self):
+        """Timing should NOT be recorded when profiler is disabled."""
+        from praisonaiagents.profiling import Profiler
+        
+        Profiler.disable()
+        Profiler.clear()
+        
+        Profiler.record_timing("test_op", 123.45)
+        
+        timings = Profiler.get_timings()
+        assert len(timings) == 0
+    
+    def test_record_timing_with_metadata(self):
+        """Timing should include metadata."""
+        from praisonaiagents.profiling import Profiler
+        
+        Profiler.enable()
+        Profiler.clear()
+        
+        Profiler.record_timing("test_op", 100.0, metadata={"key": "value"})
+        
+        timings = Profiler.get_timings()
+        assert timings[0].metadata == {"key": "value"}
+        
+        Profiler.disable()
+
+
+class TestPhaseTracking:
+    """Test phase tracking functionality."""
+    
+    def test_phase_context_manager(self):
+        """Phase context manager should record timing."""
+        from praisonaiagents.profiling import Profiler
+        
+        Profiler.enable()
+        Profiler.clear()
+        
+        with Profiler.phase("test_phase"):
+            time.sleep(0.01)  # 10ms
+        
+        timings = Profiler.get_timings(category="phase")
+        assert len(timings) == 1
+        assert timings[0].name == "test_phase"
+        assert timings[0].duration_ms >= 10  # At least 10ms
+        
+        Profiler.disable()
+    
+    def test_nested_phases(self):
+        """Nested phases should track parent."""
+        from praisonaiagents.profiling import Profiler
+        
+        Profiler.enable()
+        Profiler.clear()
+        
+        with Profiler.phase("outer"):
+            with Profiler.phase("inner"):
+                time.sleep(0.005)
+        
+        timings = Profiler.get_timings(category="phase")
+        assert len(timings) == 2
+        
+        inner = next(t for t in timings if t.name == "inner")
+        assert inner.parent == "outer"
+        
+        Profiler.disable()
+
+
+class TestStreamingTracker:
+    """Test streaming tracker functionality."""
+    
+    def test_streaming_ttft(self):
+        """Streaming tracker should measure TTFT."""
+        from praisonaiagents.profiling import Profiler, StreamingTracker
+        
+        Profiler.enable()
+        Profiler.clear()
+        
+        tracker = StreamingTracker("test_stream")
+        tracker.start()
+        time.sleep(0.01)  # 10ms to first token
+        tracker.first_token()
+        time.sleep(0.02)  # 20ms more
+        tracker.end()
+        
+        assert tracker.ttft_ms >= 10
+        assert tracker.elapsed_ms >= 30
+        
+        records = Profiler.get_streaming_records()
+        assert len(records) == 1
+        assert records[0].name == "test_stream"
+        assert records[0].ttft_ms >= 10
+        
+        Profiler.disable()
+    
+    def test_streaming_context_manager(self):
+        """Streaming context manager should work."""
+        from praisonaiagents.profiling import Profiler
+        
+        Profiler.enable()
+        Profiler.clear()
+        
+        with Profiler.streaming("test_stream") as tracker:
+            time.sleep(0.01)
+            tracker.first_token()
+            tracker.chunk()
+            tracker.chunk()
+        
+        records = Profiler.get_streaming_records()
+        assert len(records) == 1
+        assert records[0].chunk_count == 2
+        
+        Profiler.disable()
+
+
+class TestProfileDecorators:
+    """Test profiling decorators."""
+    
+    def test_profile_func_decorator(self):
+        """profile_func decorator should record function timing."""
+        from praisonaiagents.profiling import Profiler, profile_func
+        
+        Profiler.enable()
+        Profiler.clear()
+        
+        @profile_func
+        def slow_function():
+            time.sleep(0.01)
+            return "done"
+        
+        result = slow_function()
+        assert result == "done"
+        
+        timings = Profiler.get_timings(category="function")
+        assert len(timings) == 1
+        assert timings[0].name == "slow_function"
+        assert timings[0].duration_ms >= 10
+        
+        Profiler.disable()
+    
+    def test_profile_func_disabled(self):
+        """profile_func should have zero overhead when disabled."""
+        from praisonaiagents.profiling import Profiler, profile_func
+        
+        Profiler.disable()
+        Profiler.clear()
+        
+        @profile_func
+        def fast_function():
+            return "done"
+        
+        result = fast_function()
+        assert result == "done"
+        
+        timings = Profiler.get_timings()
+        assert len(timings) == 0
+
+
+class TestProfileBlock:
+    """Test profile_block context manager."""
+    
+    def test_profile_block(self):
+        """profile_block should record block timing."""
+        from praisonaiagents.profiling import Profiler, profile_block
+        
+        Profiler.enable()
+        Profiler.clear()
+        
+        with profile_block("my_block", category="custom"):
+            time.sleep(0.01)
+        
+        timings = Profiler.get_timings(category="custom")
+        assert len(timings) == 1
+        assert timings[0].name == "my_block"
+        
+        Profiler.disable()
+    
+    def test_profile_block_disabled(self):
+        """profile_block should have zero overhead when disabled."""
+        from praisonaiagents.profiling import Profiler, profile_block
+        
+        Profiler.disable()
+        Profiler.clear()
+        
+        with profile_block("my_block"):
+            pass
+        
+        timings = Profiler.get_timings()
+        assert len(timings) == 0
+
+
+class TestReporting:
+    """Test report generation."""
+    
+    def test_get_summary(self):
+        """get_summary should return structured data."""
+        from praisonaiagents.profiling import Profiler
+        
+        Profiler.enable()
+        Profiler.clear()
+        
+        Profiler.record_timing("op1", 100.0, category="cat1")
+        Profiler.record_timing("op2", 200.0, category="cat2")
+        
+        summary = Profiler.get_summary()
+        
+        assert summary['total_time_ms'] == 300.0
+        assert summary['timing_count'] == 2
+        assert 'cat1' in summary['by_category']
+        assert 'cat2' in summary['by_category']
+        assert len(summary['slowest_operations']) == 2
+        
+        Profiler.disable()
+    
+    def test_export_json(self):
+        """export_json should return valid JSON."""
+        from praisonaiagents.profiling import Profiler
+        import json
+        
+        Profiler.enable()
+        Profiler.clear()
+        
+        Profiler.record_timing("test", 100.0)
+        
+        json_str = Profiler.export_json()
+        data = json.loads(json_str)
+        
+        assert 'summary' in data
+        assert 'timings' in data
+        assert len(data['timings']) == 1
+        
+        Profiler.disable()
+
+
+class TestWarmup:
+    """Test warmup function."""
+    
+    def test_warmup_returns_timings(self):
+        """warmup should return timing information."""
+        from praisonaiagents import warmup
+        
+        result = warmup(include_litellm=False, include_openai=False)
+        assert isinstance(result, dict)
+    
+    def test_warmup_with_litellm(self):
+        """warmup with litellm should return litellm timing."""
+        from praisonaiagents import warmup
+        
+        result = warmup(include_litellm=True, include_openai=False)
+        assert 'litellm' in result
+        # Should be positive (imported) or -1 (not available)
+        assert result['litellm'] > 0 or result['litellm'] == -1
+
+
+class TestZeroOverhead:
+    """Test that profiling has zero overhead when disabled."""
+    
+    def test_record_timing_overhead(self):
+        """record_timing should be fast when disabled."""
+        from praisonaiagents.profiling import Profiler
+        
+        Profiler.disable()
+        
+        start = time.perf_counter()
+        for _ in range(10000):
+            Profiler.record_timing("test", 100.0)
+        elapsed = (time.perf_counter() - start) * 1000
+        
+        # Should complete 10k calls in under 10ms when disabled
+        assert elapsed < 10, f"record_timing took {elapsed}ms for 10k calls"
+    
+    def test_phase_context_overhead(self):
+        """phase context manager should be fast when disabled."""
+        from praisonaiagents.profiling import Profiler
+        
+        Profiler.disable()
+        
+        start = time.perf_counter()
+        for _ in range(1000):
+            with Profiler.phase("test"):
+                pass
+        elapsed = (time.perf_counter() - start) * 1000
+        
+        # Should complete 1k calls in under 10ms when disabled
+        assert elapsed < 10, f"phase took {elapsed}ms for 1k calls"

src/praisonai-agents/tests/unit/trace/__init__.py

@@ -0,0 +1 @@
+"""Tests for the trace module."""

src/praisonai-agents/tests/unit/trace/test_protocol.py

@@ -0,0 +1,450 @@
+"""
+Tests for the trace protocol module.
+
+TDD tests for ActionEvent, TraceSink protocol, and redaction.
+"""
+
+import time
+import json
+
+
+class TestActionEvent:
+    """Tests for ActionEvent dataclass."""
+    
+    def test_action_event_creation_minimal(self):
+        """Test creating ActionEvent with minimal fields."""
+        from praisonaiagents.trace.protocol import ActionEvent
+        
+        event = ActionEvent(
+            event_type="agent_start",
+            timestamp=1736665496.123,
+        )
+        
+        assert event.event_type == "agent_start"
+        assert event.timestamp == 1736665496.123
+        assert event.schema_version == "1.0"
+        assert event.agent_id is None
+        assert event.agent_name is None
+        assert event.tool_name is None
+    
+    def test_action_event_creation_full(self):
+        """Test creating ActionEvent with all fields."""
+        from praisonaiagents.trace.protocol import ActionEvent
+        
+        event = ActionEvent(
+            event_type="tool_end",
+            timestamp=1736665497.456,
+            agent_id="agent-123",
+            agent_name="researcher",
+            tool_name="web_search",
+            tool_args={"query": "AI trends"},
+            tool_result_summary="Found 10 results",
+            duration_ms=800.5,
+            status="ok",
+            error_message=None,
+            metadata={"provider": "duckduckgo"},
+        )
+        
+        assert event.event_type == "tool_end"
+        assert event.agent_name == "researcher"
+        assert event.tool_name == "web_search"
+        assert event.tool_args == {"query": "AI trends"}
+        assert event.duration_ms == 800.5
+        assert event.status == "ok"
+        assert event.metadata == {"provider": "duckduckgo"}
+    
+    def test_action_event_to_dict(self):
+        """Test ActionEvent.to_dict() method."""
+        from praisonaiagents.trace.protocol import ActionEvent
+        
+        event = ActionEvent(
+            event_type="agent_start",
+            timestamp=1736665496.123,
+            agent_name="writer",
+        )
+        
+        d = event.to_dict()
+        
+        assert d["schema_version"] == "1.0"
+        assert d["event_type"] == "agent_start"
+        assert d["timestamp"] == 1736665496.123
+        assert d["agent_name"] == "writer"
+        # None values should be excluded
+        assert "tool_name" not in d or d["tool_name"] is None
+    
+    def test_action_event_to_json(self):
+        """Test ActionEvent.to_json() method."""
+        from praisonaiagents.trace.protocol import ActionEvent
+        
+        event = ActionEvent(
+            event_type="tool_start",
+            timestamp=1736665497.0,
+            tool_name="read_file",
+            tool_args={"path": "/tmp/test.txt"},
+        )
+        
+        json_str = event.to_json()
+        parsed = json.loads(json_str)
+        
+        assert parsed["event_type"] == "tool_start"
+        assert parsed["tool_name"] == "read_file"
+        assert parsed["tool_args"] == {"path": "/tmp/test.txt"}
+    
+    def test_action_event_types(self):
+        """Test all valid event types."""
+        from praisonaiagents.trace.protocol import ActionEvent, ActionEventType
+        
+        valid_types = [
+            ActionEventType.AGENT_START,
+            ActionEventType.AGENT_END,
+            ActionEventType.TOOL_START,
+            ActionEventType.TOOL_END,
+            ActionEventType.ERROR,
+            ActionEventType.OUTPUT,
+        ]
+        
+        for event_type in valid_types:
+            event = ActionEvent(
+                event_type=event_type.value,
+                timestamp=time.time(),
+            )
+            assert event.event_type == event_type.value
+
+
+class TestRedaction:
+    """Tests for redaction functionality."""
+    
+    def test_redact_api_key(self):
+        """Test redacting api_key field."""
+        from praisonaiagents.trace.redact import redact_dict
+        
+        data = {"api_key": "sk-1234567890abcdef", "query": "hello"}
+        redacted = redact_dict(data)
+        
+        assert redacted["api_key"] == "[REDACTED]"
+        assert redacted["query"] == "hello"
+    
+    def test_redact_multiple_sensitive_keys(self):
+        """Test redacting multiple sensitive keys."""
+        from praisonaiagents.trace.redact import redact_dict
+        
+        data = {
+            "api_key": "sk-secret",
+            "password": "mypassword123",
+            "token": "bearer-token",
+            "authorization": "Bearer xyz",
+            "secret": "top-secret",
+            "normal_field": "visible",
+        }
+        redacted = redact_dict(data)
+        
+        assert redacted["api_key"] == "[REDACTED]"
+        assert redacted["password"] == "[REDACTED]"
+        assert redacted["token"] == "[REDACTED]"
+        assert redacted["authorization"] == "[REDACTED]"
+        assert redacted["secret"] == "[REDACTED]"
+        assert redacted["normal_field"] == "visible"
+    
+    def test_redact_nested_dict(self):
+        """Test redacting nested dictionaries."""
+        from praisonaiagents.trace.redact import redact_dict
+        
+        data = {
+            "config": {
+                "api_key": "sk-nested",
+                "endpoint": "https://api.example.com",
+            },
+            "name": "test",
+        }
+        redacted = redact_dict(data)
+        
+        assert redacted["config"]["api_key"] == "[REDACTED]"
+        assert redacted["config"]["endpoint"] == "https://api.example.com"
+        assert redacted["name"] == "test"
+    
+    def test_redact_list_of_dicts(self):
+        """Test redacting list containing dictionaries."""
+        from praisonaiagents.trace.redact import redact_dict
+        
+        data = {
+            "items": [
+                {"name": "item1", "secret": "hidden1"},
+                {"name": "item2", "secret": "hidden2"},
+            ]
+        }
+        redacted = redact_dict(data)
+        
+        assert redacted["items"][0]["name"] == "item1"
+        assert redacted["items"][0]["secret"] == "[REDACTED]"
+        assert redacted["items"][1]["secret"] == "[REDACTED]"
+    
+    def test_redact_case_insensitive(self):
+        """Test that redaction is case-insensitive."""
+        from praisonaiagents.trace.redact import redact_dict
+        
+        data = {
+            "API_KEY": "key1",
+            "Api_Key": "key2",
+            "apiKey": "key3",
+            "APIKEY": "key4",
+        }
+        redacted = redact_dict(data)
+        
+        assert redacted["API_KEY"] == "[REDACTED]"
+        assert redacted["Api_Key"] == "[REDACTED]"
+        assert redacted["apiKey"] == "[REDACTED]"
+        assert redacted["APIKEY"] == "[REDACTED]"
+    
+    def test_redact_connection_strings(self):
+        """Test redacting database connection strings."""
+        from praisonaiagents.trace.redact import redact_dict
+        
+        data = {
+            "connection_string": "postgresql://user:pass@host/db",
+            "database_url": "mysql://root:secret@localhost/mydb",
+            "db_url": "redis://default:password@redis:6379",
+        }
+        redacted = redact_dict(data)
+        
+        assert redacted["connection_string"] == "[REDACTED]"
+        assert redacted["database_url"] == "[REDACTED]"
+        assert redacted["db_url"] == "[REDACTED]"
+    
+    def test_redact_preserves_original(self):
+        """Test that redaction doesn't modify original dict."""
+        from praisonaiagents.trace.redact import redact_dict
+        
+        original = {"api_key": "secret", "name": "test"}
+        redacted = redact_dict(original)
+        
+        assert original["api_key"] == "secret"
+        assert redacted["api_key"] == "[REDACTED]"
+    
+    def test_redact_empty_dict(self):
+        """Test redacting empty dictionary."""
+        from praisonaiagents.trace.redact import redact_dict
+        
+        assert redact_dict({}) == {}
+    
+    def test_redact_none_values(self):
+        """Test handling None values."""
+        from praisonaiagents.trace.redact import redact_dict
+        
+        data = {"api_key": None, "name": "test"}
+        redacted = redact_dict(data)
+        
+        # None values for sensitive keys should still be redacted
+        assert redacted["api_key"] == "[REDACTED]"
+        assert redacted["name"] == "test"
+    
+    def test_redact_disabled(self):
+        """Test disabling redaction."""
+        from praisonaiagents.trace.redact import redact_dict
+        
+        data = {"api_key": "visible", "name": "test"}
+        result = redact_dict(data, enabled=False)
+        
+        assert result["api_key"] == "visible"
+        assert result["name"] == "test"
+
+
+class TestTraceSink:
+    """Tests for TraceSink protocol."""
+    
+    def test_noop_sink(self):
+        """Test NoOpSink does nothing."""
+        from praisonaiagents.trace.protocol import NoOpSink, ActionEvent
+        
+        sink = NoOpSink()
+        event = ActionEvent(event_type="agent_start", timestamp=time.time())
+        
+        # Should not raise
+        sink.emit(event)
+        sink.flush()
+        sink.close()
+    
+    def test_list_sink(self):
+        """Test ListSink collects events."""
+        from praisonaiagents.trace.protocol import ListSink, ActionEvent
+        
+        sink = ListSink()
+        
+        event1 = ActionEvent(event_type="agent_start", timestamp=1.0)
+        event2 = ActionEvent(event_type="tool_start", timestamp=2.0)
+        
+        sink.emit(event1)
+        sink.emit(event2)
+        
+        events = sink.get_events()
+        assert len(events) == 2
+        assert events[0].event_type == "agent_start"
+        assert events[1].event_type == "tool_start"
+    
+    def test_list_sink_clear(self):
+        """Test ListSink.clear() method."""
+        from praisonaiagents.trace.protocol import ListSink, ActionEvent
+        
+        sink = ListSink()
+        sink.emit(ActionEvent(event_type="test", timestamp=1.0))
+        
+        assert len(sink.get_events()) == 1
+        
+        sink.clear()
+        
+        assert len(sink.get_events()) == 0
+
+
+class TestTraceEmitter:
+    """Tests for TraceEmitter class."""
+    
+    def test_emitter_with_noop_sink(self):
+        """Test TraceEmitter with NoOpSink."""
+        from praisonaiagents.trace.protocol import TraceEmitter, NoOpSink
+        
+        emitter = TraceEmitter(sink=NoOpSink())
+        
+        # Should not raise
+        emitter.agent_start("test-agent", agent_id="123")
+        emitter.tool_start("web_search", {"query": "test"})
+        emitter.tool_end("web_search", duration_ms=100, status="ok")
+        emitter.agent_end("test-agent", duration_ms=500)
+    
+    def test_emitter_with_list_sink(self):
+        """Test TraceEmitter collects events."""
+        from praisonaiagents.trace.protocol import TraceEmitter, ListSink
+        
+        sink = ListSink()
+        emitter = TraceEmitter(sink=sink)
+        
+        emitter.agent_start("researcher")
+        emitter.tool_start("search", {"q": "AI"})
+        emitter.tool_end("search", duration_ms=100, status="ok")
+        emitter.agent_end("researcher", duration_ms=200)
+        
+        events = sink.get_events()
+        assert len(events) == 4
+        assert events[0].event_type == "agent_start"
+        assert events[1].event_type == "tool_start"
+        assert events[2].event_type == "tool_end"
+        assert events[3].event_type == "agent_end"
+    
+    def test_emitter_redacts_by_default(self):
+        """Test TraceEmitter redacts sensitive data by default."""
+        from praisonaiagents.trace.protocol import TraceEmitter, ListSink
+        
+        sink = ListSink()
+        emitter = TraceEmitter(sink=sink, redact=True)
+        
+        emitter.tool_start("api_call", {"api_key": "sk-secret", "query": "test"})
+        
+        events = sink.get_events()
+        assert events[0].tool_args["api_key"] == "[REDACTED]"
+        assert events[0].tool_args["query"] == "test"
+    
+    def test_emitter_disabled(self):
+        """Test TraceEmitter when disabled."""
+        from praisonaiagents.trace.protocol import TraceEmitter, ListSink
+        
+        sink = ListSink()
+        emitter = TraceEmitter(sink=sink, enabled=False)
+        
+        emitter.agent_start("test")
+        emitter.tool_start("tool", {})
+        
+        # No events should be emitted
+        assert len(sink.get_events()) == 0
+    
+    def test_emitter_error_event(self):
+        """Test TraceEmitter error event."""
+        from praisonaiagents.trace.protocol import TraceEmitter, ListSink
+        
+        sink = ListSink()
+        emitter = TraceEmitter(sink=sink)
+        
+        emitter.error("Something went wrong", tool_name="failing_tool")
+        
+        events = sink.get_events()
+        assert len(events) == 1
+        assert events[0].event_type == "error"
+        assert events[0].error_message == "Something went wrong"
+        assert events[0].tool_name == "failing_tool"
+    
+    def test_emitter_output_event(self):
+        """Test TraceEmitter output event for final agent output."""
+        from praisonaiagents.trace.protocol import TraceEmitter, ListSink
+        
+        sink = ListSink()
+        emitter = TraceEmitter(sink=sink)
+        
+        emitter.output("The analysis shows three major trends...")
+        
+        events = sink.get_events()
+        assert len(events) == 1
+        assert events[0].event_type == "output"
+        assert "three major trends" in events[0].tool_result_summary
+
+
+class TestActionTraceConfig:
+    """Tests for ActionTraceConfig."""
+    
+    def test_config_defaults(self):
+        """Test ActionTraceConfig default values."""
+        from praisonaiagents.trace.protocol import ActionTraceConfig
+        
+        config = ActionTraceConfig()
+        
+        assert config.enabled is True
+        assert config.redact is True
+        assert config.compact is False
+        assert config.sink_type == "noop"
+        assert config.file_path is None
+    
+    def test_config_custom(self):
+        """Test ActionTraceConfig custom values."""
+        from praisonaiagents.trace.protocol import ActionTraceConfig
+        
+        config = ActionTraceConfig(
+            enabled=True,
+            redact=False,
+            compact=True,
+            sink_type="jsonl",
+            file_path="/tmp/trace.jsonl",
+        )
+        
+        assert config.redact is False
+        assert config.compact is True
+        assert config.sink_type == "jsonl"
+        assert config.file_path == "/tmp/trace.jsonl"
+
+
+class TestPerformance:
+    """Performance tests for trace module."""
+    
+    def test_noop_sink_overhead(self):
+        """Test that NoOpSink has minimal overhead."""
+        from praisonaiagents.trace.protocol import TraceEmitter, NoOpSink, ActionEvent
+        
+        emitter = TraceEmitter(sink=NoOpSink())
+        
+        start = time.perf_counter()
+        for _ in range(10000):
+            emitter.tool_start("test", {"arg": "value"})
+            emitter.tool_end("test", duration_ms=1, status="ok")
+        elapsed = time.perf_counter() - start
+        
+        # Should complete 10000 iterations in under 100ms
+        assert elapsed < 0.1, f"NoOpSink too slow: {elapsed:.3f}s for 10000 iterations"
+    
+    def test_disabled_emitter_overhead(self):
+        """Test that disabled emitter has near-zero overhead."""
+        from praisonaiagents.trace.protocol import TraceEmitter, ListSink
+        
+        emitter = TraceEmitter(sink=ListSink(), enabled=False)
+        
+        start = time.perf_counter()
+        for _ in range(10000):
+            emitter.tool_start("test", {"arg": "value"})
+        elapsed = time.perf_counter() - start
+        
+        # Should complete 10000 iterations in under 500ms (very conservative)
+        assert elapsed < 0.5, f"Disabled emitter too slow: {elapsed:.3f}s"

src/praisonai-agents/uv.lock

@@ -2422,7 +2422,7 @@ wheels = [
 
 [[package]]
 name = "praisonaiagents"
-version = "0.11.1"
+version = "0.11.14"
 source = { editable = "." }
 dependencies = [
     { name = "aiohttp" },

src/praisonai/.gitignore

@@ -1,3 +1,4 @@
 chainlit.md
 tests/fixtures
-reports
+reports
+.agent

src/praisonai/examples/python/dynamic_context_example.py

@@ -0,0 +1,257 @@
+"""
+Dynamic Context Discovery Example.
+
+This example demonstrates how to use the Dynamic Context Discovery feature
+to efficiently manage large tool outputs, conversation history, and terminal logs.
+
+Features demonstrated:
+1. Tool output queuing - Large outputs automatically saved as artifacts
+2. Artifact tools - tail, grep, chunk operations on stored artifacts
+3. History persistence - Conversation history saved for loss recovery
+4. Terminal logging - Shell command outputs captured
+
+Usage:
+    python dynamic_context_example.py
+"""
+
+import tempfile
+
+
+def example_artifact_store():
+    """Demonstrate artifact storage and retrieval."""
+    print("\n=== Artifact Store Example ===\n")
+    
+    from praisonai.context import FileSystemArtifactStore
+    from praisonaiagents.context.artifacts import ArtifactMetadata
+    
+    # Create a temporary store
+    with tempfile.TemporaryDirectory() as tmpdir:
+        store = FileSystemArtifactStore(base_dir=tmpdir)
+        
+        # Store a large JSON response (simulating API output)
+        large_data = {
+            "records": [{"id": i, "name": f"Item {i}"} for i in range(1000)],
+            "metadata": {"total": 1000, "page": 1},
+        }
+        
+        metadata = ArtifactMetadata(
+            agent_id="data_agent",
+            run_id="run_001",
+            tool_name="api_fetch",
+        )
+        
+        ref = store.store(large_data, metadata)
+        print(f"Stored artifact: {ref.path}")
+        print(f"Size: {ref._format_size(ref.size_bytes)}")
+        print(f"Summary: {ref.summary}")
+        print(f"Inline reference: {ref.to_inline()}")
+        
+        # Demonstrate tail operation
+        print("\n--- Tail (last 5 lines) ---")
+        tail = store.tail(ref, lines=5)
+        print(tail)
+        
+        # Demonstrate grep operation
+        print("\n--- Grep for 'Item 50' ---")
+        matches = store.grep(ref, pattern="Item 50")
+        for match in matches[:3]:
+            print(f"Line {match.line_number}: {match.line_content[:60]}...")
+        
+        # List all artifacts
+        print("\n--- List artifacts ---")
+        artifacts = store.list_artifacts(run_id="run_001")
+        for art in artifacts:
+            print(f"  - {art.artifact_id}: {art.tool_name} ({art._format_size(art.size_bytes)})")
+
+
+def example_output_queue():
+    """Demonstrate automatic tool output queuing (previously called spooling)."""
+    print("\n=== Output Queue Example ===\n")
+    
+    from praisonai.context import OutputQueue
+    from praisonaiagents.context.artifacts import ArtifactMetadata, ArtifactRef, QueueConfig
+    
+    with tempfile.TemporaryDirectory() as tmpdir:
+        # Create output queue with 1KB threshold (for demo)
+        config = QueueConfig(inline_max_bytes=1024)
+        queue = OutputQueue(base_dir=tmpdir, config=config)
+        
+        # Small output - stays inline (not queued to artifact)
+        small_output = "Hello, world!"
+        metadata = ArtifactMetadata(agent_id="agent1", run_id="run1")
+        result = queue.process(small_output, metadata)
+        print(f"Small output ({len(small_output)} bytes): {'Queued' if isinstance(result, ArtifactRef) else 'Inline'}")
+        
+        # Large output - automatically queued to artifact
+        large_output = "x" * 2000
+        result = queue.process(large_output, metadata)
+        print(f"Large output ({len(large_output)} bytes): {'Queued' if isinstance(result, ArtifactRef) else 'Inline'}")
+        
+        if isinstance(result, ArtifactRef):
+            print(f"  Artifact path: {result.path}")
+
+
+def example_history_store():
+    """Demonstrate conversation history persistence."""
+    print("\n=== History Store Example ===\n")
+    
+    from praisonai.context import HistoryStore
+    
+    with tempfile.TemporaryDirectory() as tmpdir:
+        store = HistoryStore(base_dir=tmpdir)
+        
+        # Simulate a conversation
+        messages = [
+            {"role": "user", "content": "What is the capital of France?"},
+            {"role": "assistant", "content": "The capital of France is Paris."},
+            {"role": "user", "content": "What about Germany?"},
+            {"role": "assistant", "content": "The capital of Germany is Berlin."},
+            {"role": "user", "content": "And what about the error in my code?"},
+            {"role": "assistant", "content": "I see an error on line 42. The variable is undefined."},
+        ]
+        
+        for msg in messages:
+            store.append(msg, agent_id="assistant", run_id="session_001")
+        
+        # Get history reference
+        ref = store.get_ref(agent_id="assistant", run_id="session_001")
+        print(f"History stored: {ref.summary}")
+        
+        # Search history
+        print("\n--- Search for 'capital' ---")
+        matches = store.search("capital", agent_id="assistant", run_id="session_001")
+        for msg in matches:
+            print(f"  [{msg['role']}]: {msg['content'][:50]}...")
+        
+        # Get last messages
+        print("\n--- Last 2 messages ---")
+        recent = store.get_last_messages(agent_id="assistant", run_id="session_001", count=2)
+        for msg in recent:
+            print(f"  [{msg['role']}]: {msg['content'][:50]}...")
+
+
+def example_terminal_logger():
+    """Demonstrate terminal session logging."""
+    print("\n=== Terminal Logger Example ===\n")
+    
+    from praisonai.context import TerminalLogger
+    
+    with tempfile.TemporaryDirectory() as tmpdir:
+        logger = TerminalLogger(base_dir=tmpdir)
+        
+        # Log some commands
+        commands = [
+            ("ls -la", "total 100\ndrwxr-xr-x 5 user user 160 Jan 1 00:00 .\n-rw-r--r-- 1 user user 1234 Jan 1 00:00 file.txt", 0),
+            ("cat file.txt", "Hello, world!\nThis is a test file.", 0),
+            ("grep error log.txt", "Error: Something went wrong\nError: Another issue", 0),
+            ("invalid_command", "command not found: invalid_command", 127),
+        ]
+        
+        for cmd, output, exit_code in commands:
+            ref = logger.log_command(
+                command=cmd,
+                output=output,
+                exit_code=exit_code,
+                agent_id="shell_agent",
+                run_id="session_001",
+            )
+        
+        # Get session reference
+        ref = logger.get_session_ref(agent_id="shell_agent", run_id="session_001")
+        print(f"Terminal session: {ref.summary}")
+        
+        # Tail the session
+        print("\n--- Tail (last 10 lines) ---")
+        tail = logger.tail_session(agent_id="shell_agent", run_id="session_001", lines=10)
+        print(tail)
+        
+        # Search for errors
+        print("\n--- Grep for 'Error' ---")
+        matches = logger.grep_session("Error", agent_id="shell_agent", run_id="session_001")
+        for match in matches:
+            print(f"  Line {match.line_number}: {match.line_content}")
+        
+        # Get command list
+        print("\n--- Commands executed ---")
+        cmds = logger.get_commands(agent_id="shell_agent", run_id="session_001")
+        for cmd in cmds:
+            status = "✓" if cmd["exit_code"] == 0 else "✗"
+            print(f"  {status} $ {cmd['command']}")
+
+
+def example_setup_dynamic_context():
+    """Demonstrate the convenience setup function."""
+    print("\n=== Setup Dynamic Context Example ===\n")
+    
+    from praisonai.context import setup_dynamic_context
+    
+    with tempfile.TemporaryDirectory() as tmpdir:
+        # Set up dynamic context with all features
+        ctx = setup_dynamic_context(
+            base_dir=tmpdir,
+            inline_max_kb=16,  # 16KB threshold
+            redact_secrets=True,
+            history_enabled=True,
+            terminal_logging=True,
+        )
+        
+        print(f"Run ID: {ctx.run_id}")
+        print(f"Base directory: {ctx.config.base_dir}")
+        
+        # Get tools for agent
+        tools = ctx.get_tools()
+        print(f"\nAvailable tools ({len(tools)}):")
+        for tool in tools:
+            print(f"  - {tool.__name__}: {tool.__doc__.split(chr(10))[0] if tool.__doc__ else 'No description'}")
+        
+        # Get middleware
+        middleware = ctx.get_middleware()
+        print(f"\nMiddleware: {middleware.__name__}")
+
+
+def example_agent_integration():
+    """Show how to integrate with Agent (conceptual)."""
+    print("\n=== Agent Integration Example (Conceptual) ===\n")
+    
+    print("""
+# To integrate Dynamic Context Discovery with an Agent:
+
+from praisonaiagents import Agent
+from praisonai.context import setup_dynamic_context
+
+# Set up dynamic context
+ctx = setup_dynamic_context(
+    inline_max_kb=32,
+    history_enabled=True,
+)
+
+# Create agent with dynamic context tools and middleware
+agent = Agent(
+    name="DataAnalyst",
+    instructions="You are a data analyst. Use artifact tools to explore large outputs.",
+    tools=ctx.get_tools(),  # Adds artifact_tail, artifact_grep, etc.
+    hooks=[ctx.get_middleware()],  # Auto-queues large tool outputs
+)
+
+# Now when the agent uses tools that produce large outputs,
+# they will be automatically queued to artifacts.
+# The agent can then use artifact_tail, artifact_grep, etc.
+# to explore the data without loading it all into context.
+""")
+
+
+if __name__ == "__main__":
+    print("=" * 60)
+    print("Dynamic Context Discovery Examples")
+    print("=" * 60)
+    
+    example_artifact_store()
+    example_output_queue()
+    example_history_store()
+    example_terminal_logger()
+    example_setup_dynamic_context()
+    example_agent_integration()
+    
+    print("\n" + "=" * 60)
+    print("All examples completed successfully!")
+    print("=" * 60)

src/praisonai/examples/python/guardrails/guardrails-advanced.py

@@ -0,0 +1,65 @@
+"""
+Guardrails - Advanced Example
+
+Demonstrates advanced guardrails patterns including:
+- Multi-agent configuration
+- Custom callbacks/handlers
+- Error handling
+- All available options
+
+Expected Output:
+    Comprehensive demonstration of guardrails with multiple agents.
+"""
+from praisonaiagents import Agent, Task, PraisonAIAgents
+
+# ============================================================
+# Section 1: Custom Configuration
+# ============================================================
+
+# Define custom guardrails configuration
+# config = ...
+
+# ============================================================
+# Section 2: Multi-Agent Setup
+# ============================================================
+
+agent1 = Agent(
+    name="Agent1",
+    instructions="You are the first agent",
+    # guardrails=config,
+)
+
+agent2 = Agent(
+    name="Agent2",
+    instructions="You are the second agent",
+    # guardrails=config,
+)
+
+# ============================================================
+# Section 3: Task Definition
+# ============================================================
+
+task = Task(
+    description="Demonstrate guardrails with multiple agents",
+    agent=agent1,
+)
+
+# ============================================================
+# Section 4: Execution
+# ============================================================
+
+agents = PraisonAIAgents(agents=[agent1, agent2], tasks=[task])
+result = agents.start()
+print(result)
+
+# ============================================================
+# Section 5: All Options Reference
+# ============================================================
+"""
+Guardrails Options:
+──────────────────
+| Option          | Type     | Default | Description           |
+|-----------------|----------|---------|----------------------|
+| option1         | str      | None    | Description here     |
+| option2         | bool     | False   | Description here     |
+"""

src/praisonai/examples/python/guardrails/guardrails-basic.py

@@ -0,0 +1,19 @@
+"""
+Guardrails - Basic Example
+
+Demonstrates minimal usage of guardrails with a single agent.
+
+Expected Output:
+    Agent response demonstrating guardrails functionality.
+"""
+from praisonaiagents import Agent
+
+# Basic guardrails usage
+agent = Agent(
+    instructions="You are a helpful assistant",
+    # guardrails=...  # Configure guardrails here
+)
+
+# Run the agent
+result = agent.start("Hello! Demonstrate guardrails functionality.")
+print(result)

src/praisonai/praisonai.rb

@@ -3,8 +3,8 @@ class Praisonai < Formula
   
     desc "AI tools for various AI applications"
     homepage "https://github.com/MervinPraison/PraisonAI"
-    url "https://github.com/MervinPraison/PraisonAI/archive/refs/tags/v3.7.2.tar.gz"
-    sha256 `curl -sL https://github.com/MervinPraison/PraisonAI/archive/refs/tags/v3.7.2.tar.gz | shasum -a 256`.split.first
+    url "https://github.com/MervinPraison/PraisonAI/archive/refs/tags/v3.8.4.tar.gz"
+    sha256 `curl -sL https://github.com/MervinPraison/PraisonAI/archive/refs/tags/v3.8.4.tar.gz | shasum -a 256`.split.first
     license "MIT"
   
     depends_on "[email protected]"

src/praisonai/praisonai/cli/app.py

@@ -141,11 +141,18 @@ def main_callback(
     )
     set_output_controller(state.output_controller)
     
-    # If no command provided, show help or handle legacy behavior
+    # If no command provided, start interactive mode
     if ctx.invoked_subcommand is None:
-        # Check if there are remaining args (legacy behavior)
-        # This is handled by the legacy adapter in main.py
-        pass
+        from praisonai.cli.interactive.async_tui import AsyncTUI, AsyncTUIConfig
+        
+        tui_config = AsyncTUIConfig(
+            model="gpt-4o-mini",
+            show_logo=True,
+            show_status_bar=state.output_format != OutputFormat.json,
+        )
+        
+        tui = AsyncTUI(config=tui_config)
+        tui.run()
 
 
 def get_output_controller() -> OutputController:
@@ -266,6 +273,115 @@ def register_commands():
     app.add_typer(test_app, name="test", help="Run test suite with tier and provider options")
     app.add_typer(examples_app, name="examples", help="Run and manage example files")
     
+    # Register standardise command
+    try:
+        standardise_app = typer.Typer(name="standardise", help="Documentation and examples standardisation (FDEP)")
+        
+        @standardise_app.command("check")
+        def standardise_check(
+            path: str = typer.Option(".", "--path", "-p", help="Project root path"),
+            feature: str = typer.Option(None, "--feature", help="Specific feature slug"),
+            scope: str = typer.Option("all", "--scope", help="Scope: all, docs, examples, sdk, cli"),
+            ci: bool = typer.Option(False, "--ci", help="CI mode"),
+        ):
+            """Check for standardisation issues."""
+            from .commands.standardise import _run_check
+            import argparse
+            args = argparse.Namespace(path=path, feature=feature, scope=scope, ci=ci, dry_run=True)
+            _run_check(args)
+        
+        @standardise_app.command("report")
+        def standardise_report(
+            path: str = typer.Option(".", "--path", "-p", help="Project root path"),
+            format: str = typer.Option("text", "--format", "-f", help="Format: text, json, markdown"),
+            output: str = typer.Option(None, "--output", "-o", help="Output file"),
+            ci: bool = typer.Option(False, "--ci", help="CI mode"),
+        ):
+            """Generate detailed report."""
+            from .commands.standardise import _run_report
+            import argparse
+            args = argparse.Namespace(path=path, format=format, output=output, ci=ci, feature=None, scope="all", dry_run=True)
+            _run_report(args)
+        
+        @standardise_app.command("fix")
+        def standardise_fix(
+            path: str = typer.Option(".", "--path", "-p", help="Project root path"),
+            feature: str = typer.Option(None, "--feature", help="Specific feature slug"),
+            apply: bool = typer.Option(False, "--apply", help="Actually apply changes"),
+            no_backup: bool = typer.Option(False, "--no-backup", help="Don't create backups"),
+        ):
+            """Fix standardisation issues."""
+            from .commands.standardise import _run_fix
+            import argparse
+            args = argparse.Namespace(path=path, feature=feature, apply=apply, no_backup=no_backup, scope="all", ci=False, dry_run=not apply)
+            _run_fix(args)
+        
+        @standardise_app.command("init")
+        def standardise_init(
+            feature: str = typer.Argument(..., help="Feature slug to initialise"),
+            path: str = typer.Option(".", "--path", "-p", help="Project root path"),
+            apply: bool = typer.Option(False, "--apply", help="Actually create files"),
+        ):
+            """Initialise a new feature with all required artifacts."""
+            from .commands.standardise import _run_init
+            import argparse
+            args = argparse.Namespace(feature=feature, path=path, apply=apply, scope="all", ci=False, dry_run=not apply)
+            _run_init(args)
+        
+        @standardise_app.command("ai")
+        def standardise_ai(
+            feature: str = typer.Argument(..., help="Feature slug to generate content for"),
+            gen_type: str = typer.Option("all", "--type", "-t", help="Type: docs, examples, all"),
+            apply: bool = typer.Option(False, "--apply", help="Actually create files"),
+            verify: bool = typer.Option(False, "--verify", help="Verify with AI"),
+            model: str = typer.Option("gpt-4o-mini", "--model", help="LLM model"),
+            path: str = typer.Option(".", "--path", "-p", help="Project root path"),
+        ):
+            """AI-powered generation of docs/examples."""
+            from .commands.standardise import _run_ai
+            import argparse
+            args = argparse.Namespace(feature=feature, type=gen_type, apply=apply, verify=verify, model=model, path=path, scope="all", ci=False, dry_run=not apply)
+            _run_ai(args)
+        
+        @standardise_app.command("checkpoint")
+        def standardise_checkpoint(
+            message: str = typer.Option(None, "--message", "-m", help="Checkpoint message"),
+            path: str = typer.Option(".", "--path", "-p", help="Repository path"),
+        ):
+            """Create an undo checkpoint."""
+            from .commands.standardise import _run_checkpoint
+            import argparse
+            args = argparse.Namespace(message=message, path=path)
+            _run_checkpoint(args)
+        
+        @standardise_app.command("undo")
+        def standardise_undo(
+            checkpoint: str = typer.Option(None, "--checkpoint", help="Checkpoint ID"),
+            list_checkpoints: bool = typer.Option(False, "--list", help="List checkpoints"),
+            path: str = typer.Option(".", "--path", "-p", help="Repository path"),
+        ):
+            """Undo to a previous checkpoint."""
+            from .commands.standardise import _run_undo
+            import argparse
+            args = argparse.Namespace(checkpoint=checkpoint, list=list_checkpoints, path=path)
+            _run_undo(args)
+        
+        @standardise_app.command("redo")
+        def standardise_redo(
+            path: str = typer.Option(".", "--path", "-p", help="Repository path"),
+        ):
+            """Redo after an undo."""
+            from .commands.standardise import _run_redo
+            import argparse
+            args = argparse.Namespace(path=path)
+            _run_redo(args)
+        
+        app.add_typer(standardise_app, name="standardise", help="Documentation and examples standardisation (FDEP)")
+        # Also register as 'standardize' for US spelling
+        app.add_typer(standardise_app, name="standardize", help="Documentation and examples standardisation (FDEP)")
+    except Exception:
+        pass  # Graceful degradation if standardise module not available
+    
     # Register TUI and queue commands
     tui_app = create_tui_debug_app()
     queue_app = create_queue_app()

src/praisonai/praisonai/cli/branding.py

@@ -0,0 +1,94 @@
+"""
+PraisonAI Branding - Single Source of Truth.
+
+This module provides unified branding assets (logo, version, product name)
+for all interactive UIs. All frontends should import from here.
+
+Usage:
+    from praisonai.cli.branding import get_logo, get_version, PRODUCT_NAME
+"""
+
+# Product name
+PRODUCT_NAME = "Praison AI"
+
+# ASCII Art Logos - responsive to terminal width
+LOGO_LARGE = r"""
+ ██████╗ ██████╗  █████╗ ██╗███████╗ ██████╗ ███╗   ██╗     █████╗ ██╗
+ ██╔══██╗██╔══██╗██╔══██╗██║██╔════╝██╔═══██╗████╗  ██║    ██╔══██╗██║
+ ██████╔╝██████╔╝███████║██║███████╗██║   ██║██╔██╗ ██║    ███████║██║
+ ██╔═══╝ ██╔══██╗██╔══██║██║╚════██║██║   ██║██║╚██╗██║    ██╔══██║██║
+ ██║     ██║  ██║██║  ██║██║███████║╚██████╔╝██║ ╚████║    ██║  ██║██║
+ ╚═╝     ╚═╝  ╚═╝╚═╝  ╚═╝╚═╝╚══════╝ ╚═════╝ ╚═╝  ╚═══╝    ╚═╝  ╚═╝╚═╝
+"""
+
+LOGO_MEDIUM = r"""
+ ╔═╗┬─┐┌─┐┬┌─┐┌─┐┌┐┌  ╔═╗╦
+ ╠═╝├┬┘├─┤│└─┐│ ││││  ╠═╣║
+ ╩  ┴└─┴ ┴┴└─┘└─┘┘└┘  ╩ ╩╩
+"""
+
+LOGO_SMALL = "▶ Praison AI"
+
+LOGO_MINIMAL = "Praison AI"
+
+
+def get_version() -> str:
+    """Get PraisonAI version string."""
+    try:
+        from praisonai import __version__
+        return __version__
+    except Exception:
+        return "1.0.0"
+
+
+def get_logo(width: int = 80) -> str:
+    """
+    Get appropriate logo based on terminal width.
+    
+    Args:
+        width: Terminal width in characters
+        
+    Returns:
+        ASCII art logo string appropriate for the width
+    """
+    if width >= 75:
+        return LOGO_LARGE
+    elif width >= 40:
+        return LOGO_MEDIUM
+    else:
+        return LOGO_SMALL
+
+
+def get_banner(width: int = 80, show_version: bool = True, model: str = None) -> str:
+    """
+    Get full banner with logo, version, and optional model info.
+    
+    Args:
+        width: Terminal width
+        show_version: Whether to include version
+        model: Optional model name to display
+        
+    Returns:
+        Complete banner string
+    """
+    lines = []
+    
+    # Logo
+    logo = get_logo(width)
+    lines.append(logo.strip())
+    lines.append("")
+    
+    # Version and model
+    version_line = f"  v{get_version()}"
+    if model:
+        version_line += f" · Model: {model}"
+    lines.append(version_line)
+    lines.append("")
+    
+    return "\n".join(lines)
+
+
+def get_welcome_tips() -> str:
+    """Get welcome tips for new sessions."""
+    return """  Type your message and press Enter. Use /help for commands.
+  Use PageUp/PageDown or Ctrl+Up/Down to scroll."""

src/praisonai/praisonai/cli/commands/chat.py

@@ -67,6 +67,7 @@ def chat_main(
     workspace: Optional[str] = typer.Option(None, "--workspace", "-w", help="Workspace directory"),
     no_acp: bool = typer.Option(False, "--no-acp", help="Disable ACP tools"),
     no_lsp: bool = typer.Option(False, "--no-lsp", help="Disable LSP tools"),
+    safe_mode: bool = typer.Option(False, "--safe", help="Safe mode: require approval for file writes and commands"),
     autonomy: bool = typer.Option(True, "--autonomy/--no-autonomy", help="Enable agent autonomy for complex tasks"),
     # NEW: Agent-like consolidated params for ALL GREEN consistency
     knowledge: Optional[str] = typer.Option(
@@ -107,8 +108,8 @@ def chat_main(
         flag_value="true",
     ),
     output: Optional[str] = typer.Option(
-        None, "--output",
-        help="Output preset. Use --output=verbose, --output=minimal, --output=silent",
+        None, "--output", "-o",
+        help="Output mode (default: actions). Options: actions, plain, verbose, json, silent",
     ),
     execution: Optional[str] = typer.Option(
         None, "--execution",
@@ -126,8 +127,9 @@ def chat_main(
     ),
     profile: bool = typer.Option(False, "--profile", help="Enable CLI profiling (timing breakdown)"),
     profile_deep: bool = typer.Option(False, "--profile-deep", help="Enable deep profiling (cProfile stats, higher overhead)"),
+    debug: bool = typer.Option(False, "--debug", help="Enable debug logging to ~/.praisonai/async_tui_debug.log"),
     # UI backend selection
-    ui: str = typer.Option("auto", "--ui", help="UI backend: auto, plain, rich, mg (middle-ground)"),
+    ui_backend: str = typer.Option("auto", "--ui-backend", help="UI backend: auto, plain, rich, mg (middle-ground)"),
     json_output: bool = typer.Option(False, "--json", help="Output JSON (forces plain backend)"),
     no_color: bool = typer.Option(False, "--no-color", help="Disable colors"),
     theme: str = typer.Option("default", "--theme", help="UI theme: default, dark, light, minimal"),
@@ -149,25 +151,12 @@ def chat_main(
         praisonai chat "Summarize this" --file README.md
         praisonai chat "What is 2+2?" --profile
     """
-    import asyncio
     import os
     
     # Set workspace if provided
     if workspace:
         os.environ["PRAISONAI_WORKSPACE"] = workspace
     
-    # Configure UI backend
-    from praisonai.cli.ui import select_backend, UIConfig
-    ui_config = UIConfig(
-        ui_backend=ui,
-        json_output=json_output,
-        no_color=no_color,
-        theme=theme,
-        compact=compact,
-        verbose=verbose,
-    )
-    backend = select_backend(ui_config)
-    
     # Handle profiling for single prompt mode
     if prompt and (profile or profile_deep):
         _run_profiled_chat(
@@ -184,60 +173,38 @@ def chat_main(
         typer.echo("   Use: praisonai chat \"your prompt\" --profile", err=True)
     
     # Parse memory flag: --no-memory takes precedence, then --memory value
-    memory_value = _parse_memory_flag(memory, no_memory)
+    # TODO: Pass memory_value to TUI when memory support is added
+    _parse_memory_flag(memory, no_memory)
     
-    # Try InteractiveCore first (preferred terminal-native implementation)
-    try:
-        from praisonai.cli.interactive import InteractiveCore, InteractiveConfig
-        from praisonai.cli.interactive.frontends import RichFrontend
-        
-        config = InteractiveConfig(
-            model=model,
-            session_id=session_id,
-            continue_session=continue_session,
-            workspace=workspace or None,
-            verbose=verbose,
-            memory=memory_value,
-            files=list(file) if file else [],
-            autonomy=autonomy,
-        )
-        
-        core = InteractiveCore(config=config)
-        
-        if prompt:
-            # Single prompt mode - use backend for output
-            from praisonai.cli.ui.events import UIEventType
-            
-            async def run_prompt():
-                if continue_session:
-                    core.continue_session()
-                
-                backend.emit(UIEventType.MESSAGE_START, {})
-                response = await core.prompt(prompt)
-                backend.emit(UIEventType.MESSAGE_CHUNK, {'content': response})
-                backend.emit(UIEventType.MESSAGE_END, {})
-            
-            asyncio.run(run_prompt())
-        else:
-            # Interactive REPL mode - pass ui_config to frontend
-            frontend = RichFrontend(core=core, config=config, ui_config=ui_config)
-            asyncio.run(frontend.run())
-            
-    except ImportError:
-        # Fallback to legacy terminal-native interactive mode
-        _run_legacy_terminal_chat(
-            prompt=prompt,
-            model=model,
-            verbose=verbose,
-            memory=memory,
-            tools=tools,
-            user_id=user_id,
-            session_id=session_id,
-            continue_session=continue_session,
-            workspace=workspace,
-            no_acp=no_acp,
-            no_lsp=no_lsp,
-        )
+    # Set approval mode based on --safe flag
+    import os
+    if safe_mode:
+        os.environ["PRAISON_APPROVAL_MODE"] = "prompt"
+    else:
+        os.environ["PRAISON_APPROVAL_MODE"] = "auto"
+    
+    # Use the async TUI (non-blocking, scrollable output)
+    from praisonai.cli.interactive.async_tui import AsyncTUI, AsyncTUIConfig
+    
+    tui_config = AsyncTUIConfig(
+        model=model or "gpt-4o-mini",
+        show_logo=not compact,
+        show_status_bar=not compact,
+        session_id=session_id,
+        workspace=workspace,
+        debug=debug,
+    )
+    
+    tui = AsyncTUI(config=tui_config)
+    
+    if prompt:
+        # Single prompt mode - direct response, no streaming
+        response = tui.run_single(prompt)
+        if response:
+            print(response)
+    else:
+        # Interactive split-pane TUI mode
+        tui.run()
 
 
 def _run_profiled_chat(
@@ -275,7 +242,6 @@ def _run_profiled_chat(
         "name": "ChatAgent",
         "role": "Assistant",
         "goal": "Help the user",
-        "verbose": verbose,
     }
     if model:
         agent_config["llm"] = model

src/praisonai/praisonai/cli/commands/code.py

@@ -23,6 +23,7 @@ def code_main(
     file: Optional[List[str]] = typer.Option(None, "--file", "-f", help="Attach file(s) to context"),
     no_acp: bool = typer.Option(False, "--no-acp", help="Disable ACP tools (file operations)"),
     no_lsp: bool = typer.Option(False, "--no-lsp", help="Disable LSP tools (code intelligence)"),
+    safe_mode: bool = typer.Option(False, "--safe", help="Safe mode: require approval for file writes and commands"),
     session_id: Optional[str] = typer.Option(None, "--session", "-s", help="Session ID to resume"),
     continue_session: bool = typer.Option(False, "--continue", "-c", help="Continue last session"),
     autonomy: bool = typer.Option(True, "--autonomy/--no-autonomy", help="Enable agent autonomy for complex tasks"),
@@ -54,6 +55,12 @@ def code_main(
     # Enable code-specific environment
     os.environ["PRAISONAI_CODE_MODE"] = "true"
     
+    # Set approval mode based on --safe flag
+    if safe_mode:
+        os.environ["PRAISON_APPROVAL_MODE"] = "prompt"
+    else:
+        os.environ["PRAISON_APPROVAL_MODE"] = "auto"
+    
     # Handle profiling for single prompt mode
     if prompt and (profile or profile_deep):
         _run_profiled_code(

src/praisonai/praisonai/cli/commands/context.py

@@ -224,3 +224,191 @@ def context_export(
     except ImportError as e:
         typer.echo(f"Error: Context module not available: {e}", err=True)
         raise typer.Exit(1)
+
+
+# =============================================================================
+# Artifact Commands (Dynamic Context Discovery)
+# =============================================================================
+
+artifacts_app = typer.Typer(help="Artifact management for dynamic context discovery")
+app.add_typer(artifacts_app, name="artifacts")
+
+
+@artifacts_app.command("list")
+def artifacts_list(
+    run_id: str = typer.Option(None, "--run-id", "-r", help="Filter by run ID"),
+    agent_id: str = typer.Option(None, "--agent-id", "-a", help="Filter by agent ID"),
+    tool_name: str = typer.Option(None, "--tool", "-t", help="Filter by tool name"),
+    json_output: bool = typer.Option(False, "--json", "-j", help="Output as JSON"),
+):
+    """List artifacts."""
+    try:
+        from praisonai.context import FileSystemArtifactStore
+        import json as json_module
+        
+        store = FileSystemArtifactStore()
+        artifacts = store.list_artifacts(
+            run_id=run_id,
+            agent_id=agent_id,
+            tool_name=tool_name,
+        )
+        
+        if json_output:
+            data = [a.to_dict() for a in artifacts]
+            print(json_module.dumps(data, indent=2))
+        else:
+            if not artifacts:
+                typer.echo("No artifacts found.")
+                return
+            
+            from rich.console import Console
+            from rich.table import Table
+            
+            console = Console()
+            table = Table(title=f"Artifacts ({len(artifacts)} found)")
+            table.add_column("ID", style="cyan")
+            table.add_column("Tool", style="green")
+            table.add_column("Size", justify="right")
+            table.add_column("Summary")
+            table.add_column("Path", style="dim")
+            
+            for ref in artifacts:
+                size_str = ref._format_size(ref.size_bytes)
+                table.add_row(
+                    ref.artifact_id,
+                    ref.tool_name or "-",
+                    size_str,
+                    ref.summary[:40] + "..." if len(ref.summary) > 40 else ref.summary,
+                    ref.path,
+                )
+            
+            console.print(table)
+            
+    except ImportError as e:
+        typer.echo(f"Error: {e}", err=True)
+        raise typer.Exit(1)
+
+
+@artifacts_app.command("show")
+def artifacts_show(
+    artifact_path: str = typer.Argument(..., help="Path to artifact"),
+    lines: int = typer.Option(None, "--lines", "-n", help="Number of lines to show"),
+):
+    """Show artifact content."""
+    try:
+        from praisonai.context import FileSystemArtifactStore
+        from praisonaiagents.context.artifacts import ArtifactRef
+        
+        store = FileSystemArtifactStore()
+        ref = ArtifactRef(path=artifact_path, summary="", size_bytes=0)
+        
+        if lines:
+            content = store.head(ref, lines=lines)
+        else:
+            content = store.load(ref)
+            if not isinstance(content, str):
+                import json
+                content = json.dumps(content, indent=2)
+        
+        typer.echo(content)
+        
+    except FileNotFoundError:
+        typer.echo(f"Artifact not found: {artifact_path}", err=True)
+        raise typer.Exit(1)
+    except Exception as e:
+        typer.echo(f"Error: {e}", err=True)
+        raise typer.Exit(1)
+
+
+@artifacts_app.command("tail")
+def artifacts_tail(
+    artifact_path: str = typer.Argument(..., help="Path to artifact"),
+    lines: int = typer.Option(50, "--lines", "-n", help="Number of lines"),
+):
+    """Get last N lines of artifact."""
+    try:
+        from praisonai.context import FileSystemArtifactStore
+        from praisonaiagents.context.artifacts import ArtifactRef
+        
+        store = FileSystemArtifactStore()
+        ref = ArtifactRef(path=artifact_path, summary="", size_bytes=0)
+        content = store.tail(ref, lines=lines)
+        typer.echo(content)
+        
+    except FileNotFoundError:
+        typer.echo(f"Artifact not found: {artifact_path}", err=True)
+        raise typer.Exit(1)
+    except Exception as e:
+        typer.echo(f"Error: {e}", err=True)
+        raise typer.Exit(1)
+
+
+@artifacts_app.command("grep")
+def artifacts_grep(
+    artifact_path: str = typer.Argument(..., help="Path to artifact"),
+    pattern: str = typer.Argument(..., help="Search pattern (regex)"),
+    context_lines: int = typer.Option(2, "--context", "-C", help="Context lines"),
+    max_matches: int = typer.Option(50, "--max", "-m", help="Max matches"),
+):
+    """Search for pattern in artifact."""
+    try:
+        from praisonai.context import FileSystemArtifactStore
+        from praisonaiagents.context.artifacts import ArtifactRef
+        
+        store = FileSystemArtifactStore()
+        ref = ArtifactRef(path=artifact_path, summary="", size_bytes=0)
+        matches = store.grep(ref, pattern=pattern, context_lines=context_lines, max_matches=max_matches)
+        
+        if not matches:
+            typer.echo(f"No matches found for: {pattern}")
+            return
+        
+        from rich.console import Console
+        console = Console()
+        
+        console.print(f"[bold]Found {len(matches)} matches:[/bold]")
+        for match in matches:
+            console.print(f"\n[cyan]--- Line {match.line_number} ---[/cyan]")
+            for ctx in match.context_before:
+                console.print(f"  {ctx}", style="dim")
+            console.print(f"[green]> {match.line_content}[/green]")
+            for ctx in match.context_after:
+                console.print(f"  {ctx}", style="dim")
+        
+    except FileNotFoundError:
+        typer.echo(f"Artifact not found: {artifact_path}", err=True)
+        raise typer.Exit(1)
+    except ValueError as e:
+        typer.echo(f"Invalid pattern: {e}", err=True)
+        raise typer.Exit(1)
+    except Exception as e:
+        typer.echo(f"Error: {e}", err=True)
+        raise typer.Exit(1)
+
+
+@artifacts_app.command("export")
+def artifacts_export(
+    artifact_path: str = typer.Argument(..., help="Path to artifact"),
+    output: str = typer.Option(None, "--output", "-o", help="Output file path"),
+):
+    """Export artifact to file."""
+    try:
+        from pathlib import Path
+        import shutil
+        
+        src_path = Path(artifact_path)
+        if not src_path.exists():
+            typer.echo(f"Artifact not found: {artifact_path}", err=True)
+            raise typer.Exit(1)
+        
+        if output:
+            dst_path = Path(output)
+        else:
+            dst_path = Path.cwd() / src_path.name
+        
+        shutil.copy2(src_path, dst_path)
+        typer.echo(f"Exported to: {dst_path}")
+        
+    except Exception as e:
+        typer.echo(f"Error: {e}", err=True)
+        raise typer.Exit(1)

src/praisonai/praisonai/cli/commands/mcp.py

@@ -203,6 +203,164 @@ def mcp_test(
             raise typer.Exit(1)
 
 
+@app.command("sync")
+def mcp_sync(
+    server: Optional[str] = typer.Argument(None, help="Server name (None for all)"),
+):
+    """Sync tool schemas from MCP servers to local index."""
+    output = get_output_controller()
+    
+    try:
+        from praisonai.mcp_server.tool_index import MCPToolIndex
+        
+        index = MCPToolIndex()
+        loader = get_config_loader()
+        config = loader.load()
+        
+        servers_to_sync = []
+        if server:
+            if server not in config.mcp.servers:
+                output.print_error(f"Server not found: {server}")
+                raise typer.Exit(1)
+            servers_to_sync = [server]
+        else:
+            servers_to_sync = list(config.mcp.servers.keys())
+        
+        if not servers_to_sync:
+            output.print_info("No MCP servers configured")
+            return
+        
+        total_tools = 0
+        for srv in servers_to_sync:
+            output.print_info(f"Syncing {srv}...")
+            # Note: This would need actual MCP connection to fetch tools
+            # For now, we just create the index structure
+            count = index.sync(srv, tools=[])
+            total_tools += count
+            output.print(f"  Synced {count} tools")
+        
+        if output.is_json_mode:
+            output.print_json({"synced_servers": servers_to_sync, "total_tools": total_tools})
+        else:
+            output.print_success(f"Synced {len(servers_to_sync)} servers, {total_tools} tools total")
+            
+    except ImportError as e:
+        output.print_error(f"Error: {e}")
+        raise typer.Exit(1)
+
+
+@app.command("tools")
+def mcp_tools(
+    server: Optional[str] = typer.Argument(None, help="Server name (None for all)"),
+    json_output: bool = typer.Option(False, "--json", help="Output JSON"),
+):
+    """List indexed MCP tools."""
+    output = get_output_controller()
+    
+    try:
+        from praisonai.mcp_server.tool_index import MCPToolIndex
+        
+        index = MCPToolIndex()
+        
+        if server:
+            tools = index.list_tools(server)
+            if not tools:
+                output.print_info(f"No tools indexed for server: {server}")
+                output.print("\nRun 'praisonai mcp sync' to index tools")
+                return
+        else:
+            tools = index.get_all_tools()
+            if not tools:
+                output.print_info("No tools indexed")
+                output.print("\nRun 'praisonai mcp sync' to index tools")
+                return
+        
+        if output.is_json_mode or json_output:
+            output.print_json({"tools": [t.to_dict() for t in tools]})
+        else:
+            headers = ["Server", "Tool", "Description"]
+            rows = [[t.server, t.name, t.hint[:50] + "..." if len(t.hint) > 50 else t.hint] for t in tools]
+            output.print_table(headers, rows, title=f"MCP Tools ({len(tools)} total)")
+            
+    except ImportError as e:
+        output.print_error(f"Error: {e}")
+        raise typer.Exit(1)
+
+
+@app.command("describe")
+def mcp_describe(
+    server: str = typer.Argument(..., help="Server name"),
+    tool: str = typer.Argument(..., help="Tool name"),
+):
+    """Show full schema for an MCP tool."""
+    output = get_output_controller()
+    
+    try:
+        from praisonai.mcp_server.tool_index import MCPToolIndex
+        import json
+        
+        index = MCPToolIndex()
+        schema = index.describe(server, tool)
+        
+        if not schema:
+            output.print_error(f"Tool not found: {server}/{tool}")
+            raise typer.Exit(1)
+        
+        if output.is_json_mode:
+            output.print_json(schema)
+        else:
+            output.print(json.dumps(schema, indent=2))
+            
+    except ImportError as e:
+        output.print_error(f"Error: {e}")
+        raise typer.Exit(1)
+
+
+@app.command("status")
+def mcp_status(
+    server: Optional[str] = typer.Argument(None, help="Server name (None for all)"),
+    json_output: bool = typer.Option(False, "--json", help="Output JSON"),
+):
+    """Show status of MCP servers."""
+    output = get_output_controller()
+    
+    try:
+        from praisonai.mcp_server.tool_index import MCPToolIndex
+        from datetime import datetime
+        
+        index = MCPToolIndex()
+        
+        servers = [server] if server else index.list_servers()
+        
+        if not servers:
+            output.print_info("No servers indexed")
+            return
+        
+        statuses = []
+        for srv in servers:
+            status = index.get_status(srv)
+            if status:
+                statuses.append(status)
+        
+        if output.is_json_mode or json_output:
+            output.print_json({"statuses": [s.to_dict() for s in statuses]})
+        else:
+            headers = ["Server", "Status", "Tools", "Last Sync"]
+            rows = []
+            for s in statuses:
+                status_str = "✓ Available" if s.available else "✗ Unavailable"
+                if s.auth_required:
+                    status_str += " (auth required)"
+                last_sync = datetime.fromtimestamp(s.last_sync).strftime("%Y-%m-%d %H:%M") if s.last_sync else "Never"
+                rows.append([s.server, status_str, str(s.tool_count), last_sync])
+            
+            output.print_table(headers, rows, title="MCP Server Status")
+            
+    except ImportError as e:
+        output.print_error(f"Error: {e}")
+        raise typer.Exit(1)
+
+
 @app.callback(invoke_without_command=True)
 def mcp_callback(ctx: typer.Context):
     """Show MCP help or list servers."""

src/praisonai/praisonai/cli/commands/memory.py

@@ -107,3 +107,192 @@ def memory_clear(
         pass
     finally:
         sys.argv = original_argv
+
+
+# Learn subcommand group
+learn_app = typer.Typer(help="Continuous learning management")
+app.add_typer(learn_app, name="learn")
+
+
+@learn_app.command("status")
+def learn_status(
+    user_id: str = typer.Option("default", "--user-id", "-u", help="User ID"),
+    scope: str = typer.Option("private", "--scope", "-s", help="Scope (private/shared)"),
+):
+    """Show learning status and statistics."""
+    from praisonaiagents.memory.learn import LearnManager
+    from praisonaiagents.config.feature_configs import LearnConfig
+    
+    config = LearnConfig()
+    manager = LearnManager(config=config, user_id=user_id)
+    stats = manager.get_stats()
+    
+    typer.echo(f"\n📚 Learning Status for user: {user_id}")
+    typer.echo(f"   Scope: {scope}")
+    typer.echo("-" * 40)
+    
+    if not stats:
+        typer.echo("No learning stores enabled.")
+        return
+    
+    total = 0
+    for store_name, count in stats.items():
+        typer.echo(f"   {store_name}: {count} entries")
+        total += count
+    
+    typer.echo("-" * 40)
+    typer.echo(f"   Total: {total} entries")
+
+
+@learn_app.command("show")
+def learn_show(
+    store: str = typer.Argument("all", help="Store to show (persona/insights/threads/patterns/decisions/feedback/improvements/all)"),
+    user_id: str = typer.Option("default", "--user-id", "-u", help="User ID"),
+    limit: int = typer.Option(10, "--limit", "-l", help="Number of entries to show"),
+):
+    """Show learned entries."""
+    from praisonaiagents.memory.learn import LearnManager
+    from praisonaiagents.config.feature_configs import LearnConfig
+    
+    config = LearnConfig(
+        persona=True,
+        insights=True,
+        thread=True,
+        patterns=True,
+        decisions=True,
+        feedback=True,
+        improvements=True,
+    )
+    manager = LearnManager(config=config, user_id=user_id)
+    
+    if store == "all":
+        context = manager.get_learning_context(limit_per_store=limit)
+        for store_name, entries in context.items():
+            if entries:
+                typer.echo(f"\n📁 {store_name.upper()}")
+                typer.echo("-" * 40)
+                for entry in entries:
+                    typer.echo(f"  [{entry['id'][:20]}...] {entry['content'][:60]}...")
+    else:
+        context = manager.get_learning_context(limit_per_store=limit)
+        if store in context:
+            entries = context[store]
+            typer.echo(f"\n📁 {store.upper()}")
+            typer.echo("-" * 40)
+            for entry in entries:
+                typer.echo(f"  [{entry['id'][:20]}...] {entry['content']}")
+        else:
+            typer.echo(f"Store '{store}' not found or empty.")
+
+
+@learn_app.command("add")
+def learn_add(
+    content: str = typer.Argument(..., help="Content to learn"),
+    store: str = typer.Option("insights", "--store", "-s", help="Store type (persona/insights/patterns)"),
+    user_id: str = typer.Option("default", "--user-id", "-u", help="User ID"),
+):
+    """Add a learning entry."""
+    from praisonaiagents.memory.learn import LearnManager
+    from praisonaiagents.config.feature_configs import LearnConfig
+    
+    config = LearnConfig(
+        persona=(store == "persona"),
+        insights=(store == "insights"),
+        patterns=(store == "patterns"),
+        thread=(store == "threads"),
+        decisions=(store == "decisions"),
+        feedback=(store == "feedback"),
+        improvements=(store == "improvements"),
+    )
+    manager = LearnManager(config=config, user_id=user_id)
+    
+    entry = None
+    if store == "persona":
+        entry = manager.capture_persona(content)
+    elif store == "insights":
+        entry = manager.capture_insight(content)
+    elif store == "patterns":
+        entry = manager.capture_pattern(content)
+    elif store == "decisions":
+        entry = manager.capture_decision(content)
+    elif store == "feedback":
+        entry = manager.capture_feedback(content)
+    elif store == "improvements":
+        entry = manager.capture_improvement(content)
+    else:
+        typer.echo(f"Unknown store: {store}")
+        raise typer.Exit(1)
+    
+    if entry:
+        typer.echo(f"✅ Added to {store}: {entry.id}")
+    else:
+        typer.echo(f"❌ Failed to add to {store}")
+
+
+@learn_app.command("search")
+def learn_search(
+    query: str = typer.Argument(..., help="Search query"),
+    user_id: str = typer.Option("default", "--user-id", "-u", help="User ID"),
+    limit: int = typer.Option(10, "--limit", "-l", help="Max results per store"),
+):
+    """Search across all learning stores."""
+    from praisonaiagents.memory.learn import LearnManager
+    from praisonaiagents.config.feature_configs import LearnConfig
+    
+    config = LearnConfig(
+        persona=True,
+        insights=True,
+        thread=True,
+        patterns=True,
+        decisions=True,
+        feedback=True,
+        improvements=True,
+    )
+    manager = LearnManager(config=config, user_id=user_id)
+    
+    results = manager.search(query, limit=limit)
+    
+    if not results:
+        typer.echo(f"No results found for: {query}")
+        return
+    
+    typer.echo(f"\n🔍 Search results for: {query}")
+    for store_name, entries in results.items():
+        typer.echo(f"\n📁 {store_name.upper()}")
+        typer.echo("-" * 40)
+        for entry in entries:
+            typer.echo(f"  {entry['content'][:80]}...")
+
+
+@learn_app.command("clear")
+def learn_clear(
+    store: str = typer.Argument("all", help="Store to clear (or 'all')"),
+    user_id: str = typer.Option("default", "--user-id", "-u", help="User ID"),
+    force: bool = typer.Option(False, "--force", "-f", help="Skip confirmation"),
+):
+    """Clear learning entries."""
+    from praisonaiagents.memory.learn import LearnManager
+    from praisonaiagents.config.feature_configs import LearnConfig
+    
+    if not force:
+        confirm = typer.confirm(f"Clear {store} learning data for user {user_id}?")
+        if not confirm:
+            typer.echo("Cancelled.")
+            return
+    
+    config = LearnConfig(
+        persona=True,
+        insights=True,
+        thread=True,
+        patterns=True,
+        decisions=True,
+        feedback=True,
+        improvements=True,
+    )
+    manager = LearnManager(config=config, user_id=user_id)
+    
+    cleared = manager.clear_all()
+    
+    typer.echo("\n🗑️  Cleared learning data:")
+    for store_name, count in cleared.items():
+        typer.echo(f"   {store_name}: {count} entries")

src/praisonai/praisonai/cli/commands/run.py

@@ -22,13 +22,14 @@ def run_main(
     framework: Optional[str] = typer.Option(None, "--framework", "-f", help="Framework: praisonai, crewai, autogen"),
     interactive: bool = typer.Option(False, "--interactive", "-i", help="Interactive mode"),
     verbose: bool = typer.Option(False, "--verbose", "-v", help="Verbose output"),
-    stream: bool = typer.Option(True, "--stream/--no-stream", help="Stream output"),
+    stream: bool = typer.Option(False, "--stream/--no-stream", help="Stream output (default: off for production use)"),
     trace: bool = typer.Option(False, "--trace", help="Enable tracing"),
     memory: bool = typer.Option(False, "--memory", help="Enable memory"),
     tools: Optional[str] = typer.Option(None, "--tools", "-t", help="Tools file path"),
     max_tokens: int = typer.Option(16000, "--max-tokens", help="Maximum output tokens"),
     profile: bool = typer.Option(False, "--profile", help="Enable CLI profiling (timing breakdown)"),
     profile_deep: bool = typer.Option(False, "--profile-deep", help="Enable deep profiling (cProfile stats, higher overhead)"),
+    output_mode: Optional[str] = typer.Option(None, "--output", "-o", help="Output mode: silent (default), actions, verbose, json, stream"),
 ):
     """
     Run agents from a file or prompt.
@@ -108,6 +109,7 @@ def run_main(
             memory=memory,
             tools=tools,
             max_tokens=max_tokens,
+            output_mode=output_mode,
         )
     else:
         # Run as prompt
@@ -120,6 +122,7 @@ def run_main(
             memory=memory,
             tools=tools,
             max_tokens=max_tokens,
+            output_mode=output_mode,
         )
 
 
@@ -134,6 +137,7 @@ def _run_from_file(
     memory: bool = False,
     tools: Optional[str] = None,
     max_tokens: int = 16000,
+    output_mode: Optional[str] = None,
 ):
     """Run agents from a YAML file."""
     output = get_output_controller()
@@ -178,12 +182,37 @@ def _run_prompt(
     memory: bool = False,
     tools: Optional[str] = None,
     max_tokens: int = 16000,
+    output_mode: Optional[str] = None,
 ):
     """Run a direct prompt."""
     output = get_output_controller()
     
     try:
-        # Use existing handle_direct_prompt
+        # If output_mode is "actions", use direct Agent with actions preset
+        if output_mode == "actions":
+            from praisonaiagents import Agent
+            
+            agent_config = {
+                "name": "RunAgent",
+                "role": "Assistant", 
+                "goal": "Complete the task",
+                "output": "actions",  # Use actions preset
+            }
+            if model:
+                agent_config["llm"] = model
+            
+            agent = Agent(**agent_config)
+            result = agent.start(prompt)
+            
+            output.emit_result(
+                message="Prompt completed",
+                data={"result": str(result) if result else None}
+            )
+            
+            # Don't print result again - actions mode already shows output
+            return
+        
+        # Use existing handle_direct_prompt for other modes
         from praisonai.cli.main import PraisonAI
         
         praison = PraisonAI()

src/praisonai/praisonai/cli/commands/skills.py

@@ -97,3 +97,121 @@ def skills_install(
         pass
     finally:
         sys.argv = original_argv
+
+
+@app.command("search")
+def skills_search(
+    query: str = typer.Argument(..., help="Search query"),
+    skill_dirs: str = typer.Option(None, "--dirs", "-d", help="Comma-separated skill directories"),
+):
+    """Search skills by name, description, or content."""
+    try:
+        from praisonaiagents.skills import discover_skills
+        import re
+        
+        dirs = skill_dirs.split(",") if skill_dirs else None
+        skills = discover_skills(dirs, include_defaults=True)
+        
+        pattern = re.compile(query, re.IGNORECASE)
+        matches = []
+        
+        for skill in skills:
+            # Search in name and description
+            if pattern.search(skill.name) or pattern.search(skill.description):
+                matches.append(skill)
+                continue
+            
+            # Search in SKILL.md content if path available
+            if skill.path:
+                skill_md = skill.path / "SKILL.md"
+                if skill_md.exists():
+                    content = skill_md.read_text()
+                    if pattern.search(content):
+                        matches.append(skill)
+        
+        if not matches:
+            typer.echo(f"No skills found matching: {query}")
+            return
+        
+        from rich.console import Console
+        from rich.table import Table
+        
+        console = Console()
+        table = Table(title=f"Skills matching '{query}' ({len(matches)} found)")
+        table.add_column("Name", style="cyan")
+        table.add_column("Description")
+        table.add_column("Path", style="dim")
+        
+        for skill in matches:
+            table.add_row(
+                skill.name,
+                skill.description[:60] + "..." if len(skill.description) > 60 else skill.description,
+                str(skill.path) if skill.path else "-",
+            )
+        
+        console.print(table)
+        
+    except ImportError as e:
+        typer.echo(f"Error: {e}", err=True)
+        raise typer.Exit(1)
+
+
+@app.command("info")
+def skills_info(
+    name: str = typer.Argument(..., help="Skill name or path"),
+):
+    """Show detailed information about a skill."""
+    try:
+        from praisonaiagents.skills import discover_skills, SkillLoader
+        from pathlib import Path
+        
+        # Try as path first
+        path = Path(name).expanduser()
+        if path.exists() and path.is_dir():
+            loader = SkillLoader()
+            skill = loader.load_metadata(str(path))
+            if skill:
+                loader.activate(skill)
+        else:
+            # Search by name
+            skills = discover_skills(include_defaults=True)
+            skill = None
+            for s in skills:
+                if s.name == name:
+                    loader = SkillLoader()
+                    skill = loader.load_metadata(str(s.path))
+                    if skill:
+                        loader.activate(skill)
+                    break
+        
+        if not skill:
+            typer.echo(f"Skill not found: {name}", err=True)
+            raise typer.Exit(1)
+        
+        from rich.console import Console
+        from rich.panel import Panel
+        from rich.markdown import Markdown
+        
+        console = Console()
+        
+        # Show metadata
+        console.print(f"\n[bold cyan]{skill.properties.name}[/bold cyan]")
+        console.print(f"[dim]{skill.properties.description}[/dim]\n")
+        
+        if skill.properties.license:
+            console.print(f"License: {skill.properties.license}")
+        if skill.properties.compatibility:
+            console.print(f"Compatibility: {skill.properties.compatibility}")
+        if skill.properties.allowed_tools:
+            console.print(f"Allowed Tools: {skill.properties.allowed_tools}")
+        if skill.properties.metadata:
+            console.print(f"Metadata: {skill.properties.metadata}")
+        
+        # Show instructions if loaded
+        if skill.instructions:
+            console.print("\n[bold]Instructions:[/bold]")
+            console.print(Panel(Markdown(skill.instructions[:2000])))
+        
+    except ImportError as e:
+        typer.echo(f"Error: {e}", err=True)
+        raise typer.Exit(1)

src/praisonai/praisonai/cli/commands/standardise.py

@@ -2,10 +2,14 @@
 CLI commands for the FDEP standardisation system.
 
 Commands:
-- praisonai standardise check   - Check for standardisation issues
-- praisonai standardise report  - Generate detailed report
-- praisonai standardise fix     - Fix issues (with --apply)
-- praisonai standardise init    - Initialise a new feature
+- praisonai standardise check      - Check for standardisation issues
+- praisonai standardise report     - Generate detailed report
+- praisonai standardise fix        - Fix issues (with --apply)
+- praisonai standardise init       - Initialise a new feature
+- praisonai standardise ai         - AI-powered generation
+- praisonai standardise checkpoint - Create undo checkpoint
+- praisonai standardise undo       - Undo to previous checkpoint
+- praisonai standardise redo       - Redo after undo
 """
 
 import sys
@@ -88,6 +92,91 @@ def standardise_command(args=None):
         help="Actually create files (default: dry-run)",
     )
     
+    # ai subcommand
+    ai_parser = subparsers.add_parser(
+        "ai",
+        help="AI-powered generation of docs/examples",
+    )
+    ai_parser.add_argument(
+        "feature",
+        type=str,
+        help="Feature slug to generate content for",
+    )
+    ai_parser.add_argument(
+        "--type", "-t",
+        choices=["docs", "examples", "all"],
+        default="all",
+        help="Type of content to generate",
+    )
+    ai_parser.add_argument(
+        "--apply",
+        action="store_true",
+        help="Actually create files (default: dry-run preview)",
+    )
+    ai_parser.add_argument(
+        "--verify",
+        action="store_true",
+        help="Verify generated content with AI",
+    )
+    ai_parser.add_argument(
+        "--model",
+        type=str,
+        default="gpt-4o-mini",
+        help="LLM model to use (default: gpt-4o-mini)",
+    )
+    _add_common_args(ai_parser)
+    
+    # checkpoint subcommand
+    checkpoint_parser = subparsers.add_parser(
+        "checkpoint",
+        help="Create an undo checkpoint",
+    )
+    checkpoint_parser.add_argument(
+        "--message", "-m",
+        type=str,
+        help="Checkpoint message",
+    )
+    checkpoint_parser.add_argument(
+        "--path", "-p",
+        type=str,
+        default=".",
+        help="Repository path",
+    )
+    
+    # undo subcommand
+    undo_parser = subparsers.add_parser(
+        "undo",
+        help="Undo to a previous checkpoint",
+    )
+    undo_parser.add_argument(
+        "--checkpoint",
+        type=str,
+        help="Specific checkpoint ID to restore",
+    )
+    undo_parser.add_argument(
+        "--list",
+        action="store_true",
+        help="List available checkpoints",
+    )
+    undo_parser.add_argument(
+        "--path", "-p",
+        type=str,
+        default=".",
+        help="Repository path",
+    )
+    
+    # redo subcommand
+    redo_parser = subparsers.add_parser(
+        "redo",
+        help="Redo after an undo",
+    )
+    redo_parser.add_argument(
+        "--path", "-p",
+        type=str,
+        default=".",
+        help="Repository path",
+    )
+    
     # Parse args
     parsed = parser.parse_args(args)
     
@@ -105,6 +194,14 @@ def standardise_command(args=None):
         return _run_fix(parsed)
     elif parsed.subcommand == "init":
         return _run_init(parsed)
+    elif parsed.subcommand == "ai":
+        return _run_ai(parsed)
+    elif parsed.subcommand == "checkpoint":
+        return _run_checkpoint(parsed)
+    elif parsed.subcommand == "undo":
+        return _run_undo(parsed)
+    elif parsed.subcommand == "redo":
+        return _run_redo(parsed)
     
     return 0
 
@@ -258,6 +355,210 @@ def _run_init(parsed) -> int:
     return 0
 
 
+def _run_ai(parsed) -> int:
+    """Run the AI generation subcommand with ACP/LSP runtime support."""
+    import asyncio
+    from praisonai.standardise.ai_generator import AIGenerator
+    from praisonai.standardise.models import ArtifactType, FeatureSlug
+    
+    config = _create_config(parsed)
+    
+    slug = FeatureSlug.from_string(parsed.feature)
+    if not slug.is_valid:
+        print(f"Error: Invalid feature slug: {slug.validation_error}")
+        return 2
+    
+    # Start ACP/LSP runtime for context gathering
+    runtime = None
+    try:
+        from praisonai.cli.features.interactive_runtime import create_runtime
+        runtime = create_runtime(
+            workspace=str(config.sdk_root or config.project_root or "."),
+            lsp=True,
+            acp=False  # ACP not needed for generation
+        )
+        # Start runtime in background
+        asyncio.get_event_loop().run_until_complete(runtime.start())
+        if runtime.lsp_ready:
+            print("🔧 LSP server ready for code intelligence")
+    except Exception:
+        # Runtime is optional, continue without it
+        pass
+    
+    generator = AIGenerator(
+        model=parsed.model,
+        sdk_root=config.sdk_root,
+        docs_root=config.docs_root,
+        examples_root=config.examples_root,
+        use_fast_context=True,
+    )
+    
+    # Determine which artifacts to generate
+    artifact_types = []
+    gen_type = getattr(parsed, "type", "all")
+    
+    if gen_type in ["docs", "all"]:
+        artifact_types.extend([
+            ArtifactType.DOCS_CONCEPT,
+            ArtifactType.DOCS_FEATURE,
+            ArtifactType.DOCS_CLI,
+            ArtifactType.DOCS_SDK,
+        ])
+    
+    if gen_type in ["examples", "all"]:
+        artifact_types.extend([
+            ArtifactType.EXAMPLE_BASIC,
+            ArtifactType.EXAMPLE_ADVANCED,
+        ])
+    
+    print(f"🤖 AI Generation for '{slug.normalised}'")
+    print("=" * 50)
+    
+    results = []
+    for artifact_type in artifact_types:
+        print(f"\n📝 Generating {artifact_type.value}...")
+        
+        try:
+            # Generate with execution verification for examples
+            content, output_path, verification_info = generator.generate(
+                slug, artifact_type,
+                dry_run=not parsed.apply,
+                verify_execution=True,
+            )
+            
+            # Show verification results for examples
+            if verification_info:
+                if verification_info.get("execution_passed"):
+                    print("  ✅ Code verified: runs successfully")
+                elif verification_info.get("requires_external"):
+                    libs = ", ".join(verification_info.get("missing_libraries", []))
+                    print(f"  ⚠️  Requires external libraries: {libs}")
+                elif not verification_info.get("syntax_valid"):
+                    print(f"  ❌ Syntax error: {verification_info.get('error', 'Unknown')[:100]}")
+                else:
+                    print(f"  ❌ Execution failed after {verification_info.get('attempt', 1)} attempts")
+                    if verification_info.get("error"):
+                        print(f"     Error: {verification_info['error'][:150]}")
+            
+            # AI content verification (optional)
+            if parsed.verify:
+                print("  🔍 AI content review...")
+                is_valid, summary, issues = generator.verify(content, artifact_type, slug)
+                if is_valid:
+                    print(f"  ✅ {summary}")
+                else:
+                    print(f"  ⚠️  {summary}")
+                    for issue in issues[:3]:
+                        print(f"     - {issue}")
+            
+            # Check if we should write
+            can_write = True
+            if verification_info and not verification_info.get("success"):
+                if not verification_info.get("requires_external"):
+                    can_write = False
+                    print("  ⛔ Not writing: code doesn't run")
+            
+            if parsed.apply and output_path and can_write:
+                print(f"  ✓ Created: {output_path}")
+                results.append((artifact_type.value, True, output_path))
+            elif not parsed.apply:
+                print(f"  📋 Preview ({len(content)} chars)")
+                # Show first few lines
+                preview_lines = content.split("\n")[:10]
+                for line in preview_lines:
+                    print(f"     {line[:60]}")
+                if len(content.split("\n")) > 10:
+                    print("     ...")
+                results.append((artifact_type.value, True, None))
+            else:
+                results.append((artifact_type.value, False, "Verification failed"))
+            
+        except Exception as e:
+            print(f"  ❌ Error: {e}")
+            results.append((artifact_type.value, False, str(e)))
+    
+    print("\n" + "=" * 50)
+    print("Summary:")
+    success_count = sum(1 for _, success, _ in results if success)
+    print(f"  Generated: {success_count}/{len(results)}")
+    
+    if not parsed.apply:
+        print("\nRun with --apply to create these files.")
+    
+    # Cleanup runtime
+    if runtime:
+        try:
+            asyncio.get_event_loop().run_until_complete(runtime.stop())
+        except Exception:
+            pass
+    
+    return 0
+
+
+def _run_checkpoint(parsed) -> int:
+    """Run the checkpoint subcommand."""
+    from praisonai.standardise.undo_redo import UndoRedoManager
+    
+    manager = UndoRedoManager(Path(parsed.path).resolve())
+    
+    success, result = manager.create_checkpoint(parsed.message)
+    
+    if success:
+        print(f"✓ Checkpoint created: {result}")
+    else:
+        print(f"❌ Failed: {result}")
+        return 1
+    
+    return 0
+
+
+def _run_undo(parsed) -> int:
+    """Run the undo subcommand."""
+    from praisonai.standardise.undo_redo import UndoRedoManager
+    
+    manager = UndoRedoManager(Path(parsed.path).resolve())
+    
+    # List checkpoints if requested
+    if getattr(parsed, "list", False):
+        checkpoints = manager.list_checkpoints()
+        if not checkpoints:
+            print("No checkpoints found.")
+            return 0
+        
+        print("Available checkpoints:")
+        for cp_id, date, message in checkpoints:
+            print(f"  • {cp_id} ({date}): {message[:50]}")
+        return 0
+    
+    # Undo to checkpoint
+    success, result = manager.undo(parsed.checkpoint)
+    
+    if success:
+        print(f"✓ {result}")
+    else:
+        print(f"❌ {result}")
+        return 1
+    
+    return 0
+
+
+def _run_redo(parsed) -> int:
+    """Run the redo subcommand."""
+    from praisonai.standardise.undo_redo import UndoRedoManager
+    
+    manager = UndoRedoManager(Path(parsed.path).resolve())
+    
+    success, result = manager.redo()
+    
+    if success:
+        print(f"✓ {result}")
+    else:
+        print(f"❌ {result}")
+        return 1
+    
+    return 0
+
+
 # Click-based alternative for integration with main CLI
 def register_click_commands(app):
     """Register click commands with the main CLI app."""

src/praisonai/praisonai/cli/commands/test.py

@@ -10,6 +10,12 @@ Usage:
     praisonai test --provider openai    # Run OpenAI provider tests
     praisonai test --live               # Enable live API tests
     praisonai test --parallel auto      # Run with parallelization
+    
+    # Interactive mode tests
+    praisonai test interactive --csv tests.csv
+    praisonai test interactive --suite smoke
+    praisonai test interactive --list
+    praisonai test interactive --generate-template
 """
 
 import os
@@ -239,6 +245,272 @@ def run(
         raise typer.Exit(1)
 
 
+# Built-in interactive test suites
+BUILTIN_SUITES = {
+    "smoke": "smoke.csv",
+    "tools": "tools.csv",
+    "refactor": "refactor.csv",
+    "multi_agent": "multi_agent.csv",
+}
+
+# Special suites that don't use CSV
+SPECIAL_SUITES = {
+    "advanced": "Advanced Agent-Centric (5 complex autonomous scenarios)",
+}
+
+
+def _get_builtin_suite_path(suite: str):
+    """Get path to built-in test suite."""
+    from pathlib import Path
+    
+    if suite not in BUILTIN_SUITES:
+        return None
+    
+    # Find fixtures directory relative to package
+    package_dir = Path(__file__).parent.parent.parent.parent  # Up to src/praisonai
+    fixtures_dir = package_dir / "tests" / "fixtures" / "interactive"
+    
+    suite_path = fixtures_dir / BUILTIN_SUITES[suite]
+    if suite_path.exists():
+        return suite_path
+    
+    return None
+
+
+@app.command("interactive")
+def interactive(
+    csv: Optional[str] = typer.Option(
+        None, "--csv", "-c",
+        help="Path to CSV file with test cases"
+    ),
+    suite: Optional[str] = typer.Option(
+        None, "--suite", "-s",
+        help="Built-in test suite: smoke, tools, refactor, multi_agent"
+    ),
+    model: str = typer.Option(
+        "gpt-4o-mini", "--model", "-m",
+        help="LLM model for agent"
+    ),
+    judge_model: str = typer.Option(
+        "gpt-4o-mini", "--judge-model",
+        help="LLM model for judge"
+    ),
+    workspace: Optional[str] = typer.Option(
+        None, "--workspace", "-w",
+        help="Workspace directory (default: temp)"
+    ),
+    artifacts_dir: Optional[str] = typer.Option(
+        None, "--artifacts-dir",
+        help="Directory for test artifacts"
+    ),
+    fail_fast: bool = typer.Option(
+        False, "--fail-fast", "-x",
+        help="Stop on first failure"
+    ),
+    keep_artifacts: bool = typer.Option(
+        False, "--keep-artifacts",
+        help="Keep test artifacts after run"
+    ),
+    no_judge: bool = typer.Option(
+        False, "--no-judge",
+        help="Skip judge evaluation even if rubric present"
+    ),
+    verbose: bool = typer.Option(
+        False, "--verbose", "-v",
+        help="Verbose output"
+    ),
+    list_suites: bool = typer.Option(
+        False, "--list",
+        help="List available built-in suites"
+    ),
+    generate_template: bool = typer.Option(
+        False, "--generate-template",
+        help="Generate CSV template in current directory"
+    ),
+):
+    """
+    Run interactive mode tests from CSV.
+    
+    Tests are executed through the headless interactive core, using the same
+    pipeline as the interactive TUI (InteractiveRuntime, get_interactive_tools,
+    ActionOrchestrator).
+    
+    Examples:
+        praisonai test interactive --csv tests.csv
+        praisonai test interactive --suite smoke
+        praisonai test interactive --suite tools --keep-artifacts
+        praisonai test interactive --list
+        praisonai test interactive --generate-template
+    """
+    from pathlib import Path
+    
+    # List suites
+    if list_suites:
+        typer.echo("📋 Available built-in test suites:")
+        typer.echo("")
+        typer.echo("CSV-based suites:")
+        for name, filename in BUILTIN_SUITES.items():
+            suite_path = _get_builtin_suite_path(name)
+            status = "✅" if suite_path else "❌ (not found)"
+            typer.echo(f"  {name:15} {filename:20} {status}")
+        typer.echo("")
+        typer.echo("Special suites:")
+        for name, description in SPECIAL_SUITES.items():
+            typer.echo(f"  {name:15} {description}")
+        typer.echo("")
+        typer.echo("Usage: praisonai test interactive --suite <name>")
+        typer.echo("       PRAISONAI_LIVE_INTERACTIVE=1 praisonai test interactive --suite advanced")
+        return
+    
+    # Generate template
+    if generate_template:
+        # Lazy import
+        from praisonai.cli.features.csv_test_runner import generate_csv_template
+        
+        output_path = Path.cwd() / "interactive_tests_template.csv"
+        generate_csv_template(output_path)
+        typer.echo(f"✅ Generated CSV template: {output_path}")
+        return
+    
+    # Validate inputs
+    if not csv and not suite:
+        typer.echo("❌ Either --csv or --suite is required")
+        typer.echo("")
+        typer.echo("Examples:")
+        typer.echo("  praisonai test interactive --csv tests.csv")
+        typer.echo("  praisonai test interactive --suite smoke")
+        typer.echo("  praisonai test interactive --suite advanced")
+        typer.echo("  praisonai test interactive --list")
+        raise typer.Exit(1)
+    
+    # Handle special suites
+    if suite and suite in SPECIAL_SUITES:
+        if suite == "advanced":
+            _run_advanced_suite(
+                model=model,
+                artifacts_dir=Path(artifacts_dir) if artifacts_dir else None,
+                keep_artifacts=keep_artifacts,
+                verbose=verbose,
+            )
+            return
+    
+    # Get CSV path
+    if csv:
+        csv_path = Path(csv)
+        if not csv_path.exists():
+            typer.echo(f"❌ CSV file not found: {csv}")
+            raise typer.Exit(1)
+    else:
+        csv_path = _get_builtin_suite_path(suite)
+        if not csv_path:
+            typer.echo(f"❌ Built-in suite not found: {suite}")
+            typer.echo("Use --list to see available suites")
+            raise typer.Exit(1)
+    
+    # Lazy import runner
+    from praisonai.cli.features.csv_test_runner import CSVTestRunner
+    
+    typer.echo(f"🧪 Running interactive tests from: {csv_path}")
+    typer.echo(f"   Model: {model} | Judge: {judge_model}")
+    typer.echo("")
+    
+    # Create runner
+    runner = CSVTestRunner(
+        csv_path=csv_path,
+        model=model,
+        judge_model=judge_model,
+        workspace=Path(workspace) if workspace else None,
+        artifacts_dir=Path(artifacts_dir) if artifacts_dir else None,
+        fail_fast=fail_fast,
+        keep_artifacts=keep_artifacts,
+        no_judge=no_judge,
+        verbose=verbose,
+    )
+    
+    # Run tests
+    summary = runner.run()
+    
+    # Print summary
+    typer.echo("")
+    summary.print_summary()
+    
+    # Save summary if keeping artifacts
+    if keep_artifacts and runner.artifacts_dir:
+        import json
+        summary_path = runner.artifacts_dir / "summary.json"
+        summary_path.parent.mkdir(parents=True, exist_ok=True)
+        with open(summary_path, "w") as f:
+            json.dump(summary.to_dict(), f, indent=2, default=str)
+        typer.echo(f"\n📁 Artifacts saved to: {runner.artifacts_dir}")
+    
+    # Exit code based on results
+    if summary.failed > 0 or summary.errors > 0:
+        raise typer.Exit(1)
+
+
+def _run_advanced_suite(
+    model: str = "gpt-4o-mini",
+    artifacts_dir = None,
+    keep_artifacts: bool = True,
+    verbose: bool = True,
+):
+    """Run the Advanced Agent-Centric test suite."""
+    from pathlib import Path
+    
+    # Check prerequisites
+    typer.echo("🔍 Checking Advanced suite prerequisites...")
+    
+    # Check if live interactive is enabled
+    live_enabled = os.environ.get("PRAISONAI_LIVE_INTERACTIVE", "0") == "1"
+    has_api_key = bool(os.environ.get("OPENAI_API_KEY"))
+    
+    if not live_enabled:
+        typer.echo("\n⚠️  Cannot run Advanced tests: PRAISONAI_LIVE_INTERACTIVE=1 not set")
+        typer.echo("")
+        typer.echo("Prerequisites:")
+        typer.echo("  1. Set PRAISONAI_LIVE_INTERACTIVE=1")
+        typer.echo("  2. Set OPENAI_API_KEY")
+        typer.echo("")
+        typer.echo("Example:")
+        typer.echo("  PRAISONAI_LIVE_INTERACTIVE=1 praisonai test interactive --suite advanced")
+        raise typer.Exit(1)
+    
+    if not has_api_key:
+        typer.echo("\n⚠️  Cannot run Advanced tests: OPENAI_API_KEY not set")
+        raise typer.Exit(1)
+    
+    typer.echo("✅ Live interactive mode enabled")
+    typer.echo("✅ API key configured")
+    typer.echo("")
+    
+    # Lazy import runner and scenarios
+    from tests.live.interactive.runner import LiveInteractiveRunner
+    from tests.live.interactive.advanced.scenarios import ALL_ADVANCED_SCENARIOS
+    
+    # Create runner
+    runner = LiveInteractiveRunner(
+        model=model,
+        artifacts_dir=Path(artifacts_dir) if artifacts_dir else None,
+        keep_artifacts=keep_artifacts,
+        verbose=verbose,
+    )
+    
+    # Run tests
+    summary = runner.run_all(ALL_ADVANCED_SCENARIOS)
+    
+    # Print summary
+    typer.echo("")
+    summary.print_summary()
+    
+    # Show artifacts location
+    if keep_artifacts:
+        typer.echo(f"\n📁 Artifacts saved to: {runner.artifacts_dir}")
+    
+    # Exit code
+    if summary.failed > 0:
+        raise typer.Exit(1)
+
+
 @app.command()
 def info():
     """Show test configuration and available options."""

src/praisonai/praisonai/cli/features/action_orchestrator.py

@@ -160,6 +160,8 @@ class ActionOrchestrator:
     
     async def _create_fallback_plan(self, prompt: str, plan_id: str) -> ActionResult:
         """Create a fallback plan when ACP is not available for planning."""
+        import re  # Import at function level for all pattern matching
+        
         # This is a simplified fallback - in production, this would use LLM
         steps = []
         
@@ -168,7 +170,6 @@ class ActionOrchestrator:
         
         if "create" in prompt_lower or "new file" in prompt_lower:
             # Extract file path if mentioned
-            import re
             file_match = re.search(r'(?:create|new)\s+(?:a\s+)?(?:file\s+)?(?:called\s+)?([^\s]+\.\w+)', prompt_lower)
             target = file_match.group(1) if file_match else "new_file.py"
             
@@ -188,7 +189,8 @@ class ActionOrchestrator:
             ))
         
         elif "edit" in prompt_lower or "modify" in prompt_lower or "change" in prompt_lower:
-            file_match = re.search(r'(?:edit|modify|change)\s+([^\s]+\.\w+)', prompt_lower)
+            # Match patterns like "edit file X", "edit X", "modify X"
+            file_match = re.search(r'(?:edit|modify|change)\s+(?:file\s+)?([^\s]+(?:\.\w+)?)', prompt_lower)
             target = file_match.group(1) if file_match else ""
             
             if target:
@@ -197,11 +199,13 @@ class ActionOrchestrator:
                     action_type=ActionType.FILE_EDIT,
                     description=f"Edit file: {target}",
                     target=target,
-                    params={"changes": prompt}
+                    params={"changes": prompt},
+                    status=ActionStatus.APPROVED  # Auto-approve for fallback
                 ))
         
         elif "delete" in prompt_lower or "remove" in prompt_lower:
-            file_match = re.search(r'(?:delete|remove)\s+([^\s]+\.\w+)', prompt_lower)
+            # Match patterns like "delete file X", "delete X", "remove X"
+            file_match = re.search(r'(?:delete|remove)\s+(?:file\s+)?([^\s]+(?:\.\w+)?)', prompt_lower)
             target = file_match.group(1) if file_match else ""
             
             if target:
@@ -209,7 +213,8 @@ class ActionOrchestrator:
                     id=self._generate_step_id(plan_id, 0),
                     action_type=ActionType.FILE_DELETE,
                     description=f"Delete file: {target}",
-                    target=target
+                    target=target,
+                    status=ActionStatus.APPROVED  # Auto-approve for fallback
                 ))
         
         elif "rename" in prompt_lower:
@@ -218,19 +223,22 @@ class ActionOrchestrator:
                 action_type=ActionType.REFACTOR,
                 description="Rename operation",
                 target="",
-                params={"prompt": prompt}
+                params={"prompt": prompt},
+                status=ActionStatus.APPROVED  # Auto-approve for fallback
             ))
         
-        elif "run" in prompt_lower or "execute" in prompt_lower:
-            cmd_match = re.search(r'(?:run|execute)\s+[`"\']?([^`"\']+)[`"\']?', prompt_lower)
-            command = cmd_match.group(1) if cmd_match else ""
+        elif "run" in prompt_lower or "execute" in prompt_lower or "command" in prompt_lower:
+            # Match patterns like "run command: X", "execute X", "run X"
+            cmd_match = re.search(r'(?:run|execute)\s+(?:command[:\s]+)?[`"\']?(.+?)[`"\']?$', prompt_lower)
+            command = cmd_match.group(1).strip() if cmd_match else ""
             
             if command:
                 steps.append(ActionStep(
                     id=self._generate_step_id(plan_id, 0),
                     action_type=ActionType.SHELL_COMMAND,
                     description=f"Execute: {command}",
-                    target=command
+                    target=command,
+                    status=ActionStatus.APPROVED  # Auto-approve for fallback
                 ))
         
         if not steps:

src/praisonai/praisonai/cli/features/code_intelligence.py

@@ -596,20 +596,34 @@ class CodeIntelligenceRouter:
     
     def _extract_symbol_from_query(self, query: str) -> Optional[str]:
         """Extract symbol name from query."""
-        # Common patterns for symbol extraction
+        # Common patterns for symbol extraction (order matters - more specific first)
         patterns = [
-            r'(?:definition\s+of|where\s+is|find|show)\s+["\']?(\w+)["\']?',
+            # "find all references to X" - capture X after "to"
+            r'references?\s+to\s+["\']?(\w+)["\']?',
+            # "definition of X" - capture X after "of"
+            r'definition\s+of\s+["\']?(\w+)["\']?',
+            # "where is X defined/used"
+            r'where\s+is\s+["\']?(\w+)["\']?\s+(?:defined|used)',
+            # Quoted or backticked symbols
             r'["\'](\w+)["\']',
             r'`(\w+)`',
+            # "X is defined/used"
             r'(\w+)\s+(?:is\s+)?(?:defined|used)',
+            # Generic "find/show X" - but filter common words
+            r'(?:find|show)\s+["\']?(\w+)["\']?',
         ]
         
+        # Words to filter out
+        filter_words = {'the', 'a', 'an', 'is', 'are', 'in', 'of', 'to', 'for', 
+                       'all', 'any', 'some', 'this', 'that', 'where', 'what',
+                       'definition', 'references', 'reference', 'symbol', 'symbols'}
+        
         for pattern in patterns:
             match = re.search(pattern, query, re.IGNORECASE)
             if match:
                 symbol = match.group(1)
                 # Filter out common words
-                if symbol.lower() not in ['the', 'a', 'an', 'is', 'are', 'in', 'of', 'to', 'for']:
+                if symbol.lower() not in filter_words:
                     return symbol
         
         return None

src/praisonai/praisonai/cli/features/csv_test_runner.py

@@ -0,0 +1,736 @@
+"""
+CSV-driven Interactive Test Runner for PraisonAI.
+
+Provides a CSV-driven test runner that:
+- Loads test cases from CSV with defined schema
+- Runs each test in isolated temp workspace
+- Executes prompts via headless interactive core
+- Validates tool calls, files, and responses
+- Integrates LLM-as-judge for response quality
+- Emits artifacts per-test
+"""
+
+import csv
+import json
+import logging
+import os
+import time
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Any, Dict, List, Optional
+
+logger = logging.getLogger(__name__)
+
+
+# CSV Schema definition
+CSV_SCHEMA = {
+    # Required columns
+    "id": {"required": True, "type": str, "description": "Unique test identifier"},
+    "name": {"required": True, "type": str, "description": "Test name"},
+    "prompts": {"required": True, "type": str, "description": "Single prompt or JSON array"},
+    # Optional columns
+    "description": {"required": False, "type": str, "default": ""},
+    "mode": {"required": False, "type": str, "default": "headless"},
+    "workspace_fixture": {"required": False, "type": str, "default": "empty"},
+    "expected_tools": {"required": False, "type": str, "default": ""},
+    "forbidden_tools": {"required": False, "type": str, "default": ""},
+    "expected_files": {"required": False, "type": str, "default": "{}"},
+    "expected_response": {"required": False, "type": str, "default": ""},
+    "judge_rubric": {"required": False, "type": str, "default": ""},
+    "judge_threshold": {"required": False, "type": float, "default": 7.0},
+    "judge_model": {"required": False, "type": str, "default": "gpt-4o-mini"},
+    "timeout": {"required": False, "type": int, "default": 60},
+    "retries": {"required": False, "type": int, "default": 0},
+    "skip_if": {"required": False, "type": str, "default": ""},
+    "agents": {"required": False, "type": str, "default": "[]"},
+    "workflow": {"required": False, "type": str, "default": "{}"},
+}
+
+
+@dataclass
+class TestCase:
+    """A single test case from CSV."""
+    id: str
+    name: str
+    prompts: List[str]
+    description: str = ""
+    mode: str = "headless"
+    workspace_fixture: str = "empty"
+    expected_tools: List[str] = field(default_factory=list)
+    forbidden_tools: List[str] = field(default_factory=list)
+    expected_files: Dict[str, str] = field(default_factory=dict)
+    expected_response: str = ""
+    judge_rubric: str = ""
+    judge_threshold: float = 7.0
+    judge_model: str = "gpt-4o-mini"
+    timeout: int = 60
+    retries: int = 0
+    skip_if: str = ""
+    agents: List[Dict[str, Any]] = field(default_factory=list)
+    workflow: Dict[str, Any] = field(default_factory=dict)
+    
+    def should_skip(self) -> Optional[str]:
+        """Check if test should be skipped."""
+        if not self.skip_if:
+            return None
+        
+        conditions = [c.strip() for c in self.skip_if.split(",")]
+        
+        for condition in conditions:
+            if condition == "no_openai_key":
+                if not os.environ.get("OPENAI_API_KEY"):
+                    return "OPENAI_API_KEY not set"
+            elif condition == "no_anthropic_key":
+                if not os.environ.get("ANTHROPIC_API_KEY"):
+                    return "ANTHROPIC_API_KEY not set"
+            elif condition == "no_lsp":
+                # Could check if LSP is available
+                pass
+            elif condition == "no_network":
+                if os.environ.get("PRAISONAI_NO_NETWORK"):
+                    return "Network disabled"
+        
+        return None
+
+
+@dataclass
+class TestResult:
+    """Result of a single test."""
+    test_id: str
+    test_name: str
+    status: str  # passed, failed, skipped, error
+    duration: float
+    tool_calls: List[str] = field(default_factory=list)
+    response: str = ""
+    judge_score: Optional[float] = None
+    judge_passed: Optional[bool] = None
+    judge_reasoning: str = ""
+    artifacts_path: Optional[str] = None
+    error: Optional[str] = None
+    skip_reason: Optional[str] = None
+    assertions: Dict[str, bool] = field(default_factory=dict)
+    
+    def to_dict(self) -> Dict[str, Any]:
+        return {
+            "test_id": self.test_id,
+            "test_name": self.test_name,
+            "status": self.status,
+            "duration": self.duration,
+            "tool_calls": self.tool_calls,
+            "response": self.response[:500] if self.response else "",
+            "judge_score": self.judge_score,
+            "judge_passed": self.judge_passed,
+            "judge_reasoning": self.judge_reasoning,
+            "artifacts_path": self.artifacts_path,
+            "error": self.error,
+            "skip_reason": self.skip_reason,
+            "assertions": self.assertions,
+        }
+
+
+@dataclass
+class TestSummary:
+    """Summary of all test results."""
+    total: int = 0
+    passed: int = 0
+    failed: int = 0
+    skipped: int = 0
+    errors: int = 0
+    duration: float = 0.0
+    judge_avg: Optional[float] = None
+    results: List[TestResult] = field(default_factory=list)
+    
+    def to_dict(self) -> Dict[str, Any]:
+        return {
+            "total": self.total,
+            "passed": self.passed,
+            "failed": self.failed,
+            "skipped": self.skipped,
+            "errors": self.errors,
+            "duration": self.duration,
+            "judge_avg": self.judge_avg,
+            "pass_rate": self.passed / self.total if self.total > 0 else 0,
+            "results": [r.to_dict() for r in self.results],
+        }
+    
+    def print_summary(self) -> None:
+        """Print summary to console."""
+        try:
+            from rich.console import Console
+            from rich.table import Table
+            
+            console = Console()
+            
+            table = Table(title="Interactive Test Results")
+            table.add_column("ID", style="cyan")
+            table.add_column("Name", style="white")
+            table.add_column("Status")
+            table.add_column("Duration", justify="right")
+            table.add_column("Judge", justify="right")
+            
+            for r in self.results:
+                status_style = {
+                    "passed": "[green]✓ passed[/green]",
+                    "failed": "[red]✗ failed[/red]",
+                    "skipped": "[yellow]○ skipped[/yellow]",
+                    "error": "[red]! error[/red]",
+                }.get(r.status, r.status)
+                
+                judge_str = f"{r.judge_score:.1f}" if r.judge_score is not None else "-"
+                
+                table.add_row(
+                    r.test_id,
+                    r.test_name[:30],
+                    status_style,
+                    f"{r.duration:.2f}s",
+                    judge_str,
+                )
+            
+            console.print(table)
+            console.print()
+            console.print(f"[bold]Total:[/bold] {self.total} | "
+                         f"[green]Passed:[/green] {self.passed} | "
+                         f"[red]Failed:[/red] {self.failed} | "
+                         f"[yellow]Skipped:[/yellow] {self.skipped}")
+            if self.judge_avg is not None:
+                console.print(f"[bold]Average Judge Score:[/bold] {self.judge_avg:.2f}")
+            console.print(f"[bold]Total Duration:[/bold] {self.duration:.2f}s")
+            
+        except ImportError:
+            # Fallback to plain print
+            print("\n=== Interactive Test Results ===")
+            for r in self.results:
+                status = {"passed": "✓", "failed": "✗", "skipped": "○", "error": "!"}.get(r.status, "?")
+                print(f"  {status} {r.test_id}: {r.test_name} ({r.duration:.2f}s)")
+            print(f"\nTotal: {self.total} | Passed: {self.passed} | Failed: {self.failed} | Skipped: {self.skipped}")
+
+
+def parse_csv(csv_path: Path) -> List[TestCase]:
+    """
+    Parse CSV file into TestCase objects.
+    
+    Args:
+        csv_path: Path to CSV file
+        
+    Returns:
+        List of TestCase objects
+    """
+    test_cases = []
+    
+    with open(csv_path, "r", encoding="utf-8") as f:
+        reader = csv.DictReader(f)
+        
+        for row_num, row in enumerate(reader, start=2):  # Start at 2 (header is 1)
+            try:
+                # Parse prompts (single string or JSON array)
+                prompts_raw = row.get("prompts", "").strip()
+                if prompts_raw.startswith("["):
+                    prompts = json.loads(prompts_raw)
+                else:
+                    prompts = [prompts_raw] if prompts_raw else []
+                
+                # Parse expected_tools
+                expected_tools = []
+                if row.get("expected_tools"):
+                    expected_tools = [t.strip() for t in row["expected_tools"].split(",") if t.strip()]
+                
+                # Parse forbidden_tools
+                forbidden_tools = []
+                if row.get("forbidden_tools"):
+                    forbidden_tools = [t.strip() for t in row["forbidden_tools"].split(",") if t.strip()]
+                
+                # Parse expected_files JSON
+                expected_files = {}
+                if row.get("expected_files"):
+                    try:
+                        expected_files = json.loads(row["expected_files"])
+                    except json.JSONDecodeError:
+                        logger.warning(f"Row {row_num}: Invalid JSON in expected_files")
+                
+                # Parse agents JSON
+                agents = []
+                if row.get("agents"):
+                    try:
+                        agents = json.loads(row["agents"])
+                    except json.JSONDecodeError:
+                        logger.warning(f"Row {row_num}: Invalid JSON in agents")
+                
+                # Parse workflow JSON
+                workflow = {}
+                if row.get("workflow"):
+                    try:
+                        workflow = json.loads(row["workflow"])
+                    except json.JSONDecodeError:
+                        logger.warning(f"Row {row_num}: Invalid JSON in workflow")
+                
+                # Parse numeric fields
+                judge_threshold = float(row.get("judge_threshold", 7.0) or 7.0)
+                timeout = int(row.get("timeout", 60) or 60)
+                retries = int(row.get("retries", 0) or 0)
+                
+                test_case = TestCase(
+                    id=row.get("id", f"test_{row_num}"),
+                    name=row.get("name", f"Test {row_num}"),
+                    prompts=prompts,
+                    description=row.get("description", ""),
+                    mode=row.get("mode", "headless"),
+                    workspace_fixture=row.get("workspace_fixture", "empty"),
+                    expected_tools=expected_tools,
+                    forbidden_tools=forbidden_tools,
+                    expected_files=expected_files,
+                    expected_response=row.get("expected_response", ""),
+                    judge_rubric=row.get("judge_rubric", ""),
+                    judge_threshold=judge_threshold,
+                    judge_model=row.get("judge_model", "gpt-4o-mini"),
+                    timeout=timeout,
+                    retries=retries,
+                    skip_if=row.get("skip_if", ""),
+                    agents=agents,
+                    workflow=workflow,
+                )
+                
+                test_cases.append(test_case)
+                
+            except Exception as e:
+                logger.error(f"Error parsing row {row_num}: {e}")
+                continue
+    
+    return test_cases
+
+
+class CSVTestRunner:
+    """
+    CSV-driven test runner for interactive mode.
+    
+    Usage:
+        runner = CSVTestRunner(csv_path="tests.csv")
+        summary = runner.run()
+        summary.print_summary()
+    """
+    
+    def __init__(
+        self,
+        csv_path: Path,
+        model: str = "gpt-4o-mini",
+        judge_model: str = "gpt-4o-mini",
+        workspace: Optional[Path] = None,
+        artifacts_dir: Optional[Path] = None,
+        fail_fast: bool = False,
+        keep_artifacts: bool = False,
+        no_judge: bool = False,
+        verbose: bool = False,
+    ):
+        """
+        Initialize CSV test runner.
+        
+        Args:
+            csv_path: Path to CSV file with test cases
+            model: LLM model for agent
+            judge_model: LLM model for judge
+            workspace: Base workspace directory
+            artifacts_dir: Directory for artifacts
+            fail_fast: Stop on first failure
+            keep_artifacts: Keep artifacts after run
+            no_judge: Skip judge even if rubric present
+            verbose: Verbose output
+        """
+        self.csv_path = Path(csv_path)
+        self.model = model
+        self.judge_model = judge_model
+        self.workspace = workspace
+        self.artifacts_dir = artifacts_dir
+        self.fail_fast = fail_fast
+        self.keep_artifacts = keep_artifacts
+        self.no_judge = no_judge
+        self.verbose = verbose
+        self.results: List[TestResult] = []
+    
+    def run(self) -> TestSummary:
+        """
+        Run all tests from CSV.
+        
+        Returns:
+            TestSummary with all results
+        """
+        start_time = time.time()
+        
+        # Parse CSV
+        test_cases = parse_csv(self.csv_path)
+        logger.info(f"Loaded {len(test_cases)} test cases from {self.csv_path}")
+        
+        if self.verbose:
+            print(f"Running {len(test_cases)} tests from {self.csv_path}")
+        
+        # Run each test
+        for i, test_case in enumerate(test_cases):
+            if self.verbose:
+                print(f"\n[{i+1}/{len(test_cases)}] Running: {test_case.name}")
+            
+            # Check skip conditions
+            skip_reason = test_case.should_skip()
+            if skip_reason:
+                result = TestResult(
+                    test_id=test_case.id,
+                    test_name=test_case.name,
+                    status="skipped",
+                    duration=0.0,
+                    skip_reason=skip_reason,
+                )
+                self.results.append(result)
+                if self.verbose:
+                    print(f"  Skipped: {skip_reason}")
+                continue
+            
+            # Run test
+            result = self._run_single_test(test_case)
+            self.results.append(result)
+            
+            if self.verbose:
+                status_icon = {"passed": "✓", "failed": "✗", "error": "!"}.get(result.status, "?")
+                print(f"  {status_icon} {result.status} ({result.duration:.2f}s)")
+                if result.error:
+                    print(f"    Error: {result.error}")
+            
+            # Fail fast
+            if self.fail_fast and result.status in ("failed", "error"):
+                logger.info(f"Stopping due to fail-fast: {test_case.id}")
+                break
+        
+        # Create summary
+        summary = self._create_summary(time.time() - start_time)
+        
+        return summary
+    
+    def _run_single_test(self, test_case: TestCase) -> TestResult:
+        """Run a single test case."""
+        import tempfile
+        
+        start_time = time.time()
+        
+        # Create workspace
+        if self.workspace:
+            test_workspace = self.workspace / test_case.id
+        else:
+            test_workspace = Path(tempfile.mkdtemp(prefix=f"praison_test_{test_case.id}_"))
+        
+        # Create artifacts dir
+        if self.artifacts_dir:
+            test_artifacts = self.artifacts_dir / test_case.id
+        else:
+            test_artifacts = test_workspace / "artifacts"
+        
+        try:
+            # Import harness (lazy)
+            from .interactive_test_harness import InteractiveTestHarness
+            
+            # Create harness
+            harness = InteractiveTestHarness(
+                workspace=test_workspace,
+                artifacts_dir=test_artifacts,
+                keep_workspace=self.keep_artifacts,
+            )
+            
+            # Setup workspace
+            harness.setup_workspace(test_case.workspace_fixture)
+            
+            # Run prompts
+            exec_result = harness.run(
+                prompts=test_case.prompts,
+                model=self.model,
+                approval_mode="auto",
+                agents=test_case.agents if test_case.agents else None,
+                workflow=test_case.workflow if test_case.workflow else None,
+            )
+            
+            duration = time.time() - start_time
+            
+            # Get response
+            response = "\n".join(exec_result.responses) if exec_result.responses else ""
+            
+            # Verify assertions
+            assertions = {}
+            
+            # Tool call assertions
+            tool_result = harness.verify_tool_calls(
+                expected_tools=test_case.expected_tools,
+                forbidden_tools=test_case.forbidden_tools,
+            )
+            assertions["tools"] = tool_result["passed"]
+            
+            # File assertions
+            if test_case.expected_files:
+                file_results = harness.verify_files(test_case.expected_files)
+                assertions["files"] = all(file_results.values())
+            else:
+                assertions["files"] = True
+            
+            # Response assertion
+            if test_case.expected_response:
+                assertions["response"] = harness.verify_response(response, test_case.expected_response)
+            else:
+                assertions["response"] = True
+            
+            # Judge evaluation
+            judge_score = None
+            judge_passed = None
+            judge_reasoning = ""
+            
+            if test_case.judge_rubric and not self.no_judge:
+                judge_result = self._run_judge(
+                    response=response,
+                    rubric=test_case.judge_rubric,
+                    threshold=test_case.judge_threshold,
+                    model=test_case.judge_model or self.judge_model,
+                )
+                judge_score = judge_result.get("score")
+                judge_passed = judge_result.get("passed")
+                judge_reasoning = judge_result.get("reasoning", "")
+                assertions["judge"] = judge_passed if judge_passed is not None else True
+                harness.artifacts.judge_result = judge_result
+            
+            # Determine overall status
+            all_passed = all(assertions.values()) and exec_result.success
+            status = "passed" if all_passed else "failed"
+            
+            # Snapshot workspace if keeping artifacts
+            if self.keep_artifacts:
+                harness.snapshot_workspace()
+            
+            # Save artifacts
+            artifacts_path = str(harness.save_artifacts(test_case.id))
+            
+            return TestResult(
+                test_id=test_case.id,
+                test_name=test_case.name,
+                status=status,
+                duration=duration,
+                tool_calls=harness._executor.get_tools_called() if harness._executor else [],
+                response=response,
+                judge_score=judge_score,
+                judge_passed=judge_passed,
+                judge_reasoning=judge_reasoning,
+                artifacts_path=artifacts_path,
+                assertions=assertions,
+            )
+            
+        except Exception as e:
+            logger.error(f"Test {test_case.id} error: {e}", exc_info=True)
+            return TestResult(
+                test_id=test_case.id,
+                test_name=test_case.name,
+                status="error",
+                duration=time.time() - start_time,
+                error=str(e),
+            )
+        finally:
+            if not self.keep_artifacts:
+                try:
+                    import shutil
+                    if test_workspace.exists() and not self.workspace:
+                        shutil.rmtree(test_workspace)
+                except Exception:
+                    pass
+    
+    def _run_judge(
+        self,
+        response: str,
+        rubric: str,
+        threshold: float,
+        model: str,
+    ) -> Dict[str, Any]:
+        """
+        Run LLM judge on response using Agent class.
+        
+        Args:
+            response: Response to evaluate
+            rubric: Evaluation rubric
+            threshold: Pass threshold (1-10)
+            model: Judge model
+            
+        Returns:
+            Dict with score, passed, reasoning
+        """
+        try:
+            from praisonaiagents import Agent
+            
+            # Create judge agent
+            judge_agent = Agent(
+                name="Judge",
+                role="Evaluator",
+                goal="Evaluate response quality based on rubric",
+                instructions=f"""You are an expert evaluator. Your task is to evaluate a response based on the given rubric.
+
+RUBRIC:
+{rubric}
+
+RESPONSE TO EVALUATE:
+{response}
+
+Provide your evaluation in the following format:
+SCORE: [1-10]
+REASONING: [Your detailed reasoning]
+
+Be strict but fair. A score of 7+ means the response adequately meets the rubric criteria.""",
+                llm=model,
+                verbose=False,
+            )
+            
+            # Get evaluation
+            eval_response = judge_agent.chat("Please evaluate the response based on the rubric.")
+            
+            # Parse score from response
+            score = self._parse_judge_score(eval_response)
+            passed = score >= threshold if score is not None else None
+            
+            return {
+                "score": score,
+                "passed": passed,
+                "reasoning": eval_response,
+                "threshold": threshold,
+                "model": model,
+            }
+            
+        except Exception as e:
+            logger.warning(f"Judge evaluation failed: {e}")
+            return {
+                "score": None,
+                "passed": None,
+                "reasoning": f"Judge error: {e}",
+                "error": str(e),
+            }
+    
+    def _parse_judge_score(self, response: str) -> Optional[float]:
+        """Parse score from judge response."""
+        import re
+        
+        # Try to find "SCORE: X" pattern
+        match = re.search(r"SCORE:\s*(\d+(?:\.\d+)?)", response, re.IGNORECASE)
+        if match:
+            try:
+                return float(match.group(1))
+            except ValueError:
+                pass
+        
+        # Try to find standalone number at start
+        match = re.search(r"^(\d+(?:\.\d+)?)", response.strip())
+        if match:
+            try:
+                score = float(match.group(1))
+                if 1 <= score <= 10:
+                    return score
+            except ValueError:
+                pass
+        
+        return None
+    
+    def _create_summary(self, total_duration: float) -> TestSummary:
+        """Create test summary."""
+        summary = TestSummary(
+            total=len(self.results),
+            passed=sum(1 for r in self.results if r.status == "passed"),
+            failed=sum(1 for r in self.results if r.status == "failed"),
+            skipped=sum(1 for r in self.results if r.status == "skipped"),
+            errors=sum(1 for r in self.results if r.status == "error"),
+            duration=total_duration,
+            results=self.results,
+        )
+        
+        # Calculate average judge score
+        judge_scores = [r.judge_score for r in self.results if r.judge_score is not None]
+        if judge_scores:
+            summary.judge_avg = sum(judge_scores) / len(judge_scores)
+        
+        return summary
+
+
+def generate_csv_template(output_path: Path = None) -> str:
+    """
+    Generate a CSV template with all columns.
+    
+    Args:
+        output_path: Path to write template (optional)
+        
+    Returns:
+        CSV template string
+    """
+    headers = list(CSV_SCHEMA.keys())
+    
+    # Example rows
+    examples = [
+        {
+            "id": "smoke_01",
+            "name": "Basic Chat",
+            "prompts": "Hello, what is 2+2?",
+            "description": "Test basic chat response",
+            "mode": "headless",
+            "workspace_fixture": "empty",
+            "expected_tools": "",
+            "forbidden_tools": "",
+            "expected_files": "{}",
+            "expected_response": "4",
+            "judge_rubric": "Response contains the number 4",
+            "judge_threshold": "7.0",
+            "judge_model": "gpt-4o-mini",
+            "timeout": "60",
+            "retries": "0",
+            "skip_if": "",
+            "agents": "[]",
+            "workflow": "{}",
+        },
+        {
+            "id": "tools_01",
+            "name": "Create File",
+            "prompts": "Create a file called hello.py with print('hello')",
+            "description": "Test file creation",
+            "mode": "headless",
+            "workspace_fixture": "empty",
+            "expected_tools": "acp_create_file",
+            "forbidden_tools": "",
+            "expected_files": '{"hello.py": "print.*hello"}',
+            "expected_response": "",
+            "judge_rubric": "File was created successfully",
+            "judge_threshold": "7.0",
+            "judge_model": "gpt-4o-mini",
+            "timeout": "60",
+            "retries": "0",
+            "skip_if": "no_openai_key",
+            "agents": "[]",
+            "workflow": "{}",
+        },
+        {
+            "id": "multi_01",
+            "name": "Multi-step Edit",
+            "prompts": '["Create test.py with x=1", "Edit test.py to change x=1 to x=2"]',
+            "description": "Test multi-step editing",
+            "mode": "headless",
+            "workspace_fixture": "empty",
+            "expected_tools": "acp_create_file,acp_edit_file",
+            "forbidden_tools": "",
+            "expected_files": '{"test.py": "x.*=.*2"}',
+            "expected_response": "",
+            "judge_rubric": "",
+            "judge_threshold": "7.0",
+            "judge_model": "gpt-4o-mini",
+            "timeout": "120",
+            "retries": "0",
+            "skip_if": "no_openai_key",
+            "agents": "[]",
+            "workflow": "{}",
+        },
+    ]
+    
+    # Build CSV string
+    import io
+    output = io.StringIO()
+    writer = csv.DictWriter(output, fieldnames=headers)
+    writer.writeheader()
+    for example in examples:
+        writer.writerow(example)
+    
+    csv_content = output.getvalue()
+    
+    if output_path:
+        Path(output_path).write_text(csv_content)
+        logger.info(f"Generated CSV template at {output_path}")
+    
+    return csv_content

src/praisonai/praisonai/cli/features/interactive_core_headless.py

@@ -0,0 +1,450 @@
+"""
+Headless Interactive Core Executor for PraisonAI.
+
+This module provides a headless execution mode that uses the SAME interactive core
+pipeline as the interactive TUI:
+- InteractiveRuntime config
+- get_interactive_tools() tool resolution
+- ActionOrchestrator behavior (plan/approve/apply/verify)
+- Tool call tracing
+
+This is for tests and CI - no terminal/prompt_toolkit dependency.
+"""
+
+import functools
+import logging
+import os
+import time
+from dataclasses import dataclass, field
+from datetime import datetime
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass
+class ToolCallTrace:
+    """Record of a single tool call."""
+    tool_name: str
+    args: Tuple
+    kwargs: Dict[str, Any]
+    result: Any
+    success: bool
+    error: Optional[str]
+    duration: float
+    timestamp: str
+    
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert to dictionary for JSON serialization."""
+        return {
+            "tool_name": self.tool_name,
+            "args": _redact_sensitive(self.args),
+            "kwargs": _redact_sensitive(self.kwargs),
+            "result": _truncate(str(self.result), 500) if self.result else None,
+            "success": self.success,
+            "error": self.error,
+            "duration": self.duration,
+            "timestamp": self.timestamp,
+        }
+
+
+@dataclass
+class HeadlessExecutionResult:
+    """Result of a headless execution."""
+    success: bool
+    responses: List[str]
+    tool_trace: List[ToolCallTrace]
+    duration: float
+    error: Optional[str] = None
+    transcript: List[Dict[str, str]] = field(default_factory=list)
+    
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert to dictionary."""
+        return {
+            "success": self.success,
+            "responses": self.responses,
+            "tool_trace": [t.to_dict() for t in self.tool_trace],
+            "duration": self.duration,
+            "error": self.error,
+            "transcript": self.transcript,
+        }
+
+
+@dataclass
+class AgentConfig:
+    """Configuration for a single agent in headless mode."""
+    name: str = "HeadlessAgent"
+    instructions: str = "You are a helpful assistant with file and code tools."
+    role: str = "assistant"
+    llm: str = "gpt-4o-mini"
+    
+    def to_dict(self) -> Dict[str, Any]:
+        return {
+            "name": self.name,
+            "instructions": self.instructions,
+            "role": self.role,
+            "llm": self.llm,
+        }
+
+
+@dataclass 
+class HeadlessConfig:
+    """Configuration for headless interactive core execution."""
+    workspace: str = field(default_factory=os.getcwd)
+    model: str = "gpt-4o-mini"
+    approval_mode: str = "auto"  # auto, manual, scoped
+    enable_acp: bool = True
+    enable_lsp: bool = True
+    enable_basic: bool = True
+    timeout: int = 60
+    verbose: bool = False
+    # Multi-agent support
+    agents: List[AgentConfig] = field(default_factory=list)
+    workflow: Optional[Dict[str, Any]] = None  # Turn routing/handoffs
+    
+    def __post_init__(self):
+        if not self.agents:
+            # Default single agent
+            self.agents = [AgentConfig(llm=self.model)]
+
+
+def _redact_sensitive(obj: Any, sensitive_keys: set = None) -> Any:
+    """Redact sensitive information from args/kwargs."""
+    if sensitive_keys is None:
+        sensitive_keys = {"api_key", "password", "secret", "token", "key"}
+    
+    if isinstance(obj, dict):
+        return {
+            k: "[REDACTED]" if any(s in k.lower() for s in sensitive_keys) else _redact_sensitive(v, sensitive_keys)
+            for k, v in obj.items()
+        }
+    elif isinstance(obj, (list, tuple)):
+        return type(obj)(_redact_sensitive(item, sensitive_keys) for item in obj)
+    elif isinstance(obj, str) and len(obj) > 100:
+        return obj[:100] + "...[truncated]"
+    return obj
+
+
+def _truncate(s: str, max_len: int) -> str:
+    """Truncate string to max length."""
+    if len(s) <= max_len:
+        return s
+    return s[:max_len] + "...[truncated]"
+
+
+class HeadlessInteractiveCore:
+    """
+    Headless executor that uses the same interactive core pipeline as TUI.
+    
+    This class reuses:
+    - InteractiveRuntime for ACP/LSP subsystems
+    - get_interactive_tools() for tool resolution
+    - ActionOrchestrator for plan/approve/apply/verify flow
+    
+    Usage:
+        executor = HeadlessInteractiveCore(config)
+        result = executor.run(["Create a file hello.py with print('hello')"])
+        executor.cleanup()  # IMPORTANT: Always call cleanup to stop LSP/ACP
+        
+    Or use as context manager:
+        with HeadlessInteractiveCore(config) as executor:
+            result = executor.run(["Create a file"])
+    """
+    
+    def __init__(self, config: Optional[HeadlessConfig] = None):
+        self.config = config or HeadlessConfig()
+        self._runtime = None
+        self._tools: List[Callable] = []
+        self._tool_trace: List[ToolCallTrace] = []
+        self._agents: List[Any] = []  # Will hold Agent instances
+        self._initialized = False
+        self._cleanup_done = False
+    
+    def __enter__(self):
+        """Context manager entry."""
+        return self
+    
+    def __exit__(self, exc_type, exc_val, exc_tb):
+        """Context manager exit - ensures cleanup."""
+        self.cleanup()
+        return False
+    
+    def __del__(self):
+        """Destructor - attempt cleanup if not done."""
+        if not self._cleanup_done:
+            try:
+                self.cleanup()
+            except Exception:
+                pass  # Best effort cleanup in destructor
+    
+    def _wrap_tool_for_trace(self, tool: Callable) -> Callable:
+        """Wrap a tool to capture calls in trace."""
+        @functools.wraps(tool)
+        def wrapper(*args, **kwargs):
+            start = time.time()
+            timestamp = datetime.utcnow().isoformat()
+            try:
+                result = tool(*args, **kwargs)
+                trace = ToolCallTrace(
+                    tool_name=tool.__name__,
+                    args=args,
+                    kwargs=kwargs,
+                    result=result,
+                    success=True,
+                    error=None,
+                    duration=time.time() - start,
+                    timestamp=timestamp,
+                )
+                self._tool_trace.append(trace)
+                logger.debug(f"Tool {tool.__name__} succeeded in {trace.duration:.3f}s")
+                return result
+            except Exception as e:
+                trace = ToolCallTrace(
+                    tool_name=tool.__name__,
+                    args=args,
+                    kwargs=kwargs,
+                    result=None,
+                    success=False,
+                    error=str(e),
+                    duration=time.time() - start,
+                    timestamp=timestamp,
+                )
+                self._tool_trace.append(trace)
+                logger.debug(f"Tool {tool.__name__} failed: {e}")
+                raise
+        return wrapper
+    
+    def _initialize(self) -> None:
+        """Initialize runtime and tools using interactive core path."""
+        if self._initialized:
+            return
+        
+        logger.debug(f"Initializing headless core with workspace={self.config.workspace}")
+        
+        # Use the SAME tool loading path as interactive TUI
+        # This is from: praisonai/cli/features/interactive_tools.py
+        from praisonai.cli.features.interactive_tools import (
+            get_interactive_tools,
+            ToolConfig,
+        )
+        
+        # Create tool config matching interactive core
+        tool_config = ToolConfig(
+            workspace=self.config.workspace,
+            enable_acp=self.config.enable_acp,
+            enable_lsp=self.config.enable_lsp,
+            enable_basic=self.config.enable_basic,
+            approval_mode=self.config.approval_mode,
+            lsp_enabled=self.config.enable_lsp,
+            acp_enabled=self.config.enable_acp,
+        )
+        
+        # Get tools using the SAME function as TUI
+        raw_tools = get_interactive_tools(config=tool_config)
+        
+        # Wrap tools for tracing
+        self._tools = [self._wrap_tool_for_trace(t) for t in raw_tools]
+        
+        logger.debug(f"Loaded {len(self._tools)} tools: {[t.__name__ for t in self._tools]}")
+        
+        # Create agents using the SAME Agent class as TUI
+        from praisonaiagents import Agent
+        
+        for agent_config in self.config.agents:
+            agent = Agent(
+                name=agent_config.name,
+                role=agent_config.role,
+                goal="Help the user with their requests",
+                instructions=agent_config.instructions,
+                llm=agent_config.llm,
+                tools=self._tools if self._tools else None,
+            )
+            self._agents.append(agent)
+            logger.debug(f"Created agent: {agent_config.name}")
+        
+        self._initialized = True
+    
+    def run(self, prompts: Union[str, List[str]]) -> HeadlessExecutionResult:
+        """
+        Run prompts through headless interactive core.
+        
+        Args:
+            prompts: Single prompt or list of prompts for multi-step execution
+            
+        Returns:
+            HeadlessExecutionResult with responses, tool trace, and transcript
+        """
+        if isinstance(prompts, str):
+            prompts = [prompts]
+        
+        start_time = time.time()
+        self._tool_trace = []  # Reset trace
+        responses = []
+        transcript = []
+        
+        try:
+            self._initialize()
+            
+            # Execute prompts
+            for i, prompt in enumerate(prompts):
+                logger.debug(f"Executing prompt {i+1}/{len(prompts)}: {prompt[:50]}...")
+                transcript.append({"role": "user", "content": prompt})
+                
+                # Determine which agent to use
+                agent_idx = self._get_agent_for_prompt(i, prompt)
+                agent = self._agents[agent_idx]
+                
+                # Execute using the SAME chat method as TUI
+                response = agent.chat(prompt)
+                responses.append(response)
+                transcript.append({"role": "assistant", "content": response})
+                
+                logger.debug(f"Got response: {response[:100] if response else 'None'}...")
+            
+            return HeadlessExecutionResult(
+                success=True,
+                responses=responses,
+                tool_trace=self._tool_trace,
+                duration=time.time() - start_time,
+                transcript=transcript,
+            )
+            
+        except Exception as e:
+            logger.error(f"Headless execution failed: {e}", exc_info=True)
+            return HeadlessExecutionResult(
+                success=False,
+                responses=responses,
+                tool_trace=self._tool_trace,
+                duration=time.time() - start_time,
+                error=str(e),
+                transcript=transcript,
+            )
+    
+    def _get_agent_for_prompt(self, prompt_idx: int, prompt: str) -> int:
+        """Determine which agent should handle a prompt (for multi-agent)."""
+        if len(self._agents) == 1:
+            return 0
+        
+        # If workflow is defined, use it for routing
+        if self.config.workflow:
+            routing = self.config.workflow.get("routing", {})
+            # Simple round-robin or keyword-based routing
+            if "round_robin" in routing:
+                return prompt_idx % len(self._agents)
+            elif "keywords" in routing:
+                for i, agent_config in enumerate(self.config.agents):
+                    keywords = routing["keywords"].get(agent_config.name, [])
+                    if any(kw.lower() in prompt.lower() for kw in keywords):
+                        return i
+        
+        # Default: round-robin
+        return prompt_idx % len(self._agents)
+    
+    def get_tool_trace(self) -> List[Dict[str, Any]]:
+        """Get the tool call trace as list of dicts."""
+        return [t.to_dict() for t in self._tool_trace]
+    
+    def get_tools_called(self) -> List[str]:
+        """Get list of tool names that were called."""
+        return [t.tool_name for t in self._tool_trace]
+    
+    def clear_trace(self) -> None:
+        """Clear the tool trace."""
+        self._tool_trace = []
+    
+    def cleanup(self) -> None:
+        """
+        Clean up resources - MUST be called when done.
+        
+        Stops LSP and ACP subsystems that were started during initialization.
+        This is critical to prevent resource leaks and orphaned processes.
+        """
+        if self._cleanup_done:
+            return
+        
+        logger.debug("Cleaning up HeadlessInteractiveCore resources...")
+        
+        # Stop the runtime if it was created
+        if self._runtime is not None:
+            try:
+                import asyncio
+                
+                # Get or create event loop
+                try:
+                    loop = asyncio.get_running_loop()
+                except RuntimeError:
+                    loop = None
+                
+                if loop and loop.is_running():
+                    # Schedule the stop coroutine
+                    asyncio.ensure_future(self._runtime.stop())
+                else:
+                    # Create new loop and run
+                    asyncio.run(self._runtime.stop())
+                
+                logger.debug("Runtime stopped successfully")
+            except Exception as e:
+                logger.warning(f"Error stopping runtime: {e}")
+            finally:
+                self._runtime = None
+        
+        # Clear agents
+        self._agents = []
+        self._tools = []
+        self._initialized = False
+        self._cleanup_done = True
+        
+        logger.debug("HeadlessInteractiveCore cleanup complete")
+
+
+def run_headless(
+    prompts: Union[str, List[str]],
+    workspace: Optional[str] = None,
+    model: str = "gpt-4o-mini",
+    approval_mode: str = "auto",
+    timeout: int = 60,
+    agents: Optional[List[Dict[str, Any]]] = None,
+    workflow: Optional[Dict[str, Any]] = None,
+) -> HeadlessExecutionResult:
+    """
+    Convenience function to run prompts through headless interactive core.
+    
+    Args:
+        prompts: Single prompt or list of prompts
+        workspace: Working directory (default: cwd)
+        model: LLM model to use
+        approval_mode: Approval mode (auto, manual, scoped)
+        timeout: Timeout in seconds
+        agents: List of agent configs for multi-agent
+        workflow: Workflow config for turn routing
+        
+    Returns:
+        HeadlessExecutionResult
+        
+    Example:
+        result = run_headless("Create a file hello.py with print('hello')")
+        print(result.responses[0])
+        print(result.get_tools_called())
+    """
+    agent_configs = []
+    if agents:
+        for a in agents:
+            agent_configs.append(AgentConfig(
+                name=a.get("name", "Agent"),
+                instructions=a.get("instructions", "You are a helpful assistant."),
+                role=a.get("role", "assistant"),
+                llm=a.get("llm", model),
+            ))
+    
+    config = HeadlessConfig(
+        workspace=workspace or os.getcwd(),
+        model=model,
+        approval_mode=approval_mode,
+        timeout=timeout,
+        agents=agent_configs if agent_configs else None,
+        workflow=workflow,
+    )
+    
+    # Use context manager to ensure cleanup
+    with HeadlessInteractiveCore(config) as executor:
+        return executor.run(prompts)

src/praisonai/praisonai/cli/features/interactive_runtime.py

@@ -40,7 +40,7 @@ class RuntimeConfig:
     workspace: str = "."
     lsp_enabled: bool = True
     acp_enabled: bool = True
-    approval_mode: str = "manual"  # manual, auto, scoped
+    approval_mode: str = "auto"  # auto (full privileges), manual, scoped
     trace_enabled: bool = False
     trace_file: Optional[str] = None
     json_output: bool = False
@@ -146,15 +146,21 @@ class InteractiveRuntime:
     def read_only(self) -> bool:
         """Check if runtime is in read-only mode.
         
-        Note: When PRAISON_APPROVAL_MODE=auto, we allow writes even without ACP
-        to support non-interactive automation and testing scenarios.
+        Note: When approval_mode=auto (via config or env), we allow writes even 
+        without ACP to support non-interactive automation and testing scenarios.
         """
-        # Auto-approval mode bypasses read-only for automation
-        import os as _os
-        approval_mode = _os.environ.get("PRAISON_APPROVAL_MODE", "").lower()
-        if approval_mode == "auto":
+        # Check config approval_mode first (takes precedence)
+        if self.config.approval_mode == "auto":
             return False
-        return self._read_only or not self.acp_ready
+        
+        # Then check environment variable as fallback
+        import os as _os
+        env_approval_mode = _os.environ.get("PRAISON_APPROVAL_MODE", "").lower()
+        if env_approval_mode == "auto":
+            return False
+        
+        # Only enforce read-only if explicitly set or ACP not ready
+        return self._read_only
     
     def get_status(self) -> Dict[str, Any]:
         """Get runtime status."""

src/praisonai/praisonai/cli/features/interactive_test_harness.py

@@ -0,0 +1,537 @@
+"""
+Interactive Test Harness for PraisonAI.
+
+Provides workspace isolation, fixtures, tool trace wrappers, and artifact writing
+for testing interactive mode through the headless core executor.
+"""
+
+import json
+import logging
+import re
+import shutil
+import tempfile
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Any, Dict, List, Optional
+
+logger = logging.getLogger(__name__)
+
+
+# Blocked commands for safe execution
+BLOCKED_COMMANDS = {
+    "rm -rf /",
+    "rm -rf /*",
+    "sudo rm",
+    "mkfs",
+    "dd if=",
+    ":(){:|:&};:",  # Fork bomb
+    "chmod -R 777 /",
+    "curl | sh",
+    "wget | sh",
+}
+
+# Allowed commands (whitelist for strict mode)
+SAFE_COMMANDS = {
+    "echo",
+    "cat",
+    "ls",
+    "pwd",
+    "head",
+    "tail",
+    "wc",
+    "grep",
+    "find",
+    "python",
+    "python3",
+    "pip",
+    "pip3",
+    "node",
+    "npm",
+}
+
+
+@dataclass
+class WorkspaceFixture:
+    """Definition of a workspace fixture."""
+    name: str
+    files: Dict[str, str] = field(default_factory=dict)  # path -> content
+    directories: List[str] = field(default_factory=list)
+    
+    @classmethod
+    def empty(cls) -> "WorkspaceFixture":
+        """Create an empty workspace fixture."""
+        return cls(name="empty")
+    
+    @classmethod
+    def seeded(cls) -> "WorkspaceFixture":
+        """Create a seeded workspace with common files."""
+        return cls(
+            name="seeded",
+            files={
+                "README.md": "# Test Project\n\nThis is a test project.",
+                "main.py": "# Main entry point\nprint('Hello, World!')\n",
+                "config.json": '{"name": "test", "version": "1.0.0"}',
+            },
+            directories=["src", "tests"],
+        )
+    
+    @classmethod
+    def git(cls) -> "WorkspaceFixture":
+        """Create a git-initialized workspace."""
+        fixture = cls.seeded()
+        fixture.name = "git"
+        fixture.files[".gitignore"] = "__pycache__/\n*.pyc\n.env\n"
+        return fixture
+    
+    @classmethod
+    def python_project(cls) -> "WorkspaceFixture":
+        """Create a Python project workspace."""
+        return cls(
+            name="python_project",
+            files={
+                "README.md": "# Python Project\n",
+                "main.py": '''"""Main module."""
+
+def main():
+    """Main function."""
+    print("Hello from main!")
+
+if __name__ == "__main__":
+    main()
+''',
+                "utils.py": '''"""Utility functions."""
+
+def add(a: int, b: int) -> int:
+    """Add two numbers."""
+    return a + b
+
+def multiply(a: int, b: int) -> int:
+    """Multiply two numbers."""
+    return a * b
+
+class Calculator:
+    """Simple calculator class."""
+    
+    def __init__(self):
+        self.result = 0
+    
+    def add(self, x: int) -> int:
+        self.result += x
+        return self.result
+''',
+                "requirements.txt": "# No dependencies\n",
+            },
+            directories=["tests"],
+        )
+
+
+BUILTIN_FIXTURES = {
+    "empty": WorkspaceFixture.empty,
+    "seeded": WorkspaceFixture.seeded,
+    "git": WorkspaceFixture.git,
+    "python_project": WorkspaceFixture.python_project,
+}
+
+
+@dataclass
+class TestArtifacts:
+    """Container for test artifacts."""
+    transcript: List[Dict[str, str]] = field(default_factory=list)
+    tool_trace: List[Dict[str, Any]] = field(default_factory=list)
+    result: Dict[str, Any] = field(default_factory=dict)
+    workspace_snapshot: Optional[Dict[str, str]] = None
+    judge_result: Optional[Dict[str, Any]] = None
+    
+    def save(self, artifacts_dir: Path) -> None:
+        """Save artifacts to directory."""
+        artifacts_dir.mkdir(parents=True, exist_ok=True)
+        
+        # Save transcript
+        transcript_path = artifacts_dir / "transcript.txt"
+        with open(transcript_path, "w") as f:
+            for entry in self.transcript:
+                role = entry.get("role", "unknown")
+                content = entry.get("content", "")
+                f.write(f"[{role.upper()}]\n{content}\n\n")
+        
+        # Save tool trace as JSONL
+        trace_path = artifacts_dir / "tool_trace.jsonl"
+        with open(trace_path, "w") as f:
+            for trace in self.tool_trace:
+                f.write(json.dumps(trace) + "\n")
+        
+        # Save result
+        result_path = artifacts_dir / "result.json"
+        with open(result_path, "w") as f:
+            json.dump(self.result, f, indent=2, default=str)
+        
+        # Save workspace snapshot if present
+        if self.workspace_snapshot:
+            snapshot_dir = artifacts_dir / "workspace"
+            snapshot_dir.mkdir(exist_ok=True)
+            for rel_path, content in self.workspace_snapshot.items():
+                file_path = snapshot_dir / rel_path
+                file_path.parent.mkdir(parents=True, exist_ok=True)
+                file_path.write_text(content)
+        
+        # Save judge result if present
+        if self.judge_result:
+            judge_path = artifacts_dir / "judge_result.json"
+            with open(judge_path, "w") as f:
+                json.dump(self.judge_result, f, indent=2, default=str)
+        
+        logger.debug(f"Saved artifacts to {artifacts_dir}")
+
+
+class InteractiveTestHarness:
+    """
+    Test harness for interactive mode testing.
+    
+    Provides:
+    - Workspace isolation (temp directory)
+    - Fixture setup
+    - Tool trace capture
+    - Artifact generation
+    - File assertions
+    - Tool call assertions
+    """
+    
+    def __init__(
+        self,
+        workspace: Optional[Path] = None,
+        artifacts_dir: Optional[Path] = None,
+        keep_workspace: bool = False,
+        safe_commands: bool = True,
+    ):
+        """
+        Initialize test harness.
+        
+        Args:
+            workspace: Workspace directory (created if None)
+            artifacts_dir: Directory for artifacts (created if None)
+            keep_workspace: Keep workspace after cleanup
+            safe_commands: Enforce safe command execution
+        """
+        self._workspace_created = workspace is None
+        self.workspace = workspace or Path(tempfile.mkdtemp(prefix="praison_test_"))
+        self.artifacts_dir = artifacts_dir or self.workspace / "artifacts"
+        self.keep_workspace = keep_workspace
+        self.safe_commands = safe_commands
+        self.artifacts = TestArtifacts()
+        self._executor = None
+    
+    def setup_workspace(self, fixture: str = "empty") -> None:
+        """
+        Set up workspace with fixture.
+        
+        Args:
+            fixture: Fixture name (empty, seeded, git, python_project)
+        """
+        # Get fixture
+        if fixture in BUILTIN_FIXTURES:
+            fixture_obj = BUILTIN_FIXTURES[fixture]()
+        else:
+            logger.warning(f"Unknown fixture '{fixture}', using empty")
+            fixture_obj = WorkspaceFixture.empty()
+        
+        # Create directories
+        self.workspace.mkdir(parents=True, exist_ok=True)
+        for dir_path in fixture_obj.directories:
+            (self.workspace / dir_path).mkdir(parents=True, exist_ok=True)
+        
+        # Create files
+        for rel_path, content in fixture_obj.files.items():
+            file_path = self.workspace / rel_path
+            file_path.parent.mkdir(parents=True, exist_ok=True)
+            file_path.write_text(content)
+        
+        # Initialize git if needed
+        if fixture == "git":
+            import subprocess
+            try:
+                subprocess.run(
+                    ["git", "init"],
+                    cwd=self.workspace,
+                    capture_output=True,
+                    timeout=10,
+                )
+            except Exception as e:
+                logger.debug(f"Git init failed: {e}")
+        
+        logger.debug(f"Set up workspace with fixture '{fixture}' at {self.workspace}")
+    
+    def create_executor(
+        self,
+        model: str = "gpt-4o-mini",
+        approval_mode: str = "auto",
+        agents: Optional[List[Dict[str, Any]]] = None,
+        workflow: Optional[Dict[str, Any]] = None,
+    ):
+        """
+        Create headless executor using interactive core.
+        
+        Args:
+            model: LLM model
+            approval_mode: Approval mode (auto, manual, scoped)
+            agents: Multi-agent configs
+            workflow: Workflow routing config
+            
+        Returns:
+            HeadlessInteractiveCore executor
+        """
+        from .interactive_core_headless import (
+            HeadlessConfig,
+            HeadlessInteractiveCore,
+            AgentConfig,
+        )
+        
+        agent_configs = []
+        if agents:
+            for a in agents:
+                agent_configs.append(AgentConfig(
+                    name=a.get("name", "Agent"),
+                    instructions=a.get("instructions", "You are a helpful assistant."),
+                    role=a.get("role", "assistant"),
+                    llm=a.get("llm", model),
+                ))
+        
+        config = HeadlessConfig(
+            workspace=str(self.workspace),
+            model=model,
+            approval_mode=approval_mode,
+            agents=agent_configs if agent_configs else None,
+            workflow=workflow,
+        )
+        
+        self._executor = HeadlessInteractiveCore(config)
+        return self._executor
+    
+    def run(
+        self,
+        prompts: List[str],
+        model: str = "gpt-4o-mini",
+        approval_mode: str = "auto",
+        agents: Optional[List[Dict[str, Any]]] = None,
+        workflow: Optional[Dict[str, Any]] = None,
+    ):
+        """
+        Run prompts through headless interactive core.
+        
+        Args:
+            prompts: List of prompts to execute
+            model: LLM model
+            approval_mode: Approval mode
+            agents: Multi-agent configs
+            workflow: Workflow routing
+            
+        Returns:
+            HeadlessExecutionResult
+        """
+        if not self._executor:
+            self.create_executor(model, approval_mode, agents, workflow)
+        
+        result = self._executor.run(prompts)
+        
+        # Store in artifacts
+        self.artifacts.transcript = result.transcript
+        self.artifacts.tool_trace = self._executor.get_tool_trace()
+        self.artifacts.result = result.to_dict()
+        
+        return result
+    
+    def verify_files(self, expected_files: Dict[str, str]) -> Dict[str, bool]:
+        """
+        Verify expected files exist and match content patterns.
+        
+        Args:
+            expected_files: Dict of {relative_path: content_regex}
+            
+        Returns:
+            Dict of {path: passed}
+        """
+        results = {}
+        
+        for rel_path, content_pattern in expected_files.items():
+            file_path = self.workspace / rel_path
+            
+            if not file_path.exists():
+                logger.debug(f"File not found: {rel_path}")
+                results[rel_path] = False
+                continue
+            
+            content = file_path.read_text()
+            
+            if content_pattern:
+                try:
+                    if re.search(content_pattern, content, re.MULTILINE | re.DOTALL):
+                        results[rel_path] = True
+                    else:
+                        logger.debug(f"Content mismatch for {rel_path}: pattern '{content_pattern}' not found")
+                        results[rel_path] = False
+                except re.error as e:
+                    logger.warning(f"Invalid regex pattern for {rel_path}: {e}")
+                    results[rel_path] = False
+            else:
+                # Just check existence
+                results[rel_path] = True
+        
+        return results
+    
+    def verify_tool_calls(
+        self,
+        expected_tools: Optional[List[str]] = None,
+        forbidden_tools: Optional[List[str]] = None,
+    ) -> Dict[str, Any]:
+        """
+        Verify tool calls match expectations.
+        
+        Args:
+            expected_tools: Tools that MUST have been called
+            forbidden_tools: Tools that MUST NOT have been called
+            
+        Returns:
+            Dict with verification results
+        """
+        called_tools = set()
+        if self._executor:
+            called_tools = set(self._executor.get_tools_called())
+        
+        results = {
+            "called_tools": list(called_tools),
+            "expected_passed": True,
+            "forbidden_passed": True,
+            "missing_tools": [],
+            "forbidden_called": [],
+        }
+        
+        # Check expected tools
+        if expected_tools:
+            expected_set = set(expected_tools)
+            missing = expected_set - called_tools
+            if missing:
+                results["expected_passed"] = False
+                results["missing_tools"] = list(missing)
+        
+        # Check forbidden tools
+        if forbidden_tools:
+            forbidden_set = set(forbidden_tools)
+            forbidden_called = forbidden_set & called_tools
+            if forbidden_called:
+                results["forbidden_passed"] = False
+                results["forbidden_called"] = list(forbidden_called)
+        
+        results["passed"] = results["expected_passed"] and results["forbidden_passed"]
+        return results
+    
+    def verify_response(self, response: str, pattern: str) -> bool:
+        """
+        Verify response matches pattern.
+        
+        Args:
+            response: Response text
+            pattern: Regex pattern
+            
+        Returns:
+            True if matches
+        """
+        if not pattern:
+            return True
+        
+        try:
+            return bool(re.search(pattern, response, re.MULTILINE | re.DOTALL | re.IGNORECASE))
+        except re.error as e:
+            logger.warning(f"Invalid response pattern: {e}")
+            return False
+    
+    def snapshot_workspace(self) -> Dict[str, str]:
+        """
+        Create snapshot of workspace files.
+        
+        Returns:
+            Dict of {relative_path: content}
+        """
+        snapshot = {}
+        
+        for file_path in self.workspace.rglob("*"):
+            if file_path.is_file():
+                rel_path = str(file_path.relative_to(self.workspace))
+                # Skip binary files and large files
+                if file_path.suffix in {".pyc", ".so", ".dll", ".exe", ".bin"}:
+                    continue
+                if file_path.stat().st_size > 100_000:  # 100KB limit
+                    continue
+                try:
+                    snapshot[rel_path] = file_path.read_text()
+                except Exception:
+                    pass  # Skip unreadable files
+        
+        self.artifacts.workspace_snapshot = snapshot
+        return snapshot
+    
+    def save_artifacts(self, test_id: str = "test") -> Path:
+        """
+        Save all artifacts to directory.
+        
+        Args:
+            test_id: Test identifier for subdirectory
+            
+        Returns:
+            Path to artifacts directory
+        """
+        test_artifacts_dir = self.artifacts_dir / test_id
+        self.artifacts.save(test_artifacts_dir)
+        return test_artifacts_dir
+    
+    def cleanup(self) -> None:
+        """Clean up workspace and executor resources."""
+        # Clean up executor (stops LSP/ACP subsystems)
+        if self._executor is not None:
+            try:
+                self._executor.cleanup()
+                logger.debug("Cleaned up executor resources")
+            except Exception as e:
+                logger.warning(f"Failed to clean up executor: {e}")
+            finally:
+                self._executor = None
+        
+        # Clean up workspace
+        if self._workspace_created and not self.keep_workspace:
+            try:
+                shutil.rmtree(self.workspace)
+                logger.debug(f"Cleaned up workspace: {self.workspace}")
+            except Exception as e:
+                logger.warning(f"Failed to clean up workspace: {e}")
+    
+    def __enter__(self):
+        return self
+    
+    def __exit__(self, exc_type, exc_val, exc_tb):
+        self.cleanup()
+        return False
+
+
+def create_test_harness(
+    fixture: str = "empty",
+    workspace: Optional[Path] = None,
+    artifacts_dir: Optional[Path] = None,
+    keep_workspace: bool = False,
+) -> InteractiveTestHarness:
+    """
+    Create and set up a test harness.
+    
+    Args:
+        fixture: Workspace fixture name
+        workspace: Workspace path (temp if None)
+        artifacts_dir: Artifacts directory
+        keep_workspace: Keep workspace after cleanup
+        
+    Returns:
+        Configured InteractiveTestHarness
+    """
+    harness = InteractiveTestHarness(
+        workspace=workspace,
+        artifacts_dir=artifacts_dir,
+        keep_workspace=keep_workspace,
+    )
+    harness.setup_workspace(fixture)
+    return harness

src/praisonai/praisonai/cli/features/interactive_tools.py

@@ -72,7 +72,7 @@ class ToolConfig:
     enable_acp: bool = True
     enable_lsp: bool = True
     enable_basic: bool = True
-    approval_mode: str = "auto"  # auto, manual, scoped
+    approval_mode: str = "auto"  # auto (full privileges), manual, scoped
     lsp_enabled: bool = True
     acp_enabled: bool = True