Release v3.9.2

- MemoryConfig from praisonaiagents.config (not memory)
- FileMemory from praisonaiagents.memory
- OutputConfig, ExecutionConfig, PlanningConfig from config

.github/workflows/test-windows.yml

@@ -0,0 +1,204 @@
+name: Windows Tests
+
+on:
+  push:
+    branches: [main, develop]
+  pull_request:
+    branches: [main, develop]
+  workflow_dispatch:
+
+jobs:
+  test-windows:
+    runs-on: windows-latest
+    timeout-minutes: 5
+    env:
+      OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
+      OPENAI_MODEL_NAME: gpt-4o-mini
+    
+    steps:
+    - uses: actions/checkout@v4
+    
+    - name: Set up Python 3.11
+      uses: actions/setup-python@v5
+      with:
+        python-version: "3.11"
+    
+    - name: Install praisonaiagents
+      run: |
+        python -m pip install --upgrade pip
+        cd src/praisonai-agents
+        pip install .
+      shell: bash
+    
+    - name: Install praisonai wrapper
+      run: |
+        cd src/praisonai
+        pip install .[dev]
+      shell: bash
+
+    # ==================== CORE IMPORTS ====================
+    - name: Test core imports
+      run: |
+        python -c "from praisonaiagents import Agent, Agents, Workflow; print('Core imports OK')"
+        python -c "from praisonaiagents import tool, Tools; print('Tool imports OK')"
+        python -c "from praisonaiagents import Task, WorkflowStep; print('Task/Step imports OK')"
+
+    # ==================== AGENT FEATURES ====================
+    - name: Test Agent instantiation
+      run: |
+        python -c "
+        from praisonaiagents import Agent
+        a = Agent(name='Test', instructions='You are helpful')
+        print('Agent created:', a.name)
+        a2 = Agent(name='ToolAgent', instructions='Use tools', tools=[])
+        print('Agent with tools:', a2.name)
+        "
+
+    - name: Test Agents orchestration
+      run: |
+        python -c "
+        from praisonaiagents import Agent, Agents, Task
+        a1 = Agent(name='Agent1', instructions='First agent')
+        a2 = Agent(name='Agent2', instructions='Second agent')
+        t = Task(description='Do something', agent=a1)
+        agents = Agents(agents=[a1, a2], tasks=[t])
+        print('Agents:', len(agents.agents), 'Tasks:', len(agents.tasks))
+        "
+
+    # ==================== WORKFLOW FEATURES ====================
+    - name: Test Workflow and WorkflowStep
+      run: |
+        python -c "
+        from praisonaiagents import Workflow, WorkflowStep
+        from praisonaiagents.workflows import WorkflowOutputConfig
+        
+        w = Workflow(name='test')
+        s1 = WorkflowStep(name='step1')
+        s2 = WorkflowStep(name='step2', context=['step1'])
+        print('Workflow:', w.name)
+        print('Steps:', s1.name, s2.name)
+        
+        cfg = WorkflowOutputConfig(verbose=True, stream=False)
+        w2 = Workflow(name='configured', output=cfg)
+        print('Configured workflow verbose:', w2.verbose)
+        "
+
+    # ==================== MEMORY FEATURES ====================
+    - name: Test Memory classes
+      run: |
+        python -c "
+        from praisonaiagents.memory import FileMemory
+        from praisonaiagents.config import MemoryConfig
+        
+        cfg = MemoryConfig(backend='file', user_id='test_user')
+        print('MemoryConfig backend:', cfg.backend)
+        
+        mem = FileMemory(user_id='test_user', base_path='.praison/test_memory')
+        print('FileMemory user:', mem.user_id)
+        "
+
+    # ==================== TOOL DECORATOR ====================
+    - name: Test tool decorator
+      run: |
+        python -c "
+        from praisonaiagents import tool
+        
+        @tool
+        def my_tool(query: str) -> str:
+            '''A test tool'''
+            return f'Echo: {query}'
+        
+        result = my_tool('hello')
+        print('Tool result:', result)
+        "
+
+    # ==================== CONFIG CLASSES ====================
+    - name: Test config classes
+      run: |
+        python -c "
+        from praisonaiagents.config import OutputConfig, ExecutionConfig, PlanningConfig
+        
+        out_cfg = OutputConfig(verbose=True, stream=True)
+        print('OutputConfig verbose:', out_cfg.verbose)
+        
+        exec_cfg = ExecutionConfig(max_iter=50, max_retry_limit=5)
+        print('ExecutionConfig max_iter:', exec_cfg.max_iter)
+        
+        plan_cfg = PlanningConfig(reasoning=True, auto_approve=False)
+        print('PlanningConfig reasoning:', plan_cfg.reasoning)
+        "
+    # ==================== HANDOFF ====================
+    - name: Test Handoff
+      run: |
+        python -c "
+        from praisonaiagents import Handoff, handoff, Agent
+        
+        a1 = Agent(name='Agent1', instructions='First')
+        a2 = Agent(name='Agent2', instructions='Second')
+        
+        # Using Handoff class directly
+        h = Handoff(agent=a2)
+        print('Handoff tool name:', h.tool_name)
+        
+        # Using handoff function
+        h2 = handoff(a1, tool_name_override='custom_handoff')
+        print('Custom handoff name:', h2.tool_name)
+        "
+
+
+    # ==================== ASYNC FEATURES ====================
+    - name: Test async imports
+      run: |
+        python -c "
+        import asyncio
+        from praisonaiagents import Agent
+        
+        async def test_async():
+            agent = Agent(name='AsyncAgent', instructions='Test')
+            assert hasattr(agent, 'arun'), 'Missing arun'
+            assert hasattr(agent, 'astart'), 'Missing astart'
+            print('Async methods available')
+        
+        asyncio.run(test_async())
+        "
+
+    # ==================== CLI COMMANDS ====================
+    - name: Test CLI help
+      run: praisonai --help
+
+    - name: Test CLI session list
+      run: praisonai session list
+      continue-on-error: true
+
+    - name: Test CLI queue list
+      run: praisonai queue list
+      continue-on-error: true
+
+    # ==================== REAL API TESTS ====================
+    - name: Real API test - Agent run
+      if: env.OPENAI_API_KEY != ''
+      run: |
+        python -c "
+        from praisonaiagents import Agent
+        agent = Agent(name='TestAgent', instructions='Reply with exactly: WINDOWS_OK', llm='gpt-4o-mini')
+        result = agent.run('Say WINDOWS_OK')
+        print('Result:', result)
+        print('Real API test passed')
+        "
+
+    - name: Real API test - Multi-agent
+      if: env.OPENAI_API_KEY != ''
+      run: |
+        python -c "
+        from praisonaiagents import Agent, Agents, Task
+        
+        researcher = Agent(name='Researcher', instructions='You research topics', llm='gpt-4o-mini')
+        writer = Agent(name='Writer', instructions='You write content', llm='gpt-4o-mini')
+        
+        task = Task(description='Say hello world', agent=researcher)
+        
+        agents = Agents(agents=[researcher, writer], tasks=[task])
+        result = agents.start()
+        print('Multi-agent result type:', type(result))
+        print('Multi-agent test passed')
+        "

.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

README.md

@@ -23,7 +23,7 @@
 
 </div>
 
-PraisonAI is a production-ready Multi-AI Agents framework with self-reflection, designed to create AI Agents to automate and solve problems ranging from simple tasks to complex challenges. By integrating PraisonAI Agents, AG2 (Formerly AutoGen), and CrewAI into a low-code solution, it streamlines the building and management of multi-agent LLM systems, emphasising simplicity, customisation, and effective human-agent collaboration.
+PraisonAI is a production-ready Multi-AI Agents framework with self-reflection, designed to create AI Agents to automate and solve problems ranging from simple tasks to complex challenges. It provides a low-code solution to streamline the building and management of multi-agent LLM systems, emphasising simplicity, customisation, and effective human-agent collaboration.
 
 <div align="center">
   <a href="https://docs.praison.ai">

api.md

@@ -0,0 +1,441 @@
+# PraisonAI API Reference
+
+This file is auto-generated. Do not edit manually.
+Regenerate with: `praisonai docs api-md --write`
+
+# Agents
+
+Types:
+```python
+from praisonaiagents import Agent, Agents
+```
+
+Methods:
+
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">achat</a>(prompt: str, temperature = 1.0, tools = None, output_json = None, output_pydantic = None, reasoning_steps = False, task_name = None, task_description = None, task_id = None)</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">aexecute</a>(task, context = None)</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">agent_id</a>()</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">analyze_prompt</a>(prompt: str) -> set</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">arun</a>(prompt: str, **kwargs)</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">astart</a>(prompt: str, **kwargs)</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">auto_memory</a>()</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">auto_memory</a>(value)</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">background</a>()</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">background</a>(value)</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">chat</a>(prompt, temperature = 1.0, tools = None, output_json = None, output_pydantic = None, reasoning_steps = False, stream = None, task_name = None, task_description = None, task_id = None, config = None, force_retrieval = False, skip_retrieval = False)</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">chat_with_context</a>(message: str, context: 'ContextPack', **kwargs) -> str</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">checkpoints</a>()</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">checkpoints</a>(value)</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">clean_json_output</a>(output: str) -> str</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">clear_history</a>()</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">console</a>()</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">context_manager</a>()</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">context_manager</a>(value)</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">display_generating</a>(content: str, start_time: float)</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">execute</a>(task, context = None)</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">execute_tool</a>(function_name, arguments)</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">execute_tool_async</a>(function_name: str, arguments: Dict[str, Any]) -> Any</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">from_template</a>(uri: str, config: Optional[Dict[str, Any]] = None, offline: bool = False, **kwargs) -> 'Agent'</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">generate_task</a>() -> 'Task'</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">get_available_tools</a>() -> List[Any]</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">get_memory_context</a>(query: Optional[str] = None) -> str</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">get_recommended_stage</a>(prompt: str) -> str</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">get_rules_context</a>(file_path: Optional[str] = None, include_manual: Optional[List[str]] = None) -> str</code>
+* <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>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">output_style</a>(value)</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">policy</a>()</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">policy</a>(value)</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">query</a>(question: str, **kwargs) -> 'RAGResult'</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">rag</a>()</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">rag_query</a>(question: str, **kwargs) -> 'RAGResult'</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">retrieval_config</a>()</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">retrieve</a>(query: str, **kwargs) -> 'ContextPack'</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">rules_manager</a>()</code>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">run</a>(prompt: str, **kwargs)</code>
+* <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 = 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>
+* <code title="class Agent">Agent.<a href="./src/praisonai-agents/praisonaiagents/agent/agent.py">thinking_budget</a>(value)</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">add_task</a>(task)</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">aexecute_task</a>(task_id)</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">append_to_state</a>(key: str, value: Any, max_length: Optional[int] = None) -> List[Any]</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">arun_all_tasks</a>()</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">arun_task</a>(task_id)</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">astart</a>(content = None, return_dict = False, **kwargs)</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">clean_json_output</a>(output: str) -> str</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">clear_state</a>() -> None</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">context_manager</a>()</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">current_plan</a>()</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">default_completion_checker</a>(task, agent_output)</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">delete_state</a>(key: str) -> bool</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">display_token_usage</a>()</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">execute_task</a>(task_id)</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">get_agent_details</a>(agent_name)</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">get_all_state</a>() -> Dict[str, Any]</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">get_all_tasks_status</a>()</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">get_detailed_token_report</a>() -> Dict[str, Any]</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">get_plan_markdown</a>() -> str</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">get_state</a>(key: str, default: Any = None) -> Any</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">get_task_details</a>(task_id)</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">get_task_result</a>(task_id)</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">get_task_status</a>(task_id)</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">get_todo_markdown</a>() -> str</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">get_token_usage_summary</a>() -> Dict[str, Any]</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">has_state</a>(key: str) -> bool</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">increment_state</a>(key: str, amount: float = 1, default: float = 0) -> float</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">launch</a>(path: str = '/agents', port: int = 8000, host: str = '0.0.0.0', debug: bool = False, protocol: str = 'http')</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">restore_session_state</a>(session_id: str) -> bool</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">run</a>(content = None, return_dict = False, **kwargs)</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">run_all_tasks</a>()</code>
+* <code title="class Agents">Agents.<a href="./src/praisonai-agents/praisonaiagents/agents/agents.py">run_task</a>(task_id)</code>
+* <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, 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>
+
+# Tools
+
+Types:
+```python
+from praisonaiagents import Tools, tool
+```
+
+Methods:
+
+* <code title="function">praisonaiagents.<a href="./src/praisonai-agents/praisonaiagents/tools/decorator.py">tool</a>(func: Optional[Callable] = None) -> Union[FunctionTool, Callable[[Callable], FunctionTool]]</code>
+
+# Other
+
+Types:
+```python
+from praisonaiagents import Task
+```
+
+Methods:
+
+* <code title="class Task">Task.<a href="./src/praisonai-agents/praisonaiagents/task/task.py">execute_callback</a>(task_output: TaskOutput) -> None</code>
+* <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>
+
+# Wrapper (praisonai)
+
+Types:
+```python
+from praisonai import CloudProvider, Deploy, DeployConfig, DeployType, PraisonAI, __version__
+```
+
+# CLI
+
+Methods:
+
+* <code title="cli">praisonai <a href="./src/praisonai/praisonai/cli/main.py">--help</a></code>
+* <code title="cli">praisonai acp acp-main <a href="./src/praisonai/praisonai/cli/commands/acp.py">--help</a></code>
+* <code title="cli">praisonai agents create <a href="./src/praisonai/praisonai/cli/commands/agents.py">--help</a></code>
+* <code title="cli">praisonai agents info <a href="./src/praisonai/praisonai/cli/commands/agents.py">--help</a></code>
+* <code title="cli">praisonai agents list <a href="./src/praisonai/praisonai/cli/commands/agents.py">--help</a></code>
+* <code title="cli">praisonai audit agent-centric <a href="./src/praisonai/praisonai/cli/commands/audit.py">--help</a></code>
+* <code title="cli">praisonai benchmark agent <a href="./src/praisonai/praisonai/cli/commands/benchmark.py">--help</a></code>
+* <code title="cli">praisonai benchmark benchmark-callback <a href="./src/praisonai/praisonai/cli/commands/benchmark.py">--help</a></code>
+* <code title="cli">praisonai benchmark cli <a href="./src/praisonai/praisonai/cli/commands/benchmark.py">--help</a></code>
+* <code title="cli">praisonai benchmark compare <a href="./src/praisonai/praisonai/cli/commands/benchmark.py">--help</a></code>
+* <code title="cli">praisonai benchmark litellm <a href="./src/praisonai/praisonai/cli/commands/benchmark.py">--help</a></code>
+* <code title="cli">praisonai benchmark profile <a href="./src/praisonai/praisonai/cli/commands/benchmark.py">--help</a></code>
+* <code title="cli">praisonai benchmark sdk <a href="./src/praisonai/praisonai/cli/commands/benchmark.py">--help</a></code>
+* <code title="cli">praisonai benchmark workflow <a href="./src/praisonai/praisonai/cli/commands/benchmark.py">--help</a></code>
+* <code title="cli">praisonai call call-main <a href="./src/praisonai/praisonai/cli/commands/call.py">--help</a></code>
+* <code title="cli">praisonai chat <a href="./src/praisonai/praisonai/cli/main.py">--help</a></code>
+* <code title="cli">praisonai chat chat-main <a href="./src/praisonai/praisonai/cli/commands/chat.py">--help</a></code>
+* <code title="cli">praisonai code <a href="./src/praisonai/praisonai/cli/main.py">--help</a></code>
+* <code title="cli">praisonai code code-main <a href="./src/praisonai/praisonai/cli/commands/code.py">--help</a></code>
+* <code title="cli">praisonai commit commit-main <a href="./src/praisonai/praisonai/cli/commands/commit.py">--help</a></code>
+* <code title="cli">praisonai completion bash <a href="./src/praisonai/praisonai/cli/commands/completion.py">--help</a></code>
+* <code title="cli">praisonai completion completion-callback <a href="./src/praisonai/praisonai/cli/commands/completion.py">--help</a></code>
+* <code title="cli">praisonai completion fish <a href="./src/praisonai/praisonai/cli/commands/completion.py">--help</a></code>
+* <code title="cli">praisonai completion zsh <a href="./src/praisonai/praisonai/cli/commands/completion.py">--help</a></code>
+* <code title="cli">praisonai config get <a href="./src/praisonai/praisonai/cli/commands/config.py">--help</a></code>
+* <code title="cli">praisonai config list <a href="./src/praisonai/praisonai/cli/commands/config.py">--help</a></code>
+* <code title="cli">praisonai config path <a href="./src/praisonai/praisonai/cli/commands/config.py">--help</a></code>
+* <code title="cli">praisonai config reset <a href="./src/praisonai/praisonai/cli/commands/config.py">--help</a></code>
+* <code title="cli">praisonai config set <a href="./src/praisonai/praisonai/cli/commands/config.py">--help</a></code>
+* <code title="cli">praisonai context add <a href="./src/praisonai/praisonai/cli/commands/context.py">--help</a></code>
+* <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>
+* <code title="cli">praisonai debug lsp <a href="./src/praisonai/praisonai/cli/commands/debug.py">--help</a></code>
+* <code title="cli">praisonai debug trace <a href="./src/praisonai/praisonai/cli/commands/debug.py">--help</a></code>
+* <code title="cli">praisonai deploy aws <a href="./src/praisonai/praisonai/cli/commands/deploy.py">--help</a></code>
+* <code title="cli">praisonai deploy azure <a href="./src/praisonai/praisonai/cli/commands/deploy.py">--help</a></code>
+* <code title="cli">praisonai deploy docker <a href="./src/praisonai/praisonai/cli/commands/deploy.py">--help</a></code>
+* <code title="cli">praisonai deploy gcp <a href="./src/praisonai/praisonai/cli/commands/deploy.py">--help</a></code>
+* <code title="cli">praisonai diag diag-callback <a href="./src/praisonai/praisonai/cli/commands/diag.py">--help</a></code>
+* <code title="cli">praisonai diag export <a href="./src/praisonai/praisonai/cli/commands/diag.py">--help</a></code>
+* <code title="cli">praisonai docs api-md <a href="./src/praisonai/praisonai/cli/commands/docs.py">--help</a></code>
+* <code title="cli">praisonai docs generate <a href="./src/praisonai/praisonai/cli/commands/docs.py">--help</a></code>
+* <code title="cli">praisonai docs list <a href="./src/praisonai/praisonai/cli/commands/docs.py">--help</a></code>
+* <code title="cli">praisonai docs report <a href="./src/praisonai/praisonai/cli/commands/docs.py">--help</a></code>
+* <code title="cli">praisonai docs run <a href="./src/praisonai/praisonai/cli/commands/docs.py">--help</a></code>
+* <code title="cli">praisonai docs run-all <a href="./src/praisonai/praisonai/cli/commands/docs.py">--help</a></code>
+* <code title="cli">praisonai docs serve <a href="./src/praisonai/praisonai/cli/commands/docs.py">--help</a></code>
+* <code title="cli">praisonai docs stats <a href="./src/praisonai/praisonai/cli/commands/docs.py">--help</a></code>
+* <code title="cli">praisonai doctor cleanup <a href="./src/praisonai/praisonai/cli/commands/doctor.py">--help</a></code>
+* <code title="cli">praisonai doctor config <a href="./src/praisonai/praisonai/cli/commands/doctor.py">--help</a></code>
+* <code title="cli">praisonai doctor db <a href="./src/praisonai/praisonai/cli/commands/doctor.py">--help</a></code>
+* <code title="cli">praisonai doctor doctor-callback <a href="./src/praisonai/praisonai/cli/commands/doctor.py">--help</a></code>
+* <code title="cli">praisonai doctor env <a href="./src/praisonai/praisonai/cli/commands/doctor.py">--help</a></code>
+* <code title="cli">praisonai doctor mcp <a href="./src/praisonai/praisonai/cli/commands/doctor.py">--help</a></code>
+* <code title="cli">praisonai doctor network <a href="./src/praisonai/praisonai/cli/commands/doctor.py">--help</a></code>
+* <code title="cli">praisonai doctor performance <a href="./src/praisonai/praisonai/cli/commands/doctor.py">--help</a></code>
+* <code title="cli">praisonai doctor selftest <a href="./src/praisonai/praisonai/cli/commands/doctor.py">--help</a></code>
+* <code title="cli">praisonai doctor tools <a href="./src/praisonai/praisonai/cli/commands/doctor.py">--help</a></code>
+* <code title="cli">praisonai doctor troubleshoot <a href="./src/praisonai/praisonai/cli/commands/doctor.py">--help</a></code>
+* <code title="cli">praisonai endpoints list <a href="./src/praisonai/praisonai/cli/commands/endpoints.py">--help</a></code>
+* <code title="cli">praisonai endpoints test <a href="./src/praisonai/praisonai/cli/commands/endpoints.py">--help</a></code>
+* <code title="cli">praisonai environment check <a href="./src/praisonai/praisonai/cli/commands/environment.py">--help</a></code>
+* <code title="cli">praisonai environment doctor <a href="./src/praisonai/praisonai/cli/commands/environment.py">--help</a></code>
+* <code title="cli">praisonai environment view <a href="./src/praisonai/praisonai/cli/commands/environment.py">--help</a></code>
+* <code title="cli">praisonai eval accuracy <a href="./src/praisonai/praisonai/cli/commands/eval.py">--help</a></code>
+* <code title="cli">praisonai eval performance <a href="./src/praisonai/praisonai/cli/commands/eval.py">--help</a></code>
+* <code title="cli">praisonai examples info <a href="./src/praisonai/praisonai/cli/commands/examples.py">--help</a></code>
+* <code title="cli">praisonai examples list <a href="./src/praisonai/praisonai/cli/commands/examples.py">--help</a></code>
+* <code title="cli">praisonai examples report <a href="./src/praisonai/praisonai/cli/commands/examples.py">--help</a></code>
+* <code title="cli">praisonai examples run <a href="./src/praisonai/praisonai/cli/commands/examples.py">--help</a></code>
+* <code title="cli">praisonai examples run-all <a href="./src/praisonai/praisonai/cli/commands/examples.py">--help</a></code>
+* <code title="cli">praisonai examples stats <a href="./src/praisonai/praisonai/cli/commands/examples.py">--help</a></code>
+* <code title="cli">praisonai hooks add <a href="./src/praisonai/praisonai/cli/commands/hooks.py">--help</a></code>
+* <code title="cli">praisonai hooks list <a href="./src/praisonai/praisonai/cli/commands/hooks.py">--help</a></code>
+* <code title="cli">praisonai hooks remove <a href="./src/praisonai/praisonai/cli/commands/hooks.py">--help</a></code>
+* <code title="cli">praisonai knowledge add <a href="./src/praisonai/praisonai/cli/commands/knowledge.py">--help</a></code>
+* <code title="cli">praisonai knowledge index <a href="./src/praisonai/praisonai/cli/commands/knowledge.py">--help</a></code>
+* <code title="cli">praisonai knowledge list <a href="./src/praisonai/praisonai/cli/commands/knowledge.py">--help</a></code>
+* <code title="cli">praisonai knowledge search <a href="./src/praisonai/praisonai/cli/commands/knowledge.py">--help</a></code>
+* <code title="cli">praisonai lsp logs <a href="./src/praisonai/praisonai/cli/commands/lsp.py">--help</a></code>
+* <code title="cli">praisonai lsp lsp-callback <a href="./src/praisonai/praisonai/cli/commands/lsp.py">--help</a></code>
+* <code title="cli">praisonai lsp start <a href="./src/praisonai/praisonai/cli/commands/lsp.py">--help</a></code>
+* <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 run <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>
+* <code title="cli">praisonai profile imports <a href="./src/praisonai/praisonai/cli/commands/profile.py">--help</a></code>
+* <code title="cli">praisonai profile optimize <a href="./src/praisonai/praisonai/cli/commands/profile.py">--help</a></code>
+* <code title="cli">praisonai profile profile-callback <a href="./src/praisonai/praisonai/cli/commands/profile.py">--help</a></code>
+* <code title="cli">praisonai profile query <a href="./src/praisonai/praisonai/cli/commands/profile.py">--help</a></code>
+* <code title="cli">praisonai profile snapshot <a href="./src/praisonai/praisonai/cli/commands/profile.py">--help</a></code>
+* <code title="cli">praisonai profile startup <a href="./src/praisonai/praisonai/cli/commands/profile.py">--help</a></code>
+* <code title="cli">praisonai profile suite <a href="./src/praisonai/praisonai/cli/commands/profile.py">--help</a></code>
+* <code title="cli">praisonai rag chat <a href="./src/praisonai/praisonai/cli/commands/rag.py">--help</a></code>
+* <code title="cli">praisonai rag eval <a href="./src/praisonai/praisonai/cli/commands/rag.py">--help</a></code>
+* <code title="cli">praisonai rag index <a href="./src/praisonai/praisonai/cli/commands/rag.py">--help</a></code>
+* <code title="cli">praisonai rag query <a href="./src/praisonai/praisonai/cli/commands/rag.py">--help</a></code>
+* <code title="cli">praisonai rag serve <a href="./src/praisonai/praisonai/cli/commands/rag.py">--help</a></code>
+* <code title="cli">praisonai realtime realtime-main <a href="./src/praisonai/praisonai/cli/commands/realtime.py">--help</a></code>
+* <code title="cli">praisonai recipe install <a href="./src/praisonai/praisonai/cli/commands/recipe.py">--help</a></code>
+* <code title="cli">praisonai recipe list <a href="./src/praisonai/praisonai/cli/commands/recipe.py">--help</a></code>
+* <code title="cli">praisonai recipe run <a href="./src/praisonai/praisonai/cli/commands/recipe.py">--help</a></code>
+* <code title="cli">praisonai registry list <a href="./src/praisonai/praisonai/cli/commands/registry.py">--help</a></code>
+* <code title="cli">praisonai registry serve <a href="./src/praisonai/praisonai/cli/commands/registry.py">--help</a></code>
+* <code title="cli">praisonai research research-main <a href="./src/praisonai/praisonai/cli/commands/research.py">--help</a></code>
+* <code title="cli">praisonai retrieval index <a href="./src/praisonai/praisonai/cli/commands/retrieval.py">--help</a></code>
+* <code title="cli">praisonai retrieval query <a href="./src/praisonai/praisonai/cli/commands/retrieval.py">--help</a></code>
+* <code title="cli">praisonai retrieval search <a href="./src/praisonai/praisonai/cli/commands/retrieval.py">--help</a></code>
+* <code title="cli">praisonai rules add <a href="./src/praisonai/praisonai/cli/commands/rules.py">--help</a></code>
+* <code title="cli">praisonai rules clear <a href="./src/praisonai/praisonai/cli/commands/rules.py">--help</a></code>
+* <code title="cli">praisonai rules list <a href="./src/praisonai/praisonai/cli/commands/rules.py">--help</a></code>
+* <code title="cli">praisonai run <a href="./src/praisonai/praisonai/cli/main.py">--help</a></code>
+* <code title="cli">praisonai run run-main <a href="./src/praisonai/praisonai/cli/commands/run.py">--help</a></code>
+* <code title="cli">praisonai schedule delete <a href="./src/praisonai/praisonai/cli/commands/schedule.py">--help</a></code>
+* <code title="cli">praisonai schedule describe <a href="./src/praisonai/praisonai/cli/commands/schedule.py">--help</a></code>
+* <code title="cli">praisonai schedule list <a href="./src/praisonai/praisonai/cli/commands/schedule.py">--help</a></code>
+* <code title="cli">praisonai schedule logs <a href="./src/praisonai/praisonai/cli/commands/schedule.py">--help</a></code>
+* <code title="cli">praisonai schedule restart <a href="./src/praisonai/praisonai/cli/commands/schedule.py">--help</a></code>
+* <code title="cli">praisonai schedule schedule-callback <a href="./src/praisonai/praisonai/cli/commands/schedule.py">--help</a></code>
+* <code title="cli">praisonai schedule start <a href="./src/praisonai/praisonai/cli/commands/schedule.py">--help</a></code>
+* <code title="cli">praisonai schedule stats <a href="./src/praisonai/praisonai/cli/commands/schedule.py">--help</a></code>
+* <code title="cli">praisonai schedule stop <a href="./src/praisonai/praisonai/cli/commands/schedule.py">--help</a></code>
+* <code title="cli">praisonai serve serve-callback <a href="./src/praisonai/praisonai/cli/commands/serve.py">--help</a></code>
+* <code title="cli">praisonai serve start <a href="./src/praisonai/praisonai/cli/commands/serve.py">--help</a></code>
+* <code title="cli">praisonai serve status <a href="./src/praisonai/praisonai/cli/commands/serve.py">--help</a></code>
+* <code title="cli">praisonai serve stop <a href="./src/praisonai/praisonai/cli/commands/serve.py">--help</a></code>
+* <code title="cli">praisonai session delete <a href="./src/praisonai/praisonai/cli/commands/session.py">--help</a></code>
+* <code title="cli">praisonai session export <a href="./src/praisonai/praisonai/cli/commands/session.py">--help</a></code>
+* <code title="cli">praisonai session import <a href="./src/praisonai/praisonai/cli/commands/session.py">--help</a></code>
+* <code title="cli">praisonai session list <a href="./src/praisonai/praisonai/cli/commands/session.py">--help</a></code>
+* <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>
+* <code title="cli">praisonai todo list <a href="./src/praisonai/praisonai/cli/commands/todo.py">--help</a></code>
+* <code title="cli">praisonai tools info <a href="./src/praisonai/praisonai/cli/commands/tools.py">--help</a></code>
+* <code title="cli">praisonai tools list <a href="./src/praisonai/praisonai/cli/commands/tools.py">--help</a></code>
+* <code title="cli">praisonai tools test <a href="./src/praisonai/praisonai/cli/commands/tools.py">--help</a></code>
+* <code title="cli">praisonai traces disable <a href="./src/praisonai/praisonai/cli/commands/traces.py">--help</a></code>
+* <code title="cli">praisonai traces enable <a href="./src/praisonai/praisonai/cli/commands/traces.py">--help</a></code>
+* <code title="cli">praisonai traces list <a href="./src/praisonai/praisonai/cli/commands/traces.py">--help</a></code>
+* <code title="cli">praisonai traces status <a href="./src/praisonai/praisonai/cli/commands/traces.py">--help</a></code>
+* <code title="cli">praisonai train train-main <a href="./src/praisonai/praisonai/cli/commands/train.py">--help</a></code>
+* <code title="cli">praisonai ui chat <a href="./src/praisonai/praisonai/cli/commands/ui.py">--help</a></code>
+* <code title="cli">praisonai ui code <a href="./src/praisonai/praisonai/cli/commands/ui.py">--help</a></code>
+* <code title="cli">praisonai ui gradio <a href="./src/praisonai/praisonai/cli/commands/ui.py">--help</a></code>
+* <code title="cli">praisonai ui realtime <a href="./src/praisonai/praisonai/cli/commands/ui.py">--help</a></code>
+* <code title="cli">praisonai ui ui-main <a href="./src/praisonai/praisonai/cli/commands/ui.py">--help</a></code>
+* <code title="cli">praisonai version check <a href="./src/praisonai/praisonai/cli/commands/version.py">--help</a></code>
+* <code title="cli">praisonai version show <a href="./src/praisonai/praisonai/cli/commands/version.py">--help</a></code>
+* <code title="cli">praisonai version version-callback <a href="./src/praisonai/praisonai/cli/commands/version.py">--help</a></code>
+* <code title="cli">praisonai workflow create <a href="./src/praisonai/praisonai/cli/commands/workflow.py">--help</a></code>
+* <code title="cli">praisonai workflow list <a href="./src/praisonai/praisonai/cli/commands/workflow.py">--help</a></code>
+* <code title="cli">praisonai workflow run <a href="./src/praisonai/praisonai/cli/commands/workflow.py">--help</a></code>
+
+# TypeScript
+
+Types/Exports:
+```ts
+export { Agent, Agents, PraisonAIAgents, Router } from "./agent";
+export type { PraisonAIAgentsConfig, SimpleAgentConfig, SimpleRouteConfig, SimpleRouterConfig } from "./agent";
+export { ContextAgent, createContextAgent } from "./agent/context";
+export { Handoff, handoff, handoffFilters } from "./agent/handoff";
+export { ImageAgent, createImageAgent } from "./agent/image";
+export { PromptExpanderAgent, createPromptExpanderAgent } from "./agent/prompt-expander";
+export { QueryRewriterAgent, createQueryRewriterAgent } from "./agent/query-rewriter";
+export { DeepResearchAgent, createDeepResearchAgent } from "./agent/research";
+export { RouterAgent, createRouter, routeConditions } from "./agent/router";
+export { // Agent loop
+  createAgentLoop, // DevTools
+  enableDevTools, // MCP
+  createMCP, // Middleware (renamed to avoid conflicts)
+  createCachingMiddleware, // Models
+  createModel, // Multimodal
+  createImagePart, // Next.js
+  createRouteHandler, // OAuth for MCP
+  type OAuthClientProvider, // Server adapters
+  createHttpHandler, // Speech & Transcription
+  generateSpeech, // Telemetry (AI SDK v6 parity)
+  configureTelemetry, // Tool Approval (AI SDK v6 parity)
+  ApprovalManager, // Tools
+  defineTool, // UI Message (AI SDK v6 parity)
+  convertToModelMessages, AIAgentStep, AIEmbedManyResult, AIEmbedOptions, AIEmbedResult, AIFilePart, AIGenerateImageOptions, AIGenerateImageResult, AIGenerateObjectOptions, AIGenerateObjectResult, AIGenerateTextOptions, AIGenerateTextResult, AIImagePart, AIMiddleware, AIMiddlewareConfig, AIModelMessage, AISpan, AISpanKind, AISpanOptions, AISpanStatus, AIStreamObjectOptions, AIStreamObjectResult, AIStreamTextOptions, AIStreamTextResult, AITelemetryEvent, AITelemetrySettings, AITextPart, AIToolDefinition, AITracer, AgentLoop, DANGEROUS_PATTERNS, MODEL_ALIASES, SPEECH_MODELS, TRANSCRIPTION_MODELS, ToolApprovalDeniedError, ToolApprovalTimeoutError, aiEmbed, aiEmbedMany, aiGenerateImage, aiGenerateObject, aiGenerateText, aiStreamObject, aiStreamText, applyMiddleware, autoEnableDevTools, base64ToUint8Array, clearAICache, clearEvents, closeAllMCPClients, closeMCPClient, convertToUIMessages, createAILoggingMiddleware, createAISpan, createApprovalResponse, createDangerousPatternChecker, createDevToolsMiddleware, createExpressHandler, createFastifyHandler, createFilePart, createHonoHandler, createMultimodalMessage, createNestHandler, createPagesHandler, createPdfPart, createSystemMessage, createTelemetryMiddleware, createTelemetrySettings, createTextMessage, createTextPart, createToolSet, disableAITelemetry, disableDevTools, enableAITelemetry, functionToTool, getAICacheStats, getApprovalManager, getDevToolsState, getDevToolsUrl, getEvents, getMCPClient, getModel, getTelemetrySettings, getToolsNeedingApproval, getTracer, hasModelAlias, hasPendingApprovals, initOpenTelemetry, isDangerous, isDataUrl, isDevToolsEnabled, isTelemetryEnabled, isUrl, listModelAliases, mcpToolsToAITools, parseModel, pipeUIMessageStreamToResponse, recordEvent, resolveModelAlias, safeValidateUIMessages, setApprovalManager, stopAfterSteps, stopWhen, stopWhenNoToolCalls, toMessageContent, toUIMessageStreamResponse, transcribe, uint8ArrayToBase64, validateUIMessages, withApproval, withSpan, wrapModel } from "./ai";
+export { AutoAgents, createAutoAgents } from "./auto";
+export { BaseCache, FileCache, MemoryCache, createFileCache, createMemoryCache } from "./cache";
+export { CLI_SPEC_VERSION, executeCommand, parseArgs } from "./cli";
+export { // Autonomy Mode
+  AutonomyManager, // Background Jobs
+  JobQueue, // Checkpoints
+  CheckpointManager, // Cost Tracker
+  CostTracker, // External Agents
+  BaseExternalAgent, // Fast Context
+  FastContext, // Flow Display
+  FlowDisplay, // Git Integration
+  GitManager, // Interactive TUI
+  InteractiveTUI, // N8N Integration
+  N8NIntegration, // Repo Map
+  RepoMap, // Sandbox Executor
+  SandboxExecutor, // Scheduler
+  Scheduler, // Slash Commands
+  SlashCommandHandler, AiderAgent, ClaudeCodeAgent, CodexCliAgent, CommandValidator, CostTokenUsage, DEFAULT_BLOCKED_COMMANDS, DEFAULT_BLOCKED_PATHS, DEFAULT_IGNORE_PATTERNS, DiffViewer, FileCheckpointStorage, FileJobStorage, GeminiCliAgent, GenericExternalAgent, HistoryManager, MODEL_PRICING, MODE_POLICIES, MemoryCheckpointStorage, MemoryJobStorage, StatusDisplay, cliApprovalPrompt, createAutonomyManager, createCheckpointManager, createCostTracker, createDiffViewer, createExternalAgent, createFastContext, createFileCheckpointStorage, createFileJobStorage, createFlowDisplay, createGitManager, createHistoryManager, createInteractiveTUI, createJobQueue, createN8NIntegration, createRepoMap, createSandboxExecutor, createScheduler, createSlashCommandHandler, createStatusDisplay, cronExpressions, estimateTokens, executeSlashCommand, externalAgentAsTool, formatCost, getExternalAgentRegistry, getQuickContext, getRepoTree, isSlashCommand, parseSlashCommand, registerCommand, renderWorkflow, sandboxExec, triggerN8NWebhook } from "./cli/features";
+export { createDbAdapter, db, getDefaultDbAdapter, setDefaultDbAdapter } from "./db";
+export type { DbAdapter, DbConfig, DbMessage, DbRun, DbTrace } from "./db";
+export { MemoryPostgresAdapter, NeonPostgresAdapter, PostgresSessionStorage, createMemoryPostgres, createNeonPostgres, createPostgresSessionStorage } from "./db/postgres";
+export { MemoryRedisAdapter, UpstashRedisAdapter, createMemoryRedis, createUpstashRedis } from "./db/redis";
+export { SQLiteAdapter, createSQLiteAdapter } from "./db/sqlite";
+export { EvalSuite, accuracyEval, performanceEval, reliabilityEval } from "./eval";
+export { AgentEventBus, AgentEvents, EventEmitterPubSub, PubSub, createEventBus, createPubSub } from "./events";
+export { LLMGuardrail, createLLMGuardrail } from "./guardrails/llm-guardrail";
+export { // Computer Use
+  createComputerUse, ComputerUseClient, createCLIApprovalPrompt, createComputerUseAgent } from "./integrations/computer-use";
+export { BaseObservabilityProvider, ConsoleObservabilityProvider, LangfuseObservabilityProvider, MemoryObservabilityProvider, ObservabilityTraceContext, createConsoleObservability, createLangfuseObservability, createMemoryObservability } from "./integrations/observability";
+export { // Natural Language Postgres
+  createNLPostgres, NLPostgresClient, NLPostgresConfig, createPostgresTool } from "./integrations/postgres";
+export { // Slack
+  createSlackBot, SlackBot, parseSlackMessage, verifySlackSignature } from "./integrations/slack";
+export { BaseVectorStore, ChromaVectorStore, MemoryVectorStore, PineconeVectorStore, QdrantVectorStore, VectorQueryResult, WeaviateVectorStore, createChromaStore, createMemoryVectorStore, createPineconeStore, createQdrantStore, createWeaviateStore } from "./integrations/vector";
+export { BaseVoiceProvider, ElevenLabsVoiceProvider, OpenAIVoiceProvider, createElevenLabsVoice, createOpenAIVoice } from "./integrations/voice";
+export { GraphRAG, GraphStore, createGraphRAG } from "./knowledge/graph-rag";
+export { BaseReranker, CohereReranker, CrossEncoderReranker, LLMReranker, createCohereReranker, createCrossEncoderReranker, createLLMReranker } from "./knowledge/reranker";
+export { // Provider classes
+  OpenAIProvider, // Provider factory and utilities
+  createProvider, // Provider registry (extensibility API)
+  ProviderRegistry, // Types
+  type LLMProvider, AnthropicProvider, BaseProvider, GoogleProvider, ProviderMessage, ProviderToolDefinition, createProviderRegistry, getAvailableProviders, getDefaultProvider, getDefaultRegistry, hasProvider, isProviderAvailable, listProviders, parseModelString, registerBuiltinProviders, registerProvider, unregisterProvider } from "./llm/providers";
+export { ADAPTERS, AISDK_PROVIDERS, COMMUNITY_PROVIDERS, PROVIDER_ALIASES } from "./llm/providers/ai-sdk/types";
+export { AutoMemory, AutoMemoryKnowledgeBase, AutoMemoryVectorStore, DEFAULT_POLICIES, createAutoMemory, createLLMSummarizer } from "./memory/auto-memory";
+export { FileMemory, createFileMemory } from "./memory/file-memory";
+export { Memory, createMemory } from "./memory/memory";
+export type { MemoryConfig, MemoryEntry } from "./memory/memory";
+export { // Adapters
+  NoopObservabilityAdapter, // Constants
+  OBSERVABILITY_TOOLS, // Global adapter management
+  setObservabilityAdapter, // Types
+  type SpanKind, ConsoleObservabilityAdapter, MemoryObservabilityAdapter, clearAdapterCache, createConsoleAdapter, createMemoryAdapter, createObservabilityAdapter, getObservabilityAdapter, getObservabilityToolInfo, hasObservabilityToolEnvVar, listObservabilityTools, noopAdapter, resetObservabilityAdapter, trace } from "./observability";
+export { Plan, PlanStep, PlanStorage, PlanningAgent, TaskAgent, TodoItem, TodoList, createPlan, createPlanStorage, createPlanningAgent, createTaskAgent, createTodoList } from "./planning";
+export { SkillManager, createSkillManager, parseSkillFile } from "./skills";
+export { AgentTelemetry, TelemetryCollector, cleanupTelemetry, createAgentTelemetry, disableTelemetry, enableTelemetry, getTelemetry } from "./telemetry";
+export { BaseTool, FunctionTool, ToolRegistry, ToolResult, ToolValidationError, createTool, getRegistry, getTool, registerTool, tool, validateTool } from "./tools";
+export { airweaveSearch, bedrockBrowserClick, bedrockBrowserFill, bedrockBrowserNavigate, bedrockCodeInterpreter, codeExecution, codeMode, createCustomTool, exaSearch, firecrawlCrawl, firecrawlScrape, parallelSearch, perplexitySearch, registerCustomTool, registerLocalTool, registerNpmTool, superagentGuard, superagentRedact, superagentVerify, tavilyCrawl, tavilyExtract, tavilySearch, valyuBioSearch, valyuCompanyResearch, valyuEconomicsSearch, valyuFinanceSearch, valyuPaperSearch, valyuPatentSearch, valyuSecSearch, valyuWebSearch } from "./tools/builtins";
+export { MissingDependencyError, MissingEnvVarError, ToolsRegistry, composeMiddleware, createLoggingMiddleware, createRateLimitMiddleware, createRedactionMiddleware, createRetryMiddleware, createTimeoutMiddleware, createToolsRegistry, createTracingMiddleware, createValidationMiddleware, getToolsRegistry, resetToolsRegistry } from "./tools/registry";
+export type { InstallHints, PraisonTool, RedactionHooks, RegisteredTool, ToolCapabilities, ToolExecutionContext, ToolExecutionResult, ToolFactory, ToolHooks, ToolInstallStatus, ToolLimits, ToolLogger, ToolMetadata, ToolMiddleware, ToolParameterProperty, ToolParameterSchema } from "./tools/registry";
+export { registerBuiltinTools, tools } from "./tools/tools";
+export { Workflow, loop, parallel, repeat, route } from "./workflows";
+export type { StepResult, WorkflowContext, WorkflowStep } from "./workflows";
+export { createWorkflowFromYAML, loadWorkflowFromFile, parseYAMLWorkflow, validateWorkflowDefinition } from "./workflows/yaml-parser";
+```
+
+# Optional Plugins
+
+External tools are available via `praisonai-tools` package:
+
+```bash
+pip install praisonai-tools
+```
+
+See [PraisonAI-tools](https://github.com/MervinPraison/PraisonAI-tools) for available tools.

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>=3.6.2" \
+    "praisonai>=3.9.2" \
     "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>=3.6.2" \
+    "praisonai>=3.9.2" \
     "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>=3.6.2" \
+    "praisonai>=3.9.2" \
     "praisonai[ui]" \
     "praisonai[crewai]"
 

examples/consolidated_params/README.md

@@ -0,0 +1,96 @@
+# Consolidated Params Examples
+
+Examples demonstrating the **Agent-Centric API** with consolidated parameters.
+
+## Precedence Ladder
+
+```
+Instance > Config > Array > Dict > String > Bool > Default
+```
+
+## Quick Reference
+
+| Feature | Presets | Example |
+|---------|---------|---------|
+| `memory` | file, sqlite, redis, postgres, mem0, mongodb | `memory="sqlite"` |
+| `output` | minimal, normal, verbose, debug, silent | `output="verbose"` |
+| `execution` | fast, balanced, thorough, unlimited | `execution="balanced"` |
+| `planning` | reasoning, read_only, auto | `planning="reasoning"` |
+| `reflection` | minimal, standard, thorough | `reflection="standard"` |
+| `guardrails` | strict, permissive, safety | `guardrails="strict"` |
+| `web` | duckduckgo, tavily, google, bing, serper | `web="duckduckgo"` |
+| `context` | sliding_window, summarize, truncate | `context="sliding_window"` |
+| `autonomy` | suggest, auto_edit, full_auto | `autonomy="suggest"` |
+| `caching` | enabled, disabled, prompt | `caching="prompt"` |
+
+## Examples
+
+### Basic Examples (Single Feature)
+
+| File | Feature | Description |
+|------|---------|-------------|
+| `basic_agent.py` | Agent | Minimal agent with memory |
+| `basic_memory.py` | memory | SQLite memory preset |
+| `basic_output.py` | output | Verbose output preset |
+| `basic_execution.py` | execution | Execution control |
+| `basic_guardrails.py` | guardrails | Callable validator |
+| `basic_reflection.py` | reflection | Self-reflection preset |
+| `basic_web.py` | web | Web search preset |
+| `basic_planning.py` | planning | Planning preset |
+| `basic_autonomy.py` | autonomy | Autonomy preset |
+| `basic_caching.py` | caching | Caching preset |
+| `basic_knowledge.py` | knowledge | RAG with sources |
+| `basic_context.py` | context | Context management |
+| `basic_hooks.py` | hooks | Lifecycle callbacks |
+
+### Multi-Agent Examples
+
+| File | Description |
+|------|-------------|
+| `basic_agents.py` | Multi-agent with memory and planning |
+
+### Workflow Examples
+
+| File | Description |
+|------|-------------|
+| `basic_workflow.py` | Simple workflow with output preset |
+| `basic_workflow_agentlike.py` | Workflow with agent-like params |
+| `basic_step_override.py` | Step-level param overrides |
+| `advanced_workflow_full_features.py` | All consolidated params |
+
+### Advanced Examples
+
+| File | Description |
+|------|-------------|
+| `advanced_output_execution.py` | Combined output + execution |
+
+## Usage Forms
+
+Each consolidated param supports multiple input forms:
+
+```python
+# Bool - enable with defaults
+memory=True
+
+# String preset
+memory="sqlite"
+
+# Dict config
+memory={"backend": "sqlite", "user_id": "user123"}
+
+# Config instance
+memory=MemoryConfig(backend="sqlite", user_id="user123")
+
+# Array with overrides (for some params)
+guardrails=["strict", {"max_retries": 10}]
+```
+
+## Running Examples
+
+```bash
+# Set API key
+export OPENAI_API_KEY=your_key_here
+
+# Run any example
+python basic_agent.py
+```

examples/consolidated_params/basic_autonomy.py

@@ -0,0 +1,18 @@
+"""
+Basic Autonomy Example - Agent-Centric API
+
+Demonstrates autonomy with consolidated params.
+Presets: suggest, auto_edit, full_auto
+"""
+
+from praisonaiagents import Agent
+
+# Basic: Enable autonomy with preset
+agent = Agent(
+    instructions="You are an autonomous assistant.",
+    autonomy="suggest",  # Presets: suggest, auto_edit, full_auto
+)
+
+if __name__ == "__main__":
+    response = agent.start("Help me organize my project files.")
+    print(response)

examples/consolidated_params/basic_caching.py

@@ -0,0 +1,23 @@
+"""
+Basic Caching Example - Agent-Centric API
+
+Demonstrates caching with consolidated params.
+Presets: enabled, disabled, prompt
+"""
+
+from praisonaiagents import Agent
+
+# Basic: Enable caching with preset
+agent = Agent(
+    instructions="You are a helpful assistant with caching enabled.",
+    caching="prompt",  # Presets: enabled, disabled, prompt
+)
+
+if __name__ == "__main__":
+    # First call - will cache
+    response = agent.start("What is the capital of France?")
+    print(response)
+    
+    # Second call - may use cache
+    response = agent.start("What is the capital of France?")
+    print(response)

examples/consolidated_params/basic_context.py

@@ -0,0 +1,18 @@
+"""
+Basic Context Example - Agent-Centric API
+
+Demonstrates context management with consolidated params.
+Presets: sliding_window, summarize, truncate
+"""
+
+from praisonaiagents import Agent
+
+# Basic: Enable context management with preset
+agent = Agent(
+    instructions="You are a helpful assistant with context management.",
+    context="sliding_window",  # Presets: sliding_window, summarize, truncate
+)
+
+if __name__ == "__main__":
+    response = agent.start("Tell me a long story about a dragon.")
+    print(response)

examples/consolidated_params/basic_hooks.py

@@ -0,0 +1,28 @@
+"""
+Basic Hooks Example - Agent-Centric API
+
+Demonstrates lifecycle hooks with consolidated params.
+Valid keys: on_step, on_tool_call, middleware
+"""
+
+from praisonaiagents import Agent
+
+# Define hook callbacks
+def on_step_callback(step_info):
+    print(f"Step: {step_info}")
+
+def on_tool_call_callback(tool_info):
+    print(f"Tool called: {tool_info}")
+
+# Basic: Enable hooks with dict
+agent = Agent(
+    instructions="You are a helpful assistant.",
+    hooks={
+        "on_step": on_step_callback,
+        "on_tool_call": on_tool_call_callback,
+    },
+)
+
+if __name__ == "__main__":
+    response = agent.start("What is 2 + 2?")
+    print(response)

examples/consolidated_params/basic_knowledge.py

@@ -0,0 +1,18 @@
+"""
+Basic Knowledge (RAG) Example - Agent-Centric API
+
+Demonstrates knowledge with consolidated params.
+Supports: bool, list of sources, dict config, KnowledgeConfig
+"""
+
+from praisonaiagents import Agent
+
+# Basic: Enable knowledge with list of sources
+agent = Agent(
+    instructions="You are a helpful assistant with document knowledge.",
+    knowledge=["docs/"],  # List of file/folder paths
+)
+
+if __name__ == "__main__":
+    response = agent.start("What does the documentation say about installation?")
+    print(response)

examples/consolidated_params/basic_planning.py

@@ -0,0 +1,18 @@
+"""
+Basic Planning Example - Agent-Centric API
+
+Demonstrates planning with consolidated params.
+Presets: reasoning, read_only, auto
+"""
+
+from praisonaiagents import Agent
+
+# Basic: Enable planning with preset
+agent = Agent(
+    instructions="You are a strategic planner.",
+    planning="reasoning",  # Presets: reasoning, read_only, auto
+)
+
+if __name__ == "__main__":
+    response = agent.start("Plan a 3-day trip to Tokyo.")
+    print(response)

examples/consolidated_params/basic_reflection.py

@@ -0,0 +1,18 @@
+"""
+Basic Reflection Example - Agent-Centric API
+
+Demonstrates reflection (self-reflect) with consolidated params.
+Presets: minimal, standard, thorough
+"""
+
+from praisonaiagents import Agent
+
+# Basic: Enable reflection with preset
+agent = Agent(
+    instructions="You are a helpful assistant that reflects on your answers.",
+    reflection="standard",  # Presets: minimal, standard, thorough
+)
+
+if __name__ == "__main__":
+    response = agent.start("Explain quantum computing in simple terms.")
+    print(response)

examples/consolidated_params/basic_web.py

@@ -0,0 +1,18 @@
+"""
+Basic Web Search Example - Agent-Centric API
+
+Demonstrates web search with consolidated params.
+Presets: duckduckgo, tavily, google, bing, serper, search_only, fetch_only
+"""
+
+from praisonaiagents import Agent
+
+# Basic: Enable web search with preset
+agent = Agent(
+    instructions="You are a research assistant with web access.",
+    web="duckduckgo",  # Presets: duckduckgo, tavily, google, bing, serper
+)
+
+if __name__ == "__main__":
+    response = agent.start("What are the latest AI developments in 2024?")
+    print(response)

examples/cookbooks/yaml/secondary_market_research_agents.yaml

@@ -22,7 +22,7 @@ agents:  # Canonical: use 'agents' instead of 'roles'
     verbose: true
     allow_delegation: false
     llm:
-      model: "gpt-5-nano"
+      model: "gpt-4o-mini"
       temperature: 0.3
     tools:
       - "InternetSearchTool"
@@ -76,7 +76,7 @@ agents:  # Canonical: use 'agents' instead of 'roles'
     verbose: true
     allow_delegation: false
     llm:
-      model: "gpt-5-nano" 
+      model: "gpt-4o-mini" 
       temperature: 0.3
     tools:
       - "InternetSearchTool"
@@ -130,7 +130,7 @@ agents:  # Canonical: use 'agents' instead of 'roles'
     verbose: true
     allow_delegation: false
     llm:
-      model: "gpt-5-nano"
+      model: "gpt-4o-mini"
       temperature: 0.3
     tools:
       - "InternetSearchTool"
@@ -185,7 +185,7 @@ agents:  # Canonical: use 'agents' instead of 'roles'
     verbose: true
     allow_delegation: false
     llm:
-      model: "gpt-5-nano"
+      model: "gpt-4o-mini"
       temperature: 0.4
     tools:
       - "InternetSearchTool"
@@ -239,7 +239,7 @@ agents:  # Canonical: use 'agents' instead of 'roles'
     verbose: true
     allow_delegation: false
     llm:
-      model: "gpt-5-nano"
+      model: "gpt-4o-mini"
       temperature: 0.3
     tools:
       - "InternetSearchTool"
@@ -299,7 +299,7 @@ agents:  # Canonical: use 'agents' instead of 'roles'
     verbose: true
     allow_delegation: false
     llm:
-      model: "gpt-5-nano"
+      model: "gpt-4o-mini"
       temperature: 0.3
     tasks:
       report_synthesis:

examples/mcp/multi_server.py

@@ -0,0 +1,148 @@
+"""
+Multi-Server MCP Example.
+
+This example demonstrates how to use multiple MCP servers with PraisonAI Agents,
+including mixing MCP tools with regular Python function tools.
+
+Requirements:
+    - Node.js and npx installed (or uvx for Python-based servers)
+    - pip install praisonaiagents[mcp]
+    - OPENAI_API_KEY environment variable set
+
+Usage:
+    python multi_server.py
+"""
+
+from praisonaiagents import Agent, MCP
+import os
+
+
+def calculate(expression: str) -> str:
+    """
+    A simple calculator tool that evaluates mathematical expressions.
+    
+    Args:
+        expression: A mathematical expression to evaluate (e.g., "2 + 2")
+    
+    Returns:
+        The result of the calculation
+    """
+    try:
+        # Safe evaluation of mathematical expressions
+        allowed_chars = set("0123456789+-*/.() ")
+        if not all(c in allowed_chars for c in expression):
+            return "Error: Invalid characters in expression"
+        result = eval(expression)
+        return f"Result: {result}"
+    except Exception as e:
+        return f"Error: {str(e)}"
+
+
+def main():
+    # Check for API key
+    if not os.environ.get("OPENAI_API_KEY"):
+        print("Please set OPENAI_API_KEY environment variable")
+        return
+    
+    print("=" * 60)
+    print("Multi-Server MCP Example")
+    print("=" * 60)
+    
+    # Create MCP instance with Time server
+    print("\nInitializing MCP servers...")
+    try:
+        time_mcp = MCP("uvx mcp-server-time", timeout=30)
+        print("✓ Time MCP server initialized")
+    except Exception as e:
+        print(f"✗ Time MCP server failed: {e}")
+        return
+    
+    # Create an agent with multiple tool sources
+    agent = Agent(
+        name="MultiToolAgent",
+        instructions="""You are a helpful assistant with access to multiple tools:
+        1. Time tools - for getting current time in different timezones
+        2. Calculator - for mathematical calculations
+        
+        Use the appropriate tool based on the user's request.""",
+        llm="openai/gpt-4o-mini",
+        tools=[time_mcp, calculate]  # Mix MCP and regular tools
+    )
+    
+    # Show available tools
+    print("\nAvailable tools:")
+    formatted_tools = agent._format_tools_for_completion(agent.tools)
+    for tool in formatted_tools:
+        print(f"  - {tool['function']['name']}: {tool['function']['description'][:50]}...")
+    
+    # Example queries
+    queries = [
+        "What time is it in Tokyo?",
+        "Calculate 15 * 7 + 23",
+        "What's the current time in New York and London?"
+    ]
+    
+    print("\n" + "=" * 60)
+    print("Running queries...")
+    print("=" * 60)
+    
+    for i, query in enumerate(queries, 1):
+        print(f"\n--- Query {i} ---")
+        print(f"Question: {query}\n")
+        
+        try:
+            response = agent.chat(query)
+            print(f"Response:\n{response}")
+        except Exception as e:
+            print(f"Error: {e}")
+        
+        print("-" * 40)
+    
+    # Clean up
+    time_mcp.shutdown()
+    print("\nDone!")
+
+
+def demo_multiple_mcp_servers():
+    """
+    Demonstrate using multiple MCP servers simultaneously.
+    
+    This function shows how to create agents with multiple MCP servers,
+    each providing different capabilities.
+    """
+    print("\n" + "=" * 60)
+    print("Multiple MCP Servers Demo")
+    print("=" * 60)
+    
+    # Initialize multiple MCP servers
+    servers = {}
+    
+    try:
+        servers['time'] = MCP("uvx mcp-server-time", timeout=30)
+        print("✓ Time server initialized")
+    except Exception as e:
+        print(f"✗ Time server failed: {e}")
+    
+    # Create agent with all available servers
+    available_tools = list(servers.values())
+    
+    if not available_tools:
+        print("No MCP servers available")
+        return
+    
+    agent = Agent(
+        name="MultiServerAgent",
+        instructions="You have access to multiple MCP servers. Use the appropriate tools.",
+        llm="openai/gpt-4o-mini",
+        tools=available_tools
+    )
+    
+    print(f"\nAgent created with {len(available_tools)} MCP server(s)")
+    
+    # Clean up
+    for server in servers.values():
+        server.shutdown()
+
+
+if __name__ == "__main__":
+    main()

examples/mcp/sequential_thinking.py

@@ -0,0 +1,70 @@
+"""
+Sequential Thinking MCP Example.
+
+This example demonstrates how to use the Sequential Thinking MCP server
+with PraisonAI Agents to break down complex problems step by step.
+
+Requirements:
+    - Node.js and npx installed
+    - pip install praisonaiagents[mcp]
+    - OPENAI_API_KEY environment variable set
+
+Usage:
+    python sequential_thinking.py
+"""
+
+from praisonaiagents import Agent, MCP
+import os
+
+
+def main():
+    # Check for API key
+    if not os.environ.get("OPENAI_API_KEY"):
+        print("Please set OPENAI_API_KEY environment variable")
+        return
+    
+    # Create MCP instance with Sequential Thinking server
+    # The server provides tools for step-by-step problem analysis
+    sequential_mcp = MCP(
+        "npx -y @modelcontextprotocol/server-sequential-thinking",
+        timeout=60
+    )
+    
+    # Create an agent with the Sequential Thinking tools
+    agent = Agent(
+        name="ProblemSolver",
+        instructions="""You are an expert problem solver who breaks down complex problems 
+        into manageable steps. When given a problem:
+        1. Use the sequential thinking tool to analyze it step by step
+        2. Provide clear, actionable steps
+        3. Explain your reasoning at each step""",
+        llm="openai/gpt-4o-mini",
+        tools=sequential_mcp
+    )
+    
+    # Example problems to solve
+    problems = [
+        "Break down the process of making a cup of tea",
+        "How would you plan a weekend trip to a new city?",
+        "What steps would you take to learn a new programming language?"
+    ]
+    
+    print("=" * 60)
+    print("Sequential Thinking MCP Example")
+    print("=" * 60)
+    
+    for i, problem in enumerate(problems, 1):
+        print(f"\n--- Problem {i} ---")
+        print(f"Question: {problem}\n")
+        
+        response = agent.chat(problem)
+        print(f"Response:\n{response}")
+        print("-" * 40)
+    
+    # Clean up
+    sequential_mcp.shutdown()
+    print("\nDone!")
+
+
+if __name__ == "__main__":
+    main()

examples/python/agents/MULTI_PROVIDER_README.md

@@ -46,7 +46,7 @@ agent = RouterAgent(
     name="Smart Assistant",
     role="Adaptive AI Assistant",
     goal="Complete tasks using the most appropriate model",
-    models=["gpt-5-nano", "gpt-5-nano", "claude-3-5-sonnet-20241022"],
+    models=["gpt-4o-mini", "gpt-4o-mini", "claude-3-5-sonnet-20241022"],
     routing_strategy="auto"  # Automatic model selection
 )
 
@@ -90,7 +90,7 @@ analyzer = RouterAgent(
     goal="Analyze data efficiently",
     models={
         "gemini/gemini-1.5-flash": {},
-        "gpt-5-nano": {},
+        "gpt-4o-mini": {},
         "claude-3-haiku-20240307": {}
     },
     model_router=router,
@@ -207,7 +207,7 @@ for i, agent in enumerate(auto.agents):
         name=agent.name,
         role=agent.role,
         goal=agent.goal,
-        models=["gpt-5-nano", "gpt-5-nano", "claude-3-5-sonnet"],
+        models=["gpt-4o-mini", "gpt-4o-mini", "claude-3-5-sonnet"],
         routing_strategy="auto"
     )
 

examples/python/memory/gemini-embeddings.md

@@ -97,7 +97,7 @@ researcher = Agent(
     name="Researcher",
     role="Information Researcher",
     goal="Store information with semantic understanding",
-    llm="gpt-5-nano"
+    llm="gpt-4o-mini"
 )
 
 # Create task

examples/python/monitoring/README.md

@@ -17,7 +17,7 @@ def my_function():
 
 agent = Agent(
     instructions="You are a helpful assistant",
-    llm="gpt-5-nano"
+    llm="gpt-4o-mini"
 )
 
 with track_api_call("agent_request"):

examples/python/token-metrics/README.md

@@ -13,7 +13,7 @@ agent = Agent(
     role="Assistant", 
     goal="Help users",
     backstory="You are helpful.",
-    llm="gpt-5-nano",
+    llm="gpt-4o-mini",
     metrics=True  # 🎯 That's it!
 )
 

examples/python/usecases/README-secondary-market-research.md

@@ -177,7 +177,7 @@ Set your API keys:
 
 ```bash
 export OPENAI_API_KEY="your-api-key"
-export OPENAI_MODEL_NAME="gpt-5-nano"
+export OPENAI_MODEL_NAME="gpt-4o-mini"
 ```
 
 ## 📝 Customization

src/praisonai-agents/01_basic_agent_with_monitoring.py

@@ -38,7 +38,7 @@ def main():
     # Create agent
     agent = Agent(
         instructions="You are a helpful assistant that explains scientific concepts simply",
-        llm="gpt-5-nano"
+        llm="gpt-4o-mini"
     )
     
     # Track the API call performance

src/praisonai-agents/01_basic_token_tracking.py

@@ -13,8 +13,8 @@ def main():
         role="Research Specialist", 
         goal="Find information about a topic",
         backstory="You are an expert researcher who finds comprehensive information.",
-        verbose=True,
-        llm="gpt-5-nano",  # Use a cost-effective model for examples
+        output="verbose",
+        llm="gpt-4o-mini",  # 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

@@ -54,7 +54,7 @@ def create_agents():
         goal="Gather comprehensive information on renewable energy",
         backstory="Expert researcher with deep knowledge of energy sector",
         instructions="You are a thorough researcher. Provide detailed, fact-based information.",
-        llm="gpt-5-nano"
+        llm="gpt-4o-mini"
     )
     
     analyst = Agent(
@@ -63,7 +63,7 @@ def create_agents():
         goal="Analyze trends and patterns in renewable energy data",
         backstory="Experienced data analyst specializing in energy markets",
         instructions="You are an analytical thinker. Focus on trends, patterns, and insights.",
-        llm="gpt-5-nano"
+        llm="gpt-4o-mini"
     )
     
     writer = Agent(
@@ -72,7 +72,7 @@ def create_agents():
         goal="Create clear, compelling reports",
         backstory="Professional technical writer with expertise in energy sector",
         instructions="You are a skilled writer. Create clear, well-structured reports.",
-        llm="gpt-5-nano"
+        llm="gpt-4o-mini"
     )
     
     return researcher, analyst, writer
@@ -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
@@ -48,8 +48,8 @@ def main():
         instructions="""You are a research assistant that helps find information.
         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
+        llm="gpt-4o-mini",
+        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/04_async_agent_monitoring.py

@@ -54,7 +54,7 @@ async def run_async_agent(query: str):
     # Create agent
     agent = Agent(
         instructions="You are an efficient AI assistant that processes requests quickly",
-        llm="gpt-5-nano"
+        llm="gpt-4o-mini"
     )
     
     # Track the agent execution

src/praisonai-agents/05_error_handling_monitoring.py

@@ -60,7 +60,7 @@ def run_robust_agent(query: str, with_error_simulation: bool = True):
         instructions="""You are a resilient AI assistant. 
         Even when facing challenges, you provide helpful responses.
         If you encounter errors, acknowledge them and offer alternatives.""",
-        llm="gpt-5-nano"
+        llm="gpt-4o-mini"
     )
     
     try:

src/praisonai-agents/06_memory_agent_monitoring.py

@@ -65,7 +65,7 @@ def run_stateful_agent(query: str, session_id: str, conversation_turn: int):
         instructions="""You are a memory-enabled AI assistant. You can remember 
         previous conversations and build on past interactions. Use your memory 
         to provide personalized and contextual responses.""",
-        llm="gpt-5-nano"
+        llm="gpt-4o-mini"
     )
     
     # Pre-processing: retrieve relevant memories
@@ -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/07_hierarchical_agents_monitoring.py

@@ -77,7 +77,7 @@ def run_hierarchical_workflow():
         backstory="Experienced project manager with expertise in team coordination",
         instructions="""You are a project manager. Break down complex projects into tasks,
         assign them to appropriate team members, and coordinate the overall effort.""",
-        llm="gpt-5-nano"
+        llm="gpt-4o-mini"
     )
     
     # Create worker agents
@@ -87,7 +87,7 @@ def run_hierarchical_workflow():
         goal="Conduct thorough research on assigned topics",
         backstory="Expert researcher with deep analytical skills",
         instructions="You are a research specialist. Provide detailed, factual information.",
-        llm="gpt-5-nano"
+        llm="gpt-4o-mini"
     )
     
     analysis_worker = Agent(
@@ -96,7 +96,7 @@ def run_hierarchical_workflow():
         goal="Analyze data and identify patterns and insights",
         backstory="Experienced data analyst with strong analytical thinking",
         instructions="You are a data analyst. Focus on patterns, trends, and insights.",
-        llm="gpt-5-nano"
+        llm="gpt-4o-mini"
     )
     
     writing_worker = Agent(
@@ -105,7 +105,7 @@ def run_hierarchical_workflow():
         goal="Create clear, well-structured documentation and reports", 
         backstory="Professional writer specializing in technical content",
         instructions="You are a technical writer. Create clear, comprehensive reports.",
-        llm="gpt-5-nano"
+        llm="gpt-4o-mini"
     )
     
     # Project tasks

src/praisonai-agents/08_custom_tools_monitoring.py

@@ -180,7 +180,7 @@ def run_agent_with_custom_tools():
         instructions="""You are a technical assistant with access to various tools.
         Use the appropriate tools to help with data processing, API calls, database queries,
         and file operations. Always explain what tools you're using and why.""",
-        llm="gpt-5-nano",
+        llm="gpt-4o-mini",
         tools=[process_data, call_external_api, query_database, file_operation]
     )
     
@@ -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/09_streaming_monitoring.py

@@ -60,7 +60,7 @@ def get_real_time_agent_response(query: str, response_id: int):
     # Create a lightweight agent for streaming
     agent = Agent(
         instructions="You are a real-time assistant. Provide quick, helpful responses.",
-        llm="gpt-5-nano"
+        llm="gpt-4o-mini"
     )
     
     with track_api_call(f"streaming_response_{response_id}"):

src/praisonai-agents/10_comprehensive_dashboard.py

@@ -42,19 +42,19 @@ def collect_comprehensive_data():
     agents = {
         'quick_agent': Agent(
             instructions="You are a quick-response agent. Provide concise answers.",
-            llm="gpt-5-nano"
+            llm="gpt-4o-mini"
         ),
         'analytical_agent': Agent(
             instructions="You are an analytical agent. Provide detailed analysis.",
-            llm="gpt-5-nano"
+            llm="gpt-4o-mini"
         ),
         'research_agent': Agent(
             instructions="You are a research agent. Provide comprehensive information.",
-            llm="gpt-5-nano"
+            llm="gpt-4o-mini"
         ),
         'creative_agent': Agent(
             instructions="You are a creative agent. Provide imaginative responses.",
-            llm="gpt-5-nano"
+            llm="gpt-4o-mini"
         )
     }
     
@@ -98,19 +98,19 @@ def simulate_complex_workflow():
     planner = Agent(
         name="planner",
         instructions="You are a project planner. Break down tasks and plan execution.",
-        llm="gpt-5-nano"
+        llm="gpt-4o-mini"
     )
     
     executor = Agent(
         name="executor", 
         instructions="You are a task executor. Complete assigned tasks efficiently.",
-        llm="gpt-5-nano"
+        llm="gpt-4o-mini"
     )
     
     reviewer = Agent(
         name="reviewer",
         instructions="You are a quality reviewer. Review and improve work.",
-        llm="gpt-5-nano"
+        llm="gpt-4o-mini"
     )
     
     # Create workflow tasks
@@ -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/CLAUDE.md

@@ -1,279 +0,0 @@
-# CLAUDE.md
-
-This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
-
-## Project Overview
-
-PraisonAI Agents is a hierarchical AI agent framework for completing complex tasks with self-reflection capabilities. It supports multi-agent collaboration, tool integration, and various execution patterns (sequential, hierarchical, parallel).
-
-## Development Commands
-
-### Installation and Setup
-```bash
-# Install core package
-pip install -e .
-
-# Install with specific features
-pip install -e .[all]          # All features
-pip install -e .[memory]       # Memory capabilities
-pip install -e .[knowledge]    # Document processing
-pip install -e .[mcp]          # MCP server support
-pip install -e .[llm]          # Extended LLM support
-pip install -e .[api]          # API server capabilities
-```
-
-### Testing
-```bash
-# Run individual test examples (no formal test runner configured)
-python tests/basic-agents.py
-python tests/async_example.py
-python tests/knowledge-agents.py
-
-# Test specific features
-python tests/mcp-agents.py           # MCP integration
-python tests/memory_example.py      # Memory functionality
-python tests/tools_example.py       # Tool system
-```
-
-### Running Examples
-```bash
-# Basic agent usage
-python tests/single-agent.py
-
-# Multi-agent workflows
-python tests/multi-agents-api.py
-
-# Async operations
-python tests/async_example_full.py
-
-# MCP server examples
-python tests/mcp-sse-direct-server.py  # Start MCP server
-python tests/mcp-sse-direct-client.py  # Connect to server
-```
-
-## Core Architecture
-
-### Agent System (`praisonaiagents/agent/`)
-- **Agent**: Core agent class with LLM integration, tool calling, and self-reflection
-- **ImageAgent**: Specialized multimodal agent for image processing
-- Self-reflection with configurable min/max iterations (default: 1-3)
-- Delegation support for hierarchical agent structures
-
-### Multi-Agent Orchestration (`praisonaiagents/agents/`)
-- **PraisonAIAgents**: Main orchestrator for managing multiple agents and tasks
-- **AutoAgents**: Automatic agent creation and management
-- Process types: `sequential`, `hierarchical`, `parallel`
-- Context passing between agents and task dependency management
-
-### Task System (`praisonaiagents/task/`)
-- **Task**: Core task definition with context, callbacks, and output specifications
-- Supports file output, JSON/Pydantic structured output, async execution
-- Conditional logic with `condition` parameter for task flow control
-- Context passing via `context` parameter for task dependencies
-- **Guardrails**: Built-in validation and safety mechanisms for task outputs
-  - Function-based guardrails for custom validation logic
-  - LLM-based guardrails using natural language descriptions
-  - Automatic retry with configurable `max_retries` parameter
-  - Compatible with CrewAI guardrail patterns
-
-### LLM Integration (`praisonaiagents/llm/`)
-- Unified wrapper for multiple LLM providers via LiteLLM
-- Supports OpenAI, Anthropic, Gemini, DeepSeek, local models (Ollama)
-- Context length management and tool calling capabilities
-- Set via `llm` parameter on agents or global `OPENAI_API_KEY`/`ANTHROPIC_API_KEY`
-
-### Tool System (`praisonaiagents/tools/`)
-Two implementation patterns:
-1. **Function-based**: Simple tools using `@tool` decorator
-2. **Class-based**: Complex tools inheriting from `BaseTool`
-
-Built-in tools include: DuckDuckGo search, file operations, calculator, Wikipedia, arXiv, data analysis tools, shell execution.
-
-### Memory & Knowledge Systems
-- **Memory** (`praisonaiagents/memory/`): Multi-layered memory with RAG support
-  - Types: short-term, long-term, entity, user memory
-  - Providers: ChromaDB, Mem0, custom implementations
-- **Knowledge** (`praisonaiagents/knowledge/`): Document processing with chunking
-  - Chunking strategies via `chonkie` library
-  - Embedding and retrieval capabilities
-
-### MCP (Model Context Protocol) Integration
-- **MCP Server**: Server-side tool protocol for distributed execution
-- **SSE Support**: Server-sent events for real-time communication
-- Tool discovery and dynamic registration
-
-## Development Patterns
-
-### Agent Creation
-```python
-agent = Agent(
-    name="Agent Name",
-    role="Agent Role",
-    goal="Agent Goal",
-    backstory="Agent Background",
-    llm="gpt-5-nano",  # or other LLM
-    self_reflect=True,  # Enable self-reflection
-    min_reflect=1,      # Minimum reflection iterations
-    max_reflect=3,      # Maximum reflection iterations
-    tools=[tool1, tool2],  # Optional tools
-    guardrail=validate_function,  # Agent-level guardrail (function or string)
-    max_guardrail_retries=3  # Retry limit for guardrail failures
-)
-```
-
-### Task Definition
-```python
-task = Task(
-    name="task_name",
-    description="Task description",
-    expected_output="Expected output format",
-    agent=agent,
-    context=[previous_task],  # Task dependencies
-    output_pydantic=ResponseModel,  # Structured output
-    condition="condition_function"  # Conditional execution
-)
-```
-
-### Guardrails Usage
-
-#### Task-Level Guardrails
-```python
-from typing import Tuple, Any
-from praisonaiagents import GuardrailResult
-
-# Function-based guardrail (Option 1: Using GuardrailResult)
-def validate_output(task_output: TaskOutput) -> GuardrailResult:
-    """Custom validation function returning GuardrailResult."""
-    if "error" in task_output.raw.lower():
-        return GuardrailResult(
-            success=False,
-            result=None,
-            error="Output contains errors"
-        )
-    if len(task_output.raw) < 10:
-        return GuardrailResult(
-            success=False,
-            result=None,
-            error="Output is too short"
-        )
-    return GuardrailResult(
-        success=True,
-        result=task_output,
-        error=""
-    )
-
-# Function-based guardrail (Option 2: Using Tuple[bool, Any])
-def validate_output_tuple(task_output: TaskOutput) -> Tuple[bool, Any]:
-    """Custom validation function returning tuple."""
-    if "error" in task_output.raw.lower():
-        return False, "Output contains errors"
-    if len(task_output.raw) < 10:
-        return False, "Output is too short"
-    return True, task_output
-
-task = Task(
-    description="Write a professional email",
-    expected_output="A well-formatted email",
-    agent=agent,
-    guardrail=validate_output,  # Function-based guardrail
-    max_retries=3  # Retry up to 3 times if guardrail fails
-)
-
-# LLM-based guardrail
-task = Task(
-    description="Generate marketing copy",
-    expected_output="Professional marketing content",
-    agent=agent,
-    guardrail="Ensure the content is professional, engaging, and free of errors",  # String description
-    max_retries=2
-)
-```
-
-#### Agent-Level Guardrails
-```python
-# Agent guardrails apply to ALL outputs from that agent
-
-# Option 1: Using GuardrailResult
-def validate_professional_tone(task_output: TaskOutput) -> GuardrailResult:
-    """Ensure professional tone in all agent responses."""
-    content = task_output.raw.lower()
-    casual_words = ['yo', 'dude', 'awesome', 'cool']
-    for word in casual_words:
-        if word in content:
-            return GuardrailResult(
-                success=False,
-                result=None,
-                error=f"Unprofessional language detected: {word}"
-            )
-    return GuardrailResult(
-        success=True,
-        result=task_output,
-        error=""
-    )
-
-# Option 2: Using Tuple[bool, Any]
-def validate_professional_tone_tuple(task_output: TaskOutput) -> Tuple[bool, Any]:
-    """Ensure professional tone in all agent responses."""
-    content = task_output.raw.lower()
-    casual_words = ['yo', 'dude', 'awesome', 'cool']
-    for word in casual_words:
-        if word in content:
-            return False, f"Unprofessional language detected: {word}"
-    return True, task_output
-
-# Agent with function-based guardrail
-agent = Agent(
-    name="BusinessWriter",
-    instructions="You are a professional business writer",
-    guardrail=validate_professional_tone,  # Function guardrail
-    max_guardrail_retries=3
-)
-
-# Agent with LLM-based guardrail
-agent = Agent(
-    name="ContentWriter", 
-    instructions="You are a content writer",
-    guardrail="Ensure all responses are professional, accurate, and appropriate for business use",  # String guardrail
-    max_guardrail_retries=2
-)
-```
-
-### Multi-Agent Workflow
-```python
-workflow = PraisonAIAgents(
-    agents=[agent1, agent2],
-    tasks=[task1, task2],
-    process="sequential",  # or "hierarchical", "parallel"
-    verbose=True,
-    manager_agent=manager_agent  # For hierarchical process
-)
-result = workflow.start()
-```
-
-### Async Support
-All major components support async execution:
-```python
-result = await workflow.astart()
-result = await agent.aexecute(task)
-```
-
-## Key Dependencies
-
-- **Core**: `pydantic`, `rich`, `openai`, `mcp`
-- **Memory**: `chromadb`, `mem0ai`
-- **Knowledge**: `markitdown`, `chonkie`
-- **LLM**: `litellm` for unified provider access
-- **API**: `fastapi`, `uvicorn` for server capabilities
-
-## Error Handling
-
-- Global error logging via `error_logs` list
-- Callback system for real-time error reporting
-- Context length exception handling with automatic retry
-- Graceful degradation for optional dependencies
-
-## Testing Strategy
-
-The project uses example-driven testing with 100+ test files in `tests/` directory. Each test file demonstrates specific usage patterns and serves as both test and documentation. Run individual examples to test functionality rather than using a formal test runner.
-
-Use conda activate praisonai-agents to activate the environment.

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/agent_guardrails_example.py

@@ -1,98 +1,75 @@
 #!/usr/bin/env python3
 """
 Agent-level guardrails example.
-
-This example demonstrates how to use guardrails at the Agent level,
+This example demonstrates how to use guardrails at the Agent level
 which will apply to all tasks executed by that agent.
 """
-
 from typing import Tuple, Any
 from praisonaiagents import Agent, TaskOutput
-
 def validate_content_length(task_output: TaskOutput) -> Tuple[bool, Any]:
     """
     Validate that task output content meets minimum length requirement.
-    
     Args:
         task_output: The task output to validate
-        
     Returns:
         Tuple of (success, result_or_error_message)
     """
     if len(task_output.raw) < 50:
         return False, "Content too short - must be at least 50 characters"
     return True, task_output
-
 def validate_professional_tone(task_output: TaskOutput) -> Tuple[bool, Any]:
     """
     Validate that the content has a professional tone.
-    
     Args:
         task_output: The task output to validate
-        
     Returns:
         Tuple of (success, result_or_error_message)
     """
     content = task_output.raw.lower()
     unprofessional_words = ['yo', 'dude', 'awesome', 'cool', 'lol']
-    
     for word in unprofessional_words:
         if word in content:
             return False, f"Content contains unprofessional word: '{word}'"
-    
     return True, task_output
-
 def main():
     """Demonstrate Agent-level guardrails with function-based and LLM-based validation."""
-    
     print("=== Agent Guardrail Examples ===\n")
-    
     # Example 1: Function-based guardrail
     print("1. Function-based guardrail (content length validation):")
     agent1 = Agent(
-        name="ContentWriter",
-        instructions="You are a professional content writer who creates detailed responses",
-        guardrail=validate_content_length,
-        max_guardrail_retries=2
+        name="ContentWriter"
+        instructions="You are a professional content writer who creates detailed responses"
+        guardrails=validate_content_length
     )
-    
     try:
         result1 = agent1.start("Write a brief welcome message")
         print(f"Result: {result1}\n")
     except Exception as e:
         print(f"Error: {e}\n")
-    
     # Example 2: LLM-based guardrail (string description)
     print("2. LLM-based guardrail (professional tone validation):")
     agent2 = Agent(
         name="BusinessWriter", 
-        instructions="You are a business communication expert",
-        guardrail="Ensure the content is professional, formal, and suitable for business communication. No casual language or slang.",
-        max_guardrail_retries=3
+        instructions="You are a business communication expert"
+        guardrails="Ensure the content is professional, formal, and suitable for business communication. No casual language or slang."
     )
-    
     try:
         result2 = agent2.start("Write a welcome message for new employees")
         print(f"Result: {result2}\n")
     except Exception as e:
         print(f"Error: {e}\n")
-    
     # Example 3: Multiple agents with different guardrails
     print("3. Professional tone function-based guardrail:")
     agent3 = Agent(
-        name="ProfessionalWriter",
-        instructions="Write professional business content",
-        guardrail=validate_professional_tone,
-        max_guardrail_retries=2
+        name="ProfessionalWriter"
+        instructions="Write professional business content"
+        guardrails=validate_professional_tone
     )
-    
     try:
         result3 = agent3.start("Write a casual greeting message")
         print(f"Result: {result3}\n")
     except Exception as e:
         print(f"Error: {e}\n")
-    
     print("=== Agent Guardrails Demonstration Complete ===")
-
 if __name__ == "__main__":
     main()

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

@@ -10,7 +10,7 @@ import os
 logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
 
 llm_config = {
-    "model": "gpt-5-nano",  # Changed from Gemini to gpt-5-nano for better tool support
+    "model": "gpt-4o-mini",  # Changed from Gemini to gpt-5-nano for better tool support
     "temperature": 0.7,
     "response_format": {
         "type": "text"
@@ -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/autoagents-tavily.py

@@ -52,7 +52,7 @@ agents = AutoAgents(
     tools=[evaluate, get_stock_price],
     process="sequential",
     llm="gemini/gemini-2.5-flash-lite-preview-06-17",
-    self_reflect=False,
+    reflection=False,
     verbose=False,
     max_agents=3  # Maximum number of agents to create
 )

src/praisonai-agents/basic-agents-tools.py

@@ -5,8 +5,9 @@ def get_weather(city: str) -> str:
 
 agent = Agent(
     instructions="You are a helpful assistant",
-    llm="gpt-5-nano",
-    tools=[get_weather]
+    llm="gpt-4o-mini",
+    tools=[get_weather],
+    output="status"
 )
 
 agent.start("What is the weather in Tokyo?")

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 14:42:59
 **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 | 2.46s | 2.92s | PraisonAI 1.19x faster |
+| 2 Agents | 3.21s | 19.39s | PraisonAI 6.05x faster |
+| 3 Agents | 8.26s | 15.01s | PraisonAI 1.82x 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: 2.46s
+- Min: 1.93s
+- Max: 2.90s
+- Std Dev: 0.49s
+- Times: ['1.93s', '2.55s', '2.90s']
 
 ### agno_single
-- Average: 3.18s
-- Min: 3.00s
-- Max: 3.51s
-- Std Dev: 0.28s
-- Times: ['3.51s', '3.00s', '3.04s']
+- Average: 2.92s
+- Min: 2.85s
+- Max: 2.97s
+- Std Dev: 0.06s
+- Times: ['2.97s', '2.85s', '2.94s']
 
 ### praisonai_two_agents
-- Average: 2.49s
-- Min: 2.22s
-- Max: 2.65s
-- Std Dev: 0.24s
-- Times: ['2.65s', '2.22s', '2.62s']
+- Average: 3.21s
+- Min: 2.80s
+- Max: 3.67s
+- Std Dev: 0.44s
+- Times: ['3.67s', '2.80s', '3.15s']
 
 ### agno_two_agents
-- Average: 20.49s
-- Min: 9.96s
-- Max: 35.05s
-- Std Dev: 13.02s
-- Times: ['16.46s', '9.96s', '35.05s']
+- Average: 19.39s
+- Min: 10.62s
+- Max: 26.06s
+- Std Dev: 7.93s
+- Times: ['26.06s', '10.62s', '21.50s']
 
 ### praisonai_three_agents
-- Average: 6.52s
-- Min: 5.62s
-- Max: 7.39s
-- Std Dev: 0.88s
-- Times: ['5.62s', '6.54s', '7.39s']
+- Average: 8.26s
+- Min: 6.44s
+- Max: 11.88s
+- Std Dev: 3.13s
+- Times: ['6.47s', '6.44s', '11.88s']
 
 ### agno_three_agents
-- Average: 13.97s
-- Min: 11.69s
-- Max: 15.32s
-- Std Dev: 1.99s
-- Times: ['14.91s', '11.69s', '15.32s']
+- Average: 15.01s
+- Min: 14.46s
+- Max: 15.90s
+- Std Dev: 0.78s
+- Times: ['15.90s', '14.46s', '14.68s']
 
 ## Package Versions
 
 | Package | Version |
 |---------|--------|
-| PraisonAI | 0.1.5 |
-| Agno | 2.3.14 |
+| PraisonAI | 0.11.22 |
+| 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/context_basic.py

@@ -1,5 +1,5 @@
 from praisonaiagents import ContextAgent
 
-agent = ContextAgent(llm="gpt-5-nano", auto_analyze=False)
+agent = ContextAgent(llm="gpt-4o-mini", auto_analyze=False)
 
 agent.start("https://github.com/MervinPraison/PraisonAI/ Need to add Authentication")

src/praisonai-agents/debug_comparison.py

@@ -14,7 +14,7 @@ def test_agent_direct():
         agent = Agent(
             instructions="""You are a helpful assistant that can break down complex problems.
             Use the available tools when relevant to perform step-by-step analysis.""",
-            llm="gpt-5-nano",
+            llm="gpt-4o-mini",
             tools=MCP("npx -y @modelcontextprotocol/server-sequential-thinking")
         )
         

src/praisonai-agents/gemini_self_reflection.py

@@ -14,8 +14,8 @@ llm_config = {
 agent = Agent(
     instructions="You are a helpful assistant. Be concise and clear.",
     llm=llm_config,
-    verbose=True,
-    self_reflect=True,
+    output="verbose",
+    reflection=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/guardrail_agents_example.py

@@ -12,7 +12,7 @@ agent = Agent(
 
 task = Task(
     description="Write a welcome message",
-    guardrail=validate_content,
+    guardrails=validate_content,
     agent=agent
 )
 

src/praisonai-agents/guardrails_example.py

@@ -62,7 +62,7 @@ def main():
         description="Write a professional email to a client about project completion",
         expected_output="A well-formatted professional email",
         agent=agent,
-        guardrail=email_validator,  # Function-based guardrail
+        guardrails=email_validator,  # Function-based guardrail
         max_retries=2
     )
     
@@ -110,7 +110,7 @@ Project Team""",
         description="Write a marketing email for a new product launch",
         expected_output="Engaging marketing content",
         agent=agent,
-        guardrail="Ensure the content is professional, engaging, includes a clear call-to-action, and is free of errors",
+        guardrails="Ensure the content is professional, engaging, includes a clear call-to-action, and is free of errors",
         max_retries=3
     )
     

src/praisonai-agents/langchain_example.py

@@ -9,7 +9,7 @@ agent = Agent(
     goal="Search for information from multiple sources",
     backstory="I am an AI assistant that can search YouTube and Wikipedia.",
     tools=[YouTubeSearchTool, WikipediaAPIWrapper],
-    self_reflect=False
+    reflection=False
 )
 
 # Create tasks to demonstrate both tools

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

@@ -32,9 +32,9 @@ 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
+    reflection=True,              # Enable self-reflection
     max_reflect=3,                  # Maximum reflection iterations
     min_reflect=1                   # Minimum reflection iterations
 )

src/praisonai-agents/mcp-agents.py

@@ -3,7 +3,7 @@ from praisonaiagents import Agent, MCP
 agent = Agent(
     instructions="""You are a helpful assistant that can check stock prices and perform other tasks.
     Use the available tools when relevant to answer user questions.""",
-    llm="gpt-5-nano",
+    llm="gpt-4o-mini",
     tools = MCP("/Users/praison/miniconda3/envs/mcp/bin/python /Users/praison/stockprice/app.py")
 )
 

src/praisonai-agents/mongodb_comprehensive_example.py

@@ -71,7 +71,7 @@ def main():
         tools=[mongodb_tools],
         memory=True,
         verbose=True,
-        llm="gpt-5-nano"
+        llm="gpt-4o-mini"
     )
     
     # 2. Knowledge Curator Agent - manages knowledge base
@@ -85,7 +85,7 @@ def main():
         knowledge_config=mongodb_knowledge_config,
         memory=True,
         verbose=True,
-        llm="gpt-5-nano"
+        llm="gpt-4o-mini"
     )
     
     # 3. Business Analyst Agent - analyzes data and provides insights
@@ -99,7 +99,7 @@ def main():
         tools=[mongodb_tools],
         memory=True,
         verbose=True,
-        llm="gpt-5-nano"
+        llm="gpt-4o-mini"
     )
     
     # 4. Customer Service Agent - handles customer interactions
@@ -114,7 +114,7 @@ def main():
         tools=[mongodb_tools],
         memory=True,
         verbose=True,
-        llm="gpt-5-nano"
+        llm="gpt-4o-mini"
     )
     
     # Create comprehensive business tasks

src/praisonai-agents/mongodb_knowledge_basic.py

@@ -55,11 +55,9 @@ 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"
+        llm="gpt-4o-mini"
     )
     
     # Create a research assistant agent
@@ -72,8 +70,7 @@ def main():
         knowledge base. You ensure information is accurate, well-structured, 
         and properly categorized.""",
         memory=True,
-        verbose=True,
-        llm="gpt-5-nano"
+        llm="gpt-4o-mini"
     )
     
     # Create tasks for knowledge management
@@ -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/mongodb_knowledge_example.py

@@ -59,7 +59,7 @@ def main():
         knowledge=[os.path.join(os.path.dirname(__file__), "llms.md")],
         memory=True,
         verbose=True,
-        llm="gpt-5-nano"
+        llm="gpt-4o-mini"
     )
     
     # Create a research assistant agent
@@ -73,7 +73,7 @@ def main():
         and properly categorized.""",
         memory=True,
         verbose=True,
-        llm="gpt-5-nano"
+        llm="gpt-4o-mini"
     )
     
     # Create tasks for knowledge management

src/praisonai-agents/mongodb_memory_basic.py

@@ -3,7 +3,7 @@ from praisonaiagents.memory import Memory
 from praisonaiagents.agent import ContextAgent
 import pymongo
 
-context_agent = ContextAgent(llm="gpt-5-nano", auto_analyze=False)
+context_agent = ContextAgent(llm="gpt-4o-mini", auto_analyze=False)
 
 context_output = context_agent.start("https://github.com/MervinPraison/PraisonAI/ Need to add Authentication")
 
@@ -25,7 +25,7 @@ implementation_agent = Agent(
     goal="Implement authentication features based on project requirements",
     backstory="Expert software implementer specializing in authentication systems, security features, and seamless integration with existing codebases",
     memory=True,
-    llm="gpt-5-nano",
+    llm="gpt-4o-mini",
 )
 
 implementation_task = Task(

src/praisonai-agents/mongodb_memory_example.py

@@ -50,7 +50,7 @@ def main():
         retrieve relevant information from your persistent memory store.""",
         memory=True,
         verbose=True,
-        llm="gpt-5-nano"
+        llm="gpt-4o-mini"
     )
     
     # Create research tasks

src/praisonai-agents/mongodb_tools_example.py

@@ -38,7 +38,7 @@ def main():
         for data management and retrieval.""",
         tools=[mongodb_tools],
         verbose=True,
-        llm="gpt-5-nano"
+        llm="gpt-4o-mini"
     )
     
     # Create a Data Analysis Agent
@@ -51,7 +51,7 @@ def main():
         identify patterns and trends in the stored information.""",
         tools=[mongodb_tools],
         verbose=True,
-        llm="gpt-5-nano"
+        llm="gpt-4o-mini"
     )
     
     # Create tasks for MongoDB operations

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/openai-basic.py

@@ -2,6 +2,6 @@ from praisonaiagents import Agent
 
 agent = Agent(
     instructions="You are a helpful assistant",
-    llm="gpt-5-nano"
+    llm="gpt-4o-mini"
 )
 agent.start("Why sky is Blue?")

src/praisonai-agents/openai-multiturn.py

@@ -2,7 +2,7 @@ from praisonaiagents import Agent
 
 agent = Agent(
     instructions="You are a helpful assistant",
-    llm="gpt-5-nano"
+    llm="gpt-4o-mini"
 )
 agent.start("Why sky is Blue?")
 agent.start("What was my previous question?")

src/praisonai-agents/openai-run.py

@@ -3,7 +3,7 @@ import asyncio
 
 agent = Agent(
     instructions="You are a helpful assistant",
-    llm="gpt-5-nano"
+    llm="gpt-4o-mini"
 )
 agent.run("Why sea is Blue?")
 

src/praisonai-agents/openai-sequential.py

@@ -22,7 +22,7 @@ def multiply(a: int, b: int) -> int:
 
 agent = Agent(
     instructions="You are a helpful assistant. You can use the tools provided to you to help the user.",
-    llm="gpt-5-nano",
+    llm="gpt-4o-mini",
     tools=[get_stock_price, multiply]
 )
 

src/praisonai-agents/planning_simple.py

@@ -14,8 +14,7 @@ task2 = Task(description="Write a short article about meditation benefits", agen
 agents = Agents(
     agents=[researcher, writer],
     tasks=[task1, task2],
-    planning=True,
-    auto_approve_plan=True  # Auto-approve the generated plan
+    planning=True
 )
 
 agents.start()

src/praisonai-agents/praisonai-openai.py

@@ -22,7 +22,7 @@ def multiply(a: int, b: int) -> int:
 
 agent = Agent(
     instructions="You are a helpful assistant. You can use the tools provided to you to help the user.",
-    llm="gpt-5-nano",
+    llm="gpt-4o-mini",
     tools=[get_stock_price, multiply]
 )
 

src/praisonai-agents/praisonaiagents/__init__.py

@@ -5,6 +5,29 @@ This module uses lazy loading to minimize import time and memory usage.
 Heavy dependencies like litellm are only loaded when actually needed.
 """
 
+# =============================================================================
+# NAMESPACE PACKAGE GUARD
+# =============================================================================
+# Detect if we're loaded as a namespace package (which indicates stale artifacts
+# in site-packages). This happens when there's a praisonaiagents/ directory in
+# site-packages without an __init__.py file.
+#
+# Root cause: Partial uninstall or version mismatch leaves directories that
+# Python's PathFinder treats as namespace packages, shadowing the real package.
+# =============================================================================
+if __file__ is None:
+    import warnings as _ns_warnings
+    _ns_warnings.warn(
+        "praisonaiagents is loaded as a namespace package, which indicates "
+        "stale artifacts in site-packages. This will cause import errors. "
+        "Fix: Remove the stale directory with:\n"
+        "  rm -rf $(python -c \"import site; print(site.getsitepackages()[0])\")/praisonaiagents/\n"
+        "Then reinstall: pip install praisonaiagents",
+        ImportWarning,
+        stacklevel=1
+    )
+    del _ns_warnings
+
 # Apply warning patch BEFORE any imports to intercept warnings at the source
 from . import _warning_patch  # noqa: F401
 
@@ -25,6 +48,14 @@ from .tools.registry import get_registry, register_tool, get_tool, ToolRegistry
 from .db import db
 from .obs import obs
 
+# Sub-packages for organized imports (pa.config, pa.tools, etc.)
+# These enable: import praisonaiagents as pa; pa.config.MemoryConfig
+from . import config
+from . import tools
+from . import memory
+from . import workflows
+# Note: knowledge and mcp are lazy-loaded via __getattr__ due to heavy deps
+
 # Workflows - lightweight module
 from .workflows import (
     Workflow, WorkflowStep, WorkflowContext, StepResult,
@@ -68,7 +99,18 @@ from .main import (
 )
 
 # Module-level caches for lazy-loaded classes
-_lazy_cache = {}
+# Using threading.local() ensures thread-safe access to cached values
+import threading
+_lazy_cache_local = threading.local()
+
+def _get_lazy_cache():
+    """Get thread-local lazy cache dict. Thread-safe for concurrent access."""
+    if not hasattr(_lazy_cache_local, 'cache'):
+        _lazy_cache_local.cache = {}
+    return _lazy_cache_local.cache
+
+# Backward-compatible alias (module-level access still works, now thread-safe)
+_lazy_cache = {}  # Kept for compatibility, but prefer _get_lazy_cache()
 
 # Flags for optional modules
 _mcp_available = None
@@ -442,6 +484,51 @@ def __getattr__(name):
         _lazy_cache[name] = ContextManager
         return ContextManager
     
+    # =========================================================================
+    # Sub-package fallback: delegate to sub-packages for additional symbols
+    # This enables: from praisonaiagents import duckduckgo, FileMemory, etc.
+    # =========================================================================
+    
+    # Try tools sub-package (110+ tool functions)
+    try:
+        from . import tools as _tools_module
+        if hasattr(_tools_module, name):
+            result = getattr(_tools_module, name)
+            _lazy_cache[name] = result
+            return result
+    except (ImportError, AttributeError):
+        pass
+    
+    # Try memory sub-package (FileMemory, RulesManager, etc.)
+    try:
+        from . import memory as _memory_module
+        if hasattr(_memory_module, name):
+            result = getattr(_memory_module, name)
+            _lazy_cache[name] = result
+            return result
+    except (ImportError, AttributeError):
+        pass
+    
+    # Try config sub-package (already handled above, but catch any stragglers)
+    try:
+        from . import config as _config_module
+        if hasattr(_config_module, name):
+            result = getattr(_config_module, name)
+            _lazy_cache[name] = result
+            return result
+    except (ImportError, AttributeError):
+        pass
+    
+    # Try workflows sub-package (Pipeline, Route, etc. - already imported, catch aliases)
+    try:
+        from . import workflows as _workflows_module
+        if hasattr(_workflows_module, name):
+            result = getattr(_workflows_module, name)
+            _lazy_cache[name] = result
+            return result
+    except (ImportError, AttributeError):
+        pass
+    
     raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
 
 
@@ -476,174 +563,117 @@ 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
+
+
+# ============================================================================
+# PUBLIC API: __all__ (controls IDE autocomplete and `from X import *`)
+# ============================================================================
+# DESIGN: Keep __all__ minimal for clean IDE experience.
+# All 186+ symbols are still accessible via __getattr__ for backwards compat.
+# Organized imports available via sub-packages: config, tools, memory, workflows
+# ============================================================================
+
 __all__ = [
-    # Core classes (lazy loaded)
+    # Core classes - the essentials
     'Agent',
-    'ImageAgent',
-    'ContextAgent',
-    'create_context_agent',
-    'PraisonAIAgents',
     'Agents',
-    'Tools',
     'Task',
-    # Plugin system
-    'BaseTool',
-    'ToolResult',
-    'ToolValidationError',
-    'validate_tool',
+    
+    # Tool essentials
     'tool',
-    'FunctionTool',
-    'ToolRegistry',
-    'get_registry',
-    'register_tool',
-    'get_tool',
-    'TaskOutput',
-    'ReflectionOutput',
-    'AutoAgents',
-    'AutoRagAgent',
-    'AutoRagConfig',
-    'RagRetrievalPolicy',
-    'Session',
-    'Memory',
-    'db',
-    'obs',
-    'display_interaction',
-    'display_self_reflection',
-    'display_instruction',
-    'display_tool_call',
-    'display_error',
-    'display_generating',
-    'clean_triple_backticks',
-    'error_logs',
-    'register_display_callback',
-    'sync_display_callbacks',
-    'async_display_callbacks',
-    'Knowledge',
-    'Chunking',
-    'GuardrailResult',
-    'LLMGuardrail',
-    'Handoff',
-    'handoff',
-    'handoff_filters',
-    'RECOMMENDED_PROMPT_PREFIX',
-    'prompt_with_handoff_instructions',
-    'HandoffConfig',
-    'HandoffResult',
-    'HandoffInputData',
-    'ContextPolicy',
-    'HandoffError',
-    'HandoffCycleError',
-    'HandoffDepthError',
-    'HandoffTimeoutError',
-    'get_telemetry',
-    'enable_telemetry',
-    'disable_telemetry',
-    'enable_performance_mode',
-    'disable_performance_mode',
-    'cleanup_telemetry_resources',
-    'MinimalTelemetry',
-    'TelemetryCollector',
-    'DeepResearchAgent',
-    'DeepResearchResponse',
-    'Citation',
-    'ReasoningStep',
-    'WebSearchCall',
-    'CodeExecutionStep',
-    'MCPCall',
-    'FileSearchCall',
-    'Provider',
-    'QueryRewriterAgent',
-    'RewriteStrategy',
-    'RewriteResult',
-    'PromptExpanderAgent',
-    'ExpandStrategy',
-    'ExpandResult',
-    # Workflows
-    'Workflow',
-    'WorkflowStep', 
-    'WorkflowContext',
-    'StepResult',
-    'Route',
-    'Parallel',
-    'Loop',
-    'Repeat',
-    'route',
-    'parallel',
-    'loop',
-    'repeat',
-    'Pipeline',
-    # MCP
-    'MCP',
-    # Flow display
-    'FlowDisplay',
-    'track_workflow',
-    # FastContext
-    'FastContext',
-    'FastContextResult',
-    'FileMatch',
-    'LineRange',
-    # Agent Skills
-    'SkillManager',
-    'SkillProperties',
-    'SkillMetadata',
-    'SkillLoader',
-    # Planning
-    'Plan',
-    'PlanStep',
-    'TodoList',
-    'TodoItem',
-    'PlanStorage',
-    'PlanningAgent',
-    'ApprovalCallback',
-    'READ_ONLY_TOOLS',
-    'RESTRICTED_TOOLS',
-    # UI
-    'AGUI',
-    'A2A',
-    # RAG
-    'RAG',
-    'RAGConfig',
-    'RAGResult',
-    'RAGCitation',
-    # Feature Configs (agent-centric API)
-    'MemoryConfig',
-    'KnowledgeConfig',
-    'PlanningConfig',
-    'ReflectionConfig',
-    'GuardrailConfig',
-    'WebConfig',
-    'OutputConfig',
-    'ExecutionConfig',
-    'TemplateConfig',
-    'CachingConfig',
-    'HooksConfig',
-    'SkillsConfig',
-    'AutonomyConfig',
-    'MemoryBackend',
-    'ChunkingStrategy',
-    'GuardrailAction',
-    'WebSearchProvider',
-    'OutputPreset',
-    'ExecutionPreset',
-    'AutonomyLevel',
-    # Multi-Agent Feature Configs
-    'MultiAgentHooksConfig',
-    'MultiAgentOutputConfig',
-    'MultiAgentExecutionConfig',
-    'MultiAgentPlanningConfig',
-    'MultiAgentMemoryConfig',
-    # Context Management
-    'ManagerConfig',
-    'ContextManager',
-    # Parameter Resolution (agent-centric API)
-    'resolve',
-    'ArrayMode',
-    'resolve_guardrails',
-    'resolve_guardrail_policies',
-    # Presets (agent-centric API)
-    'GUARDRAIL_PRESETS',
-    'KNOWLEDGE_PRESETS',
-    # Parse Utilities
-    'is_policy_string',
-    'parse_policy_string',
+    'Tools',
+    
+    # Sub-packages for organized imports
+    # Usage: import praisonaiagents as pa; pa.config.MemoryConfig
+    'config',
+    'tools',
+    'memory',
+    'workflows',
 ]
+
+
+def __dir__():
+    """
+    Return clean list for dir() - matches __all__ plus standard attributes.
+    
+    This keeps IDE autocomplete clean while preserving full backwards
+    compatibility via __getattr__ for all 186+ legacy exports.
+    """
+    return list(__all__) + [
+        # Standard module attributes
+        '__name__', '__doc__', '__file__', '__path__', '__package__',
+        '__loader__', '__spec__', '__cached__', '__builtins__',
+    ]
+
+
+# ============================================================================
+# BACKWARDS COMPATIBILITY: Legacy __all__ items (for reference)
+# ============================================================================
+# All items below are still importable via __getattr__ but NOT in autocomplete:
+# - ImageAgent, ContextAgent, create_context_agent, PraisonAIAgents
+# - BaseTool, ToolResult, ToolValidationError, validate_tool, FunctionTool
+# - ToolRegistry, get_registry, register_tool, get_tool
+# - TaskOutput, ReflectionOutput, AutoAgents, AutoRagAgent, AutoRagConfig
+# - Session, Memory, db, obs, Knowledge, Chunking
+# - GuardrailResult, LLMGuardrail, Handoff, handoff, handoff_filters
+# - MemoryConfig, KnowledgeConfig, PlanningConfig, OutputConfig, etc.
+# - Workflow, WorkflowStep, Route, Parallel, Loop, Repeat, Pipeline, etc.
+# - MCP, FlowDisplay, track_workflow, FastContext, etc.
+# - Plan, PlanStep, TodoList, PlanningAgent, ApprovalCallback, etc.
+# - RAG, RAGConfig, RAGResult, AGUI, A2A, etc.
+# - All telemetry, display, and utility functions
+# ============================================================================

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

@@ -30,6 +30,13 @@ from .prompt_expander_agent import (
     ExpandStrategy,
     ExpandResult
 )
+from .protocols import (
+    AgentProtocol,
+    RunnableAgentProtocol,
+    ToolAwareAgentProtocol,
+    MemoryAwareAgentProtocol,
+    FullAgentProtocol,
+)
 
 __all__ = [
     'Agent',
@@ -64,5 +71,11 @@ __all__ = [
     'RewriteResult',
     'PromptExpanderAgent',
     'ExpandStrategy',
-    'ExpandResult'
+    'ExpandResult',
+    # Protocols
+    'AgentProtocol',
+    'RunnableAgentProtocol',
+    'ToolAwareAgentProtocol',
+    'MemoryAwareAgentProtocol',
+    'FullAgentProtocol',
 ]

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:
@@ -254,10 +254,11 @@ class Agent:
         instructions: Optional[str] = None,
         # LLM configuration
         llm: Optional[Union[str, Any]] = None,
-        function_calling_llm: Optional[Any] = None,
-        llm_config: Optional[Dict[str, Any]] = None,
-        base_url: Optional[str] = None,
-        api_key: Optional[str] = None,
+        model: Optional[Union[str, Any]] = None,  # Alias for llm=
+        function_calling_llm: Optional[Any] = None,  # DEPRECATED: use llm= with LLMConfig
+        llm_config: Optional[Dict[str, Any]] = None,  # DEPRECATED: use llm= with LLMConfig
+        base_url: Optional[str] = None,  # Kept separate (connection/auth)
+        api_key: Optional[str] = None,  # Kept separate (connection/auth)
         # Tools
         tools: Optional[List[Any]] = None,
         allow_delegation: bool = False,
@@ -290,96 +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:
@@ -446,14 +438,31 @@ 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
+        # Track if user explicitly configured output (for respecting in start())
+        if output is None:
+            output = DEFAULT_OUTPUT_MODE
+            _has_explicit_output = False
+        else:
+            _has_explicit_output = True
+        
         _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
@@ -462,8 +471,45 @@ 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)
+            status_trace = getattr(_output_config, 'status_trace', False)  # New: clean inline status
+            simple_output = getattr(_output_config, 'simple_output', False)  # status preset: no timestamps
         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
+            status_trace = False
+            simple_output = False
+        
+        # Enable trace output mode if configured (takes priority)
+        # This provides timestamped inline status with duration
+        if status_trace:
+            try:
+                from ..output.trace import enable_trace_output, is_trace_output_enabled
+                if not is_trace_output_enabled():
+                    enable_trace_output(use_color=True, show_timestamps=True)
+            except ImportError:
+                pass  # Trace module not available
+        # Enable status output mode if configured (simple progress, no timestamps)
+        # This registers callbacks to capture tool calls and final output
+        elif actions_trace:
+            try:
+                from ..output.status import enable_status_output, is_status_output_enabled
+                if not is_status_output_enabled():
+                    output_format = "jsonl" if json_output else "text"
+                    # simple_output=True means status preset (no timestamps)
+                    # metrics=True means debug preset (show token/cost info)
+                    enable_status_output(
+                        redact=True,
+                        use_color=True,
+                        format=output_format,
+                        show_timestamps=not simple_output,
+                        show_metrics=metrics
+                    )
+            except ImportError:
+                pass  # Status module not available
         
         # ─────────────────────────────────────────────────────────────────────
         # Resolve EXECUTION param using unified resolver
@@ -754,7 +800,15 @@ class Agent:
 
         # If instructions are provided, use them to set role, goal, and backstory
         if instructions:
-            self.name = name or "Agent"
+            # Only use explicitly provided name, don't auto-generate from instructions
+            # Auto-generation was producing confusing names like "You Are Agent" from
+            # instructions like "You are a helpful assistant"
+            if name:
+                self.name = name
+            else:
+                # Don't auto-generate - None signals "no explicit name provided"
+                # Display logic will skip Agent Info panel when name is None
+                self.name = None
             self.role = role or "Assistant"
             self.goal = goal or instructions
             self.backstory = backstory or instructions
@@ -783,13 +837,30 @@ class Agent:
         self._llm_config = llm_config or {}
         self._llm_configurable = self._llm_config.get('configurable', False)
         
+        # ============================================================
+        # LLM CONSOLIDATION: Handle model= alias and deprecation warnings
+        # Precedence: llm= > model= > default
+        # ============================================================
+        import warnings
+        
+        # Handle model= alias for llm= (NO warnings - both are valid)
+        if llm is None and model is not None:
+            llm = model  # model= is an alias for llm=
+        
+        # Legacy params (llm_config, function_calling_llm) are accepted silently
+        # for backward compatibility - NO deprecation warnings per policy
+        
         # Store rate limiter (optional, zero overhead when None)
         self._rate_limiter = rate_limiter
         
-        # Store OpenAI client parameters for lazy initialization
+        # Store OpenAI client parameters for lazy initialization (kept separate)
         self._openai_api_key = api_key
         self._openai_base_url = base_url
         self.__openai_client = None
+        
+        # Expose base_url and api_key as properties for tests
+        self.base_url = base_url
+        self.api_key = api_key
 
         # If base_url is provided, always create a custom LLM instance
         if base_url:
@@ -806,7 +877,7 @@ 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,
@@ -869,7 +940,7 @@ 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
@@ -893,6 +964,7 @@ class Agent:
         self._memory_instance = None
         self._init_memory(memory, user_id)
         self.verbose = verbose
+        self._has_explicit_output_config = _has_explicit_output  # Track if user set output mode
         self.allow_delegation = allow_delegation
         self.step_callback = step_callback
         self.cache = cache
@@ -915,7 +987,7 @@ 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
@@ -957,9 +1029,14 @@ Your Goal: {self.goal}
                 self.tools.append(internet_search)
                 logging.info("Model does not support native web search. Added DuckDuckGo fallback tool.")
         
-        # Log warning if web_fetch is enabled but model doesn't support it
+        # Raise error if web_fetch is explicitly enabled but model doesn't support it
+        # Web fetch is only supported by Anthropic/Claude models
         if web_fetch and not self._model_supports_web_fetch():
-            logging.warning(f"Model '{self.llm}' does not support native web fetch. Web fetch will be ignored.")
+            raise ValueError(
+                f"web_fetch is only supported on Anthropic/Claude models. "
+                f"Model '{self.llm}' does not support web fetch. "
+                f"Either use a Claude model (e.g., 'anthropic/claude-sonnet-4') or disable fetch with web=WebConfig(fetch=False)."
+            )
         
         # Initialize guardrail settings
         self.guardrail = guardrail
@@ -1042,6 +1119,16 @@ 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
+
+        # Telemetry
+        try:
+            from ..telemetry import get_telemetry
+            self._telemetry = get_telemetry()
+        except (ImportError, AttributeError):
+            self._telemetry = None
 
     @property
     def auto_memory(self):
@@ -1161,7 +1248,11 @@ Your Goal: {self.goal}
 
     @property
     def console(self):
-        """Lazily initialize Rich Console only when needed."""
+        """Lazily initialize Rich Console only when needed AND verbose is True."""
+        # Only return console if verbose mode is enabled
+        # This prevents panels from being shown in status/silent modes
+        if not self.verbose:
+            return None
         if self._console is None:
             from rich.console import Console
             self._console = Console()
@@ -1697,7 +1788,7 @@ 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)
     
@@ -1719,7 +1810,7 @@ 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)
     
@@ -1963,7 +2054,7 @@ 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
         """
@@ -1973,7 +2064,7 @@ 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."""
@@ -2794,6 +2885,9 @@ Your Goal: {self.goal}"""
         """
         logging.debug(f"{self.name} executing tool {function_name} with arguments: {arguments}")
         
+        # NOTE: tool_call callback is triggered by display_tool_call in openai_client.py
+        # Do NOT call it here to avoid duplicate output
+        
         # Set up injection context for tools with Injected parameters
         from ..tools.injected import AgentState, with_injection_context
         state = AgentState(
@@ -3108,21 +3202,20 @@ Your Goal: {self.goal}"""
         
         This centralizes the logic for callback execution and display to avoid duplication.
         """
-        # Always execute callbacks regardless of verbose setting (only when not using custom LLM)
-        if not self._using_custom_llm:
-            execute_sync_callback(
-                'interaction',
-                message=prompt,
-                response=response,
-                markdown=self.markdown,
-                generation_time=generation_time,
-                agent_name=self.name,
-                agent_role=self.role,
-                agent_tools=[t.__name__ for t in self.tools] if self.tools else None,
-                task_name=task_name,
-                task_description=task_description, 
-                task_id=task_id
-            )
+        # Always execute callbacks for status/trace output (regardless of LLM backend)
+        execute_sync_callback(
+            'interaction',
+            message=prompt,
+            response=response,
+            markdown=self.markdown,
+            generation_time=generation_time,
+            agent_name=self.name,
+            agent_role=self.role,
+            agent_tools=[t.__name__ for t in self.tools] if self.tools else None,
+            task_name=task_name,
+            task_description=task_description, 
+            task_id=task_id
+        )
         # Always display final interaction when verbose is True to ensure consistent formatting
         # This ensures both OpenAI and custom LLM providers (like Gemini) show formatted output
         if self.verbose and not self._final_display_shown:
@@ -3537,7 +3630,7 @@ Your Goal: {self.goal}"""
                     output_pydantic=output_pydantic,
                     verbose=self.verbose,
                     markdown=self.markdown,
-                    self_reflect=self.self_reflect,
+                    reflection=self.self_reflect,
                     max_reflect=self.max_reflect,
                     min_reflect=self.min_reflect,
                     console=self.console,
@@ -3673,7 +3766,7 @@ 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)
@@ -3825,6 +3918,10 @@ Output MUST be JSON with 'reflection' and 'satisfactory'.
 
     async def achat(self, prompt: str, temperature=1.0, tools=None, output_json=None, output_pydantic=None, reasoning_steps=False, task_name=None, task_description=None, task_id=None):
         """Async version of chat method with self-reflection support.""" 
+        # Track execution via telemetry
+        if hasattr(self, '_telemetry') and self._telemetry:
+            self._telemetry.track_agent_execution(self.name, success=True)
+            
         # Reset the final display flag for each new conversation
         self._final_display_shown = False
         
@@ -3891,7 +3988,7 @@ Output MUST be JSON with 'reflection' and 'satisfactory'.
                         output_pydantic=output_pydantic,
                         verbose=self.verbose,
                         markdown=self.markdown,
-                        self_reflect=self.self_reflect,
+                        reflection=self.self_reflect,
                         max_reflect=self.max_reflect,
                         min_reflect=self.min_reflect,
                         console=self.console,
@@ -4198,7 +4295,7 @@ 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
 
@@ -4212,16 +4309,94 @@ 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
+        """
+        # Remove stream from kwargs since achat() doesn't accept it
+        kwargs.pop('stream', None)
+        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."""
@@ -4355,26 +4530,297 @@ 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: agent's explicit output config > start() override > TTY auto
+        # ─────────────────────────────────────────────────────────────────────
+        original_verbose = self.verbose
+        original_markdown = self.markdown
+        output_override = kwargs.pop('output', None)  # Pop to prevent passing to chat()
+        
+        # Check if agent was configured with explicit output mode (not default)
+        # If so, respect it and don't auto-enable verbose for TTY
+        has_explicit_output = getattr(self, '_has_explicit_output_config', False)
+        
+        try:
+            # Apply output override from start() call if provided
+            if 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)
+            # Only auto-enable verbose for TTY if NO explicit output was configured
+            elif is_tty and not has_explicit_output:
+                self.verbose = True
+                self.markdown = True
+            
+            # 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 (deduplicated)
+                    if tools_called:
+                        # Use dict.fromkeys to preserve order while removing duplicates
+                        unique_tools = list(dict.fromkeys(tools_called))
+                        tools_summary = ", ".join(unique_tools)
+                        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/prompt_expander_agent.py

@@ -17,7 +17,7 @@ Supported Expansion Strategies:
 Example:
     from praisonaiagents import PromptExpanderAgent, ExpandStrategy
     
-    agent = PromptExpanderAgent(verbose=True)
+    agent = PromptExpanderAgent()
     
     # Basic expansion
     result = agent.expand("write a movie script in 3 lines")
@@ -29,7 +29,7 @@ Example:
     
     # With tools for context gathering
     from praisonaiagents.tools import internet_search
-    agent = PromptExpanderAgent(tools=[internet_search], verbose=True)
+    agent = PromptExpanderAgent(tools=[internet_search])
     result = agent.expand("latest AI developments")
 """
 
@@ -73,13 +73,12 @@ class PromptExpanderAgent:
     
     Attributes:
         name: Name of the agent
-        model: LLM model to use for expansion
-        verbose: Whether to print detailed logs
+        model: LLM model to use for expansion (alias for llm=)
         temperature: Temperature for LLM generation (higher = more creative)
         tools: Optional tools for context gathering
         
     Example:
-        agent = PromptExpanderAgent(model="gpt-4o-mini", verbose=True)
+        agent = PromptExpanderAgent(model="gpt-4o-mini")
         result = agent.expand("write a poem", strategy=ExpandStrategy.CREATIVE)
         print(result.expanded_prompt)
     """
@@ -230,8 +229,7 @@ Expanded prompt:"""
                 backstory="You are an expert at understanding user intent and transforming brief requests into comprehensive, detailed prompts.",
                 tools=self.tools,
                 llm=self.model,
-                verbose=self.verbose,
-                markdown=False
+                output={"verbose": self.verbose, "markdown": False}
             )
         return self._agent
     

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

@@ -0,0 +1,151 @@
+"""
+Agent Protocol Definitions.
+
+Provides Protocol interfaces that define the minimal contract for agent implementations.
+This enables:
+- Mocking agents in tests without real LLM calls
+- Creating custom agent implementations
+- Type checking with static analyzers
+
+These protocols are lightweight and have zero performance impact.
+"""
+from typing import Protocol, runtime_checkable, Optional, Any, Dict, List
+
+
+@runtime_checkable
+class AgentProtocol(Protocol):
+    """
+    Minimal Protocol for agent implementations.
+    
+    This defines the essential interface that any agent must provide.
+    It enables proper mocking and testing without real LLM dependencies.
+    
+    Example:
+        ```python
+        # Create a mock agent for testing
+        class MockAgent:
+            @property
+            def name(self) -> str:
+                return "TestAgent"
+            
+            def chat(self, prompt: str, **kwargs) -> str:
+                return "Mock response"
+            
+            async def achat(self, prompt: str, **kwargs) -> str:
+                return "Async mock response"
+        
+        # Use in tests
+        agent: AgentProtocol = MockAgent()
+        result = agent.chat("Hello")
+        ```
+    """
+    
+    @property
+    def name(self) -> str:
+        """Agent's name identifier."""
+        ...
+    
+    def chat(self, prompt: str, **kwargs) -> str:
+        """
+        Synchronous chat with the agent.
+        
+        Args:
+            prompt: The user prompt or message
+            **kwargs: Additional optional parameters (temperature, tools, etc.)
+            
+        Returns:
+            The agent's response as a string
+        """
+        ...
+    
+    async def achat(self, prompt: str, **kwargs) -> str:
+        """
+        Asynchronous chat with the agent.
+        
+        Args:
+            prompt: The user prompt or message
+            **kwargs: Additional optional parameters (temperature, tools, etc.)
+            
+        Returns:
+            The agent's response as a string
+        """
+        ...
+
+
+@runtime_checkable
+class RunnableAgentProtocol(AgentProtocol, Protocol):
+    """
+    Extended Protocol for agents that support run/start methods.
+    
+    This extends AgentProtocol with the run/start interface used
+    by the Agent class for task execution.
+    """
+    
+    def run(self, prompt: str, **kwargs) -> str:
+        """Run the agent with a prompt (alias for chat in most cases)."""
+        ...
+    
+    def start(self, prompt: str, **kwargs) -> str:
+        """Start the agent with a prompt."""
+        ...
+    
+    async def arun(self, prompt: str, **kwargs) -> str:
+        """Async run the agent with a prompt."""
+        ...
+    
+    async def astart(self, prompt: str, **kwargs) -> str:
+        """Async start the agent with a prompt."""
+        ...
+
+
+@runtime_checkable
+class ToolAwareAgentProtocol(AgentProtocol, Protocol):
+    """
+    Protocol for agents that support tools.
+    
+    This extends AgentProtocol with tool-related properties.
+    """
+    
+    @property
+    def tools(self) -> List[Any]:
+        """List of tools available to the agent."""
+        ...
+
+
+@runtime_checkable  
+class MemoryAwareAgentProtocol(AgentProtocol, Protocol):
+    """
+    Protocol for agents that support memory.
+    
+    This extends AgentProtocol with memory-related properties.
+    """
+    
+    @property
+    def chat_history(self) -> List[Dict[str, Any]]:
+        """The agent's conversation history."""
+        ...
+    
+    def clear_history(self) -> None:
+        """Clear the agent's conversation history."""
+        ...
+
+
+# Composite protocol for fully-featured agents
+@runtime_checkable
+class FullAgentProtocol(RunnableAgentProtocol, ToolAwareAgentProtocol, MemoryAwareAgentProtocol, Protocol):
+    """
+    Complete Protocol for full-featured agents.
+    
+    This combines all agent protocol features for agents that support
+    the full interface (chat, run, tools, memory).
+    """
+    pass
+
+
+__all__ = [
+    'AgentProtocol',
+    'RunnableAgentProtocol', 
+    'ToolAwareAgentProtocol',
+    'MemoryAwareAgentProtocol',
+    'FullAgentProtocol',
+]

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

@@ -15,7 +15,7 @@ Supported Rewriting Strategies:
 Example:
     from praisonaiagents import QueryRewriterAgent, RewriteStrategy
     
-    agent = QueryRewriterAgent(verbose=True)
+    agent = QueryRewriterAgent()
     
     # Basic rewriting
     result = agent.rewrite("AI trends")
@@ -34,7 +34,7 @@ Example:
     
     # With search tools for context-aware rewriting
     from praisonaiagents.tools import internet_search
-    agent = QueryRewriterAgent(tools=[internet_search], verbose=True)
+    agent = QueryRewriterAgent(tools=[internet_search])
     result = agent.rewrite("latest AI developments")  # Searches first, then rewrites
 """
 
@@ -96,13 +96,12 @@ class QueryRewriterAgent:
     
     Attributes:
         name: Name of the agent
-        model: LLM model to use for rewriting
-        verbose: Whether to print detailed logs
+        model: LLM model to use for rewriting (alias for llm=)
         max_queries: Maximum number of queries to generate for multi-query strategy
         abbreviations: Dictionary of common abbreviations to expand
         
     Example:
-        agent = QueryRewriterAgent(model="gpt-4o-mini", verbose=True)
+        agent = QueryRewriterAgent(model="gpt-4o-mini")
         result = agent.rewrite("ML best practices", strategy=RewriteStrategy.MULTI_QUERY)
         for query in result.rewritten_queries:
             print(query)
@@ -288,8 +287,7 @@ Improved version:"""
                 backstory="You are an expert at understanding user intent and transforming queries for optimal information retrieval.",
                 tools=self.tools,
                 llm=self.model,
-                verbose=self.verbose,
-                markdown=False
+                output={"verbose": self.verbose, "markdown": False}
             )
         return self._agent
     

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

@@ -65,6 +65,52 @@ def process_video(video_path: str, seconds_per_frame=2):
     video.release()
     return base64_frames
 
+
+def get_multimodal_message(text_prompt: str, images: list) -> list:
+    """
+    Build multimodal message content for LLM with text and images.
+    
+    DRY helper - replaces duplicate _get_multimodal_message in aexecute_task/execute_task.
+    
+    Args:
+        text_prompt: The text content of the message
+        images: List of image paths (local or URL)
+        
+    Returns:
+        List of content items for multimodal LLM message
+    """
+    content = [{"type": "text", "text": text_prompt}]
+    
+    for img in images:
+        # If local file path for a valid image
+        if os.path.exists(img):
+            ext = os.path.splitext(img)[1].lower()
+            # If it's a .mp4, convert to frames
+            if ext == ".mp4":
+                frames = process_video(img, seconds_per_frame=1)
+                content.append({"type": "text", "text": "These are frames from the video."})
+                for f in frames:
+                    content.append({
+                        "type": "image_url",
+                        "image_url": {"url": f"data:image/jpg;base64,{f}"}
+                    })
+            else:
+                encoded = encode_file_to_base64(img)
+                content.append({
+                    "type": "image_url",
+                    "image_url": {
+                        "url": f"data:image/{ext.lstrip('.')};base64,{encoded}"
+                    }
+                })
+        else:
+            # Treat as a remote URL
+            content.append({
+                "type": "image_url",
+                "image_url": {"url": img}
+            })
+    return content
+
+
 def process_task_context(context_item, verbose=0, user_id=None):
     """
     Process a single context item for task execution.
@@ -148,6 +194,7 @@ class Agents:
         guardrails: Optional[Any] = None,  # Union[bool, Callable, GuardrailConfig] - validation
         web: Optional[Any] = None,  # Union[bool, WebConfig] - web search/fetch
         reflection: Optional[Any] = None,  # Union[bool, ReflectionConfig] - self-reflection
+        caching: Optional[Any] = None,  # Union[bool, CachingConfig] - caching
     ):
         """
         Initialize PraisonAIAgents with consolidated feature parameters.
@@ -177,6 +224,7 @@ class Agents:
         self._guardrails = guardrails
         self._web = web
         self._reflection = reflection
+        self._caching = caching
         # ─────────────────────────────────────────────────────────────────────
         # Extract values from consolidated params using UNIFIED CANONICAL resolver
         # Precedence: Instance > Config > Dict > Array > String > Bool > Default
@@ -361,7 +409,7 @@ 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:
@@ -464,10 +512,16 @@ class Agents:
         
         if self.shared_memory:
             for task in tasks:
-                if not task.memory:
                     task.memory = self.shared_memory
                     logger.info(f"Assigned shared memory to task {task.id}")
 
+        # Telemetry
+        try:
+            from ..telemetry import get_telemetry
+            self._telemetry = get_telemetry()
+        except (ImportError, AttributeError):
+            self._telemetry = None
+
     def add_task(self, task):
         task_id = self.task_id_counter
         task.id = task_id
@@ -601,40 +655,9 @@ Context:
         logger.debug(f"Starting execution of task {task_id} with prompt:\n{task_prompt}")
 
         if task.images:
-            def _get_multimodal_message(text_prompt, images):
-                content = [{"type": "text", "text": text_prompt}]
-
-                for img in images:
-                    # If local file path for a valid image
-                    if os.path.exists(img):
-                        ext = os.path.splitext(img)[1].lower()
-                        # If it's a .mp4, convert to frames
-                        if ext == ".mp4":
-                            frames = process_video(img, seconds_per_frame=1)
-                            content.append({"type": "text", "text": "These are frames from the video."})
-                            for f in frames:
-                                content.append({
-                                    "type": "image_url",
-                                    "image_url": {"url": f"data:image/jpg;base64,{f}"}
-                                })
-                        else:
-                            encoded = encode_file_to_base64(img)
-                            content.append({
-                                "type": "image_url",
-                                "image_url": {
-                                    "url": f"data:image/{ext.lstrip('.')};base64,{encoded}"
-                                }
-                            })
-                    else:
-                        # Treat as a remote URL
-                        content.append({
-                            "type": "image_url",
-                            "image_url": {"url": img}
-                        })
-                return content
-
+            # Use shared multimodal helper (DRY - defined at module level)
             agent_output = await executor_agent.achat(
-                _get_multimodal_message(task_prompt, task.images),
+                get_multimodal_message(task_prompt, task.images),
                 tools=tools,
                 output_json=task.output_json,
                 output_pydantic=task.output_pydantic,
@@ -819,13 +842,17 @@ Context:
                     await loop.run_in_executor(None, self.run_task, task_id)
 
     async def astart(self, content=None, return_dict=False, **kwargs):
-        """Async version of start method
+        """Async version of start method.
         
         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
             **kwargs: Additional arguments
         """
+        # Track execution via telemetry
+        if hasattr(self, '_telemetry') and self._telemetry:
+            self._telemetry.track_agent_execution(self.name, success=True, async_mode=True)
+            
         if content:
             # Add content to context of all tasks
             for task in self.tasks.values():
@@ -959,40 +986,9 @@ Context:
         logger.debug(f"Starting execution of task {task_id} with prompt:\n{task_prompt}")
 
         if task.images:
-            def _get_multimodal_message(text_prompt, images):
-                content = [{"type": "text", "text": text_prompt}]
-
-                for img in images:
-                    # If local file path for a valid image
-                    if os.path.exists(img):
-                        ext = os.path.splitext(img)[1].lower()
-                        # If it's a .mp4, convert to frames
-                        if ext == ".mp4":
-                            frames = process_video(img, seconds_per_frame=1)
-                            content.append({"type": "text", "text": "These are frames from the video."})
-                            for f in frames:
-                                content.append({
-                                    "type": "image_url",
-                                    "image_url": {"url": f"data:image/jpg;base64,{f}"}
-                                })
-                        else:
-                            encoded = encode_file_to_base64(img)
-                            content.append({
-                                "type": "image_url",
-                                "image_url": {
-                                    "url": f"data:image/{ext.lstrip('.')};base64,{encoded}"
-                                }
-                            })
-                    else:
-                        # Treat as a remote URL
-                        content.append({
-                            "type": "image_url",
-                            "image_url": {"url": img}
-                        })
-                return content
-
+            # Use shared multimodal helper (DRY - defined at module level)
             agent_output = executor_agent.chat(
-                _get_multimodal_message(task_prompt, task.images),
+                get_multimodal_message(task_prompt, task.images),
                 tools=task.tools,
                 output_json=task.output_json,
                 output_pydantic=task.output_pydantic,
@@ -1191,30 +1187,147 @@ 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")
+            ```
         """
+        # Track execution via telemetry
+        if hasattr(self, '_telemetry') and self._telemetry:
+            self._telemetry.track_agent_execution(self.name, success=True)
+        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
+                    from rich.markdown import Markdown
+                    console.print(Panel(
+                        Markdown(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)
@@ -1222,10 +1335,8 @@ 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
@@ -1236,7 +1347,6 @@ 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]
@@ -1244,12 +1354,21 @@ 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"""
@@ -2189,7 +2308,7 @@ Context:
                 # Inherit from original task if available
                 memory=original_task.memory if original_task else None,
                 callback=original_task.callback if original_task else None,
-                guardrail=original_task.guardrail if original_task else None,
+                guardrails=original_task.guardrail if original_task else None,
                 max_retries=original_task.max_retries if original_task else 3,
                 output_json=original_task.output_json if original_task else None,
                 output_pydantic=original_task.output_pydantic if original_task else None,

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

@@ -74,6 +74,17 @@ class AutoAgents(Agents):
         api_key: Optional[str] = None,
         max_agents: int = 3,  # Maximum number of agents
         autonomy: Optional[Any] = None,  # Union[bool, dict, AutonomyConfig] - Enable autonomy for all agents
+        # ============================================================
+        # CONSOLIDATED PARAMS (new SDK contract - take precedence)
+        # ============================================================
+        output: Optional[Any] = None,  # Union[str, OutputConfig] - overrides verbose/markdown
+        reflection: Optional[Any] = None,  # Union[bool, ReflectionConfig] - overrides self_reflect/max_reflect/min_reflect/reflect_llm
+        caching: Optional[Any] = None,  # Union[bool, CachingConfig] - overrides cache
+        knowledge: Optional[Any] = None,  # Union[bool, List[str], KnowledgeConfig] - overrides knowledge_sources/embedder_config
+        execution: Optional[Any] = None,  # Union[str, ExecutionConfig] - overrides max_iter
+        guardrails: Optional[Any] = None,  # Union[bool, Callable, GuardrailConfig]
+        web: Optional[Any] = None,  # Union[bool, WebConfig]
+        hooks: Optional[Any] = None,  # HooksConfig
     ):
         """Initialize AutoAgents with configuration for automatic agent and task creation."""
         if max_agents < 1:
@@ -92,7 +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
@@ -112,6 +123,16 @@ class AutoAgents(Agents):
         self.api_key = api_key
         self.autonomy = autonomy  # Store autonomy config for agents
         
+        # Store consolidated params (new SDK contract - take precedence over legacy)
+        self._output = output
+        self._reflection = reflection
+        self._caching = caching
+        self._knowledge = knowledge
+        self._execution = execution
+        self._guardrails = guardrails
+        self._web = web
+        self._hooks = hooks
+        
         # Display initial instruction
         if self.verbose:
             display_instruction(f"🎯 Main Task: {self.instructions}")
@@ -129,14 +150,21 @@ class AutoAgents(Agents):
             self._display_agents_and_tasks(agents, tasks)
         
         # Initialize parent class with generated agents and tasks
+        # Use consolidated output= instead of legacy verbose=
         super().__init__(
             agents=agents,
             tasks=tasks,
-            verbose=verbose,
-            completion_checker=completion_checker,
-            max_retries=max_retries,
             process=process,
-            manager_llm=manager_llm or self.llm
+            manager_llm=manager_llm or self.llm,
+            output=self._output,  # Consolidated output param
+            execution=self._execution,
+            hooks=self._hooks,
+            memory=self.memory,
+            reflection=self._reflection,
+            caching=self._caching,
+            guardrails=self._guardrails,
+            web=self._web,
+            autonomy=self.autonomy,
         )
 
     def _display_agents_and_tasks(self, agents: List[Agent], tasks: List[Task]):
@@ -431,38 +459,33 @@ DO NOT use strings for tasks. Each task MUST be a complete object with all four
             agent_tools = self._assign_tools_to_agent(agent_config)
             
             # Create the agent with all parameters
+            # Consolidated params take precedence over legacy params
+            # Use consolidated knowledge if provided, else fall back to legacy knowledge_sources
+            effective_knowledge = self._knowledge if self._knowledge is not None else self.knowledge_sources
+            
             agent = Agent(
                 name=agent_config.name,
                 role=agent_config.role,
                 goal=agent_config.goal,
                 backstory=agent_config.backstory,
                 tools=agent_tools,  # Use assigned tools
-                verbose=self.verbose >= 1,
                 allow_code_execution=self.allow_code_execution,
                 memory=self.memory,
-                markdown=self.markdown,
-                self_reflect=self.self_reflect,
-                max_reflect=self.max_reflect,
-                min_reflect=self.min_reflect,
-                llm=self.llm,
-                function_calling_llm=self.function_calling_llm,
+                llm=self.llm,  # Consolidated LLM param
                 code_execution_mode=self.code_execution_mode,
-                embedder_config=self.embedder_config,
-                knowledge=self.knowledge_sources,
-                use_system_prompt=self.use_system_prompt,
-                cache=self.cache,
                 allow_delegation=self.allow_delegation,
-                step_callback=self.step_callback,
-                system_template=self.system_template,
-                prompt_template=self.prompt_template,
-                response_template=self.response_template,
-                max_rpm=self.max_rpm,
-                max_execution_time=self.max_execution_time,
-                max_iter=self.max_iter,
-                reflect_llm=self.reflect_llm,
                 base_url=self.base_url,
                 api_key=self.api_key,
-                autonomy=self.autonomy  # Pass autonomy config to agents
+                # Consolidated params (new SDK contract)
+                output=self._output,  # Overrides verbose/markdown
+                reflection=self._reflection,  # Overrides self_reflect/max_reflect/min_reflect/reflect_llm
+                caching=self._caching,  # Overrides cache
+                knowledge=effective_knowledge,  # Overrides knowledge_sources/embedder_config
+                execution=self._execution,  # Overrides max_iter
+                guardrails=self._guardrails,
+                web=self._web,
+                hooks=self._hooks,
+                autonomy=self.autonomy,
             )
             agents.append(agent)
             

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

@@ -376,7 +376,7 @@ class SubagentDelegator:
         return result
     
     async def _create_agent(self, profile: AgentProfile, task: DelegationTask) -> Any:
-        """Create an agent from profile."""
+        """Create an agent from profile using consolidated params."""
         if self.agent_factory:
             return self.agent_factory(profile.name)
         
@@ -387,14 +387,17 @@ class SubagentDelegator:
             # Filter tools based on profile
             tools = self._get_tools_for_profile(profile)
             
+            # Use consolidated params (protocol-compliant)
+            # output='silent' replaces verbose=False
+            # execution={max_iter: N} replaces max_iter=N
             agent = Agent(
                 name=profile.name,
                 role=profile.name.title(),
                 goal=task.objective,
                 backstory=profile.system_prompt,
                 tools=tools,
-                verbose=False,
-                max_iter=min(task.max_steps, profile.max_steps),
+                output='silent',
+                execution={'max_iter': min(task.max_steps, profile.max_steps)},
             )
             
             return agent

src/praisonai-agents/praisonaiagents/approval.py

@@ -13,10 +13,35 @@ from typing import Dict, Set, Optional, Callable, Any, Literal, List
 from functools import wraps
 from contextvars import ContextVar
 from dataclasses import dataclass, field
-from rich.console import Console
-from rich.panel import Panel
-from rich.text import Text
-from rich.prompt import Confirm
+
+# Lazy import for rich components
+_rich_console = None
+_rich_panel = None
+_rich_confirm = None
+
+def _get_rich_console():
+    """Lazy import rich Console."""
+    global _rich_console
+    if _rich_console is None:
+        from rich.console import Console
+        _rich_console = Console
+    return _rich_console
+
+def _get_rich_panel():
+    """Lazy import rich Panel."""
+    global _rich_panel
+    if _rich_panel is None:
+        from rich.panel import Panel
+        _rich_panel = Panel
+    return _rich_panel
+
+def _get_rich_confirm():
+    """Lazy import rich Confirm."""
+    global _rich_confirm
+    if _rich_confirm is None:
+        from rich.prompt import Confirm
+        _rich_confirm = Confirm
+    return _rich_confirm
 
 # Global registries for approval requirements
 APPROVAL_REQUIRED_TOOLS: Set[str] = set()
@@ -308,6 +333,10 @@ def console_approval_callback(function_name: str, arguments: Dict[str, Any], ris
     
     Displays tool information and prompts user for approval via console.
     """
+    Console = _get_rich_console()
+    Panel = _get_rich_panel()
+    Confirm = _get_rich_confirm()
+    
     console = Console()
     
     # Create risk level styling

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 +520,24 @@ 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
+    
+    # Simple output mode - just print response without panels
+    simple_output: bool = False
+    
+    # Show LLM parameters (for debug mode)
+    show_parameters: bool = False
+    
+    # Status trace mode - clean inline status updates
+    # Shows: [timestamp] Calling LLM..., Executing tool..., Response: ...
+    status_trace: bool = False
+    
     def to_dict(self) -> Dict[str, Any]:
         """Convert to dictionary."""
         return {
@@ -430,6 +546,11 @@ class OutputConfig:
             "stream": self.stream,
             "metrics": self.metrics,
             "reasoning_steps": self.reasoning_steps,
+            "actions_trace": self.actions_trace,
+            "json_output": self.json_output,
+            "simple_output": self.simple_output,
+            "show_parameters": self.show_parameters,
+            "status_trace": self.status_trace,
         }
 
 
@@ -857,9 +978,259 @@ 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 +1238,7 @@ __all__ = [
     "ExecutionPreset",
     "AutonomyLevel",
     # Config classes (Agent)
+    "LearnConfig",
     "MemoryConfig",
     "KnowledgeConfig",
     "PlanningConfig",
@@ -900,4 +1272,15 @@ __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

@@ -39,8 +39,107 @@ MEMORY_URL_SCHEMES: Dict[str, str] = {
 # =============================================================================
 # Output Presets
 # =============================================================================
+# 
+# Preset Hierarchy (from least to most output):
+#   silent  → Nothing (default for SDK, max performance)
+#   text    → Response only, simple format
+#   verbose → Task + Tools inline + Response panel
+#   debug   → verbose + metrics footer (tokens, cost, model)
+#   stream  → Real-time token streaming
+#   json    → Machine-readable JSONL events
+#
+# =============================================================================
 
 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,
+    },
+    # Status preset - Shows tool calls and response without timestamps
+    # Simple progress indicator: ▸ tool → result ✓
+    "status": {
+        "verbose": False,
+        "markdown": False,
+        "stream": False,
+        "metrics": False,
+        "reasoning_steps": False,
+        "actions_trace": True,
+        "simple_output": True,  # Simple format without timestamps
+    },
+    # Trace preset - Full execution trace with timestamps
+    # Shows: [HH:MM:SS] ▸ tool → result [0.2s] ✓
+    # Ideal for debugging and monitoring
+    "trace": {
+        "verbose": False,
+        "markdown": False,
+        "stream": False,
+        "metrics": False,
+        "reasoning_steps": False,
+        "actions_trace": True,
+        "status_trace": True,  # Enable timestamped status output
+    },
+    # Verbose preset - Full interactive output with panels
+    # Shows: Task prompt, Tool calls (inline), Response panel
+    # Best for interactive terminal use
+    "verbose": {
+        "verbose": True,
+        "markdown": True,
+        "stream": False,
+        "metrics": False,
+        "reasoning_steps": False,
+    },
+    # Debug preset - trace + metrics (NO boxes)
+    # Shows: [timestamp] tool calls, metrics footer, reasoning steps
+    # Best for developers debugging agent behavior
+    "debug": {
+        "verbose": False,  # No boxes
+        "markdown": False,
+        "stream": False,
+        "metrics": True,  # Shows metrics footer
+        "reasoning_steps": True,
+        "actions_trace": True,
+        "status_trace": True,  # Timestamps like trace
+        "show_parameters": True,
+    },
+    # Streaming preset - enables streaming by default
+    "stream": {
+        "verbose": True,
+        "markdown": True,
+        "stream": True,
+        "metrics": False,
+        "reasoning_steps": False,
+    },
+    # JSON preset - JSONL output for piping/scripting
+    "json": {
+        "verbose": False,
+        "markdown": False,
+        "stream": False,
+        "metrics": False,
+        "reasoning_steps": False,
+        "actions_trace": True,
+        "json_output": True,  # Output as JSONL
+    },
+    
+    # ==========================================================================
+    # ALIASES (for backward compatibility)
+    # ==========================================================================
+    # plain → silent (identical behavior)
+    "plain": {
+        "verbose": False,
+        "markdown": False,
+        "stream": False,
+        "metrics": False,
+        "reasoning_steps": False,
+        "actions_trace": False,
+    },
+    # minimal → silent (identical behavior)
     "minimal": {
         "verbose": False,
         "markdown": False,
@@ -48,6 +147,7 @@ OUTPUT_PRESETS: Dict[str, Dict[str, Any]] = {
         "metrics": False,
         "reasoning_steps": False,
     },
+    # normal → verbose (consolidated)
     "normal": {
         "verbose": True,
         "markdown": True,
@@ -55,29 +155,34 @@ OUTPUT_PRESETS: Dict[str, Dict[str, Any]] = {
         "metrics": False,
         "reasoning_steps": False,
     },
-    "verbose": {
-        "verbose": True,
-        "markdown": True,
-        "stream": False,
-        "metrics": True,
-        "reasoning_steps": True,
-    },
-    "debug": {
-        "verbose": True,
-        "markdown": True,
-        "stream": False,
-        "metrics": True,
-        "reasoning_steps": True,
-    },
-    "silent": {
+    # actions → status (renamed for clarity)
+    "actions": {
         "verbose": False,
         "markdown": False,
         "stream": False,
         "metrics": False,
         "reasoning_steps": False,
+        "actions_trace": True,
+        "simple_output": True,
+    },
+    # text → status (old name, kept for backward compat)
+    "text": {
+        "verbose": False,
+        "markdown": False,
+        "stream": False,
+        "metrics": False,
+        "reasoning_steps": False,
+        "actions_trace": True,
+        "simple_output": True,
     },
 }
 
+# Default output mode - can be overridden by PRAISONAI_OUTPUT env var
+# "silent" = zero overhead, fastest performance (DEFAULT)
+# "verbose" = full interactive display
+# "debug" = verbose + metrics
+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/flow_display.py

@@ -7,19 +7,35 @@ Visual display with agents in center and tools on sides.
 from typing import Dict, List, Set, Tuple
 from collections import defaultdict
 import threading
-from rich.console import Console
-from rich.panel import Panel
-from rich.text import Text
-from rich.align import Align
-from rich.columns import Columns
-from rich.table import Table
-from rich import box
+
+# Lazy imports for rich components
+_rich_cache = {}
+
+def _get_rich_components():
+    """Lazy import all rich components."""
+    if not _rich_cache:
+        from rich.console import Console
+        from rich.panel import Panel
+        from rich.text import Text
+        from rich.align import Align
+        from rich.columns import Columns
+        from rich.table import Table
+        from rich import box
+        _rich_cache['Console'] = Console
+        _rich_cache['Panel'] = Panel
+        _rich_cache['Text'] = Text
+        _rich_cache['Align'] = Align
+        _rich_cache['Columns'] = Columns
+        _rich_cache['Table'] = Table
+        _rich_cache['box'] = box
+    return _rich_cache
 
 class FlowDisplay:
     """Displays agent workflow with agents centered and tools on sides."""
     
     def __init__(self):
-        self.console = Console()
+        rich = _get_rich_components()
+        self.console = rich['Console']()
         self.agents = []  # List of agents in order
         self.agent_tools = defaultdict(list)  # agent -> [tools]
         self.tracking = False
@@ -98,6 +114,13 @@ class FlowDisplay:
         
     def _display_agent_with_tools(self, agent: str):
         """Display agent with tools on the sides."""
+        rich = _get_rich_components()
+        Table = rich['Table']
+        Panel = rich['Panel']
+        Text = rich['Text']
+        Align = rich['Align']
+        box = rich['box']
+        
         tools = self.agent_tools.get(agent, [])
         
         if not tools:
@@ -149,8 +172,13 @@ class FlowDisplay:
                     
                 self.console.print(Align.center(Text("".join(arrow_parts))))
                 
-    def _create_tools_panel(self, tools: List[str]) -> Panel:
+    def _create_tools_panel(self, tools: List[str]):
         """Create a panel for tools."""
+        rich = _get_rich_components()
+        Panel = rich['Panel']
+        Text = rich['Text']
+        box = rich['box']
+        
         if not tools:
             return ""
             
@@ -173,6 +201,12 @@ class FlowDisplay:
             
     def _display_centered_node(self, label: str, color: str):
         """Display a centered node."""
+        rich = _get_rich_components()
+        Panel = rich['Panel']
+        Text = rich['Text']
+        Align = rich['Align']
+        box = rich['box']
+        
         panel = Panel(
             Text(label, style=f"white on {color}", justify="center"),
             style=f"white on {color}",
@@ -183,6 +217,9 @@ class FlowDisplay:
         
     def _display_arrow_down(self):
         """Display a downward arrow."""
+        rich = _get_rich_components()
+        Align = rich['Align']
+        
         self.console.print()
         self.console.print(Align.center("↓"))
         self.console.print()

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
@@ -379,7 +379,7 @@ Respond with ONLY a valid JSON tool call in this format:
             base_url=self.base_url,
             verbose=self.verbose,
             markdown=self.markdown,
-            self_reflect=self.self_reflect,
+            reflection=self.self_reflect,
             max_reflect=self.max_reflect,
             min_reflect=self.min_reflect,
             reasoning_steps=self.reasoning_steps,
@@ -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"""
@@ -1456,7 +1458,7 @@ Now provide your final answer using this result."""
             base_url=self.base_url,
             verbose=self.verbose,
             markdown=self.markdown,
-            self_reflect=self.self_reflect,
+            reflection=self.self_reflect,
             max_reflect=self.max_reflect,
             min_reflect=self.min_reflect,
             reasoning_steps=self.reasoning_steps
@@ -1474,7 +1476,7 @@ Now provide your final answer using this result."""
             output_pydantic=output_pydantic,
             verbose=verbose,
             markdown=markdown,
-            self_reflect=self_reflect,
+            reflection=self_reflect,
             max_reflect=max_reflect,
             min_reflect=min_reflect,
             agent_name=agent_name,
@@ -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
 
@@ -1532,6 +1548,10 @@ Now provide your final answer using this result."""
                 try:
                     # Get response from LiteLLM
                     current_time = time.time()
+                    
+                    # Trigger LLM callback for subsequent iterations (after tool calls)
+                    # Initial callback is in agent._chat_completion, so this triggers for all loop iterations
+                    execute_sync_callback('llm_start', model=self.model, agent_name=agent_name)
 
                     # If reasoning_steps is True, do a single non-streaming call
                     if reasoning_steps:
@@ -1554,6 +1574,39 @@ Now provide your final answer using this result."""
                         if self.metrics:
                             self._track_token_usage(final_response, self.model)
                         
+                        # Trigger llm_end callback with metrics for debug output
+                        llm_latency_ms = (time.time() - current_time) * 1000
+                        
+                        # Extract usage - handle both dict and ModelResponse object
+                        tokens_in = 0
+                        tokens_out = 0
+                        if isinstance(final_response, dict):
+                            usage = final_response.get("usage", {})
+                            tokens_in = usage.get("prompt_tokens", 0)
+                            tokens_out = usage.get("completion_tokens", 0)
+                        else:
+                            # ModelResponse object
+                            usage = getattr(final_response, 'usage', None)
+                            if usage:
+                                tokens_in = getattr(usage, 'prompt_tokens', 0) or 0
+                                tokens_out = getattr(usage, 'completion_tokens', 0) or 0
+                        
+                        # Calculate cost if available
+                        llm_cost = None
+                        try:
+                            llm_cost = litellm.completion_cost(completion_response=final_response)
+                        except Exception:
+                            pass
+                        
+                        execute_sync_callback(
+                            'llm_end',
+                            model=self.model,
+                            tokens_in=tokens_in,
+                            tokens_out=tokens_out,
+                            cost=llm_cost,
+                            latency_ms=llm_latency_ms
+                        )
+                        
                         # Execute callbacks and display based on verbose setting
                         generation_time_val = time.time() - current_time
                         response_content = f"Reasoning:\n{reasoning_content}\n\nAnswer:\n{response_text}" if reasoning_content else response_text
@@ -2139,6 +2192,16 @@ Now provide your final answer using this result."""
                                 
                                 logging.debug(f"[TOOL_EXEC_DEBUG] About to display tool call with message: {display_message}")
                                 display_tool_call(display_message, console=self.console)
+                            
+                            # Always trigger callback for status output (even when verbose=False)
+                            result_str = json.dumps(tool_result) if tool_result else "empty"
+                            execute_sync_callback(
+                                'tool_call',
+                                message=f"Calling function: {function_name}",
+                                tool_name=function_name,
+                                tool_input=arguments,
+                                tool_output=result_str[:200] if result_str else None
+                            )
                                 
                             # Check if this is Ollama provider
                             if self._is_ollama_provider():
@@ -2917,7 +2980,7 @@ Output MUST be JSON with 'reflection' and 'satisfactory'.
                 base_url=self.base_url,
                 verbose=self.verbose,
                 markdown=self.markdown,
-                self_reflect=self.self_reflect,
+                reflection=self.self_reflect,
                 max_reflect=self.max_reflect,
                 min_reflect=self.min_reflect,
                 reasoning_steps=self.reasoning_steps
@@ -2935,7 +2998,7 @@ Output MUST be JSON with 'reflection' and 'satisfactory'.
                 output_pydantic=output_pydantic,
                 verbose=verbose,
                 markdown=markdown,
-                self_reflect=self_reflect,
+                reflection=self_reflect,
                 max_reflect=max_reflect,
                 min_reflect=min_reflect,
                 agent_name=agent_name,
@@ -3621,12 +3684,14 @@ Output MUST be JSON with 'reflection' and 'satisfactory'.
             self.session_token_metrics = self.session_token_metrics + metrics
             
             # Track in global collector
+            # Extract provider from model string (e.g., 'openai' from 'openai/gpt-4o-mini')
+            provider = model.split('/')[0] if '/' in model else 'litellm'
             _token_collector.track_tokens(
                 model=model,
                 agent=self.current_agent_name,
                 metrics=metrics,
                 metadata={
-                    "provider": self.provider,
+                    "provider": provider,
                     "stream": False
                 }
             )
@@ -3691,9 +3756,19 @@ Output MUST be JSON with 'reflection' and 'satisfactory'.
         output_json = override_params.get('output_json')
         output_pydantic = override_params.get('output_pydantic')
         
-        # Always remove these from params as they're not native litellm parameters
-        params.pop('output_json', None)
-        params.pop('output_pydantic', None)
+        # Always remove internal PraisonAI parameters that are not native litellm parameters
+        # These are used internally by PraisonAI but should not be passed to the LLM API
+        internal_params = [
+            'output_json', 'output_pydantic',  # Structured output handling
+            'reflection', 'self_reflect', 'max_reflect', 'min_reflect', 'reflect_llm',  # Self-reflection
+            'verbose', 'markdown',  # Display options
+            'agent_name', 'agent_role', 'agent_tools',  # Agent metadata
+            'task_name', 'task_description', 'task_id',  # Task metadata
+            'execute_tool_fn', 'stream_callback', 'emit_events',  # Callbacks
+            'console',  # Rich console
+        ]
+        for param in internal_params:
+            params.pop(param, None)
         
         if output_json or output_pydantic:
             
@@ -3712,18 +3787,18 @@ Output MUST be JSON with 'reflection' and 'satisfactory'.
                     logging.debug(f"Using Gemini native structured output with schema: {json.dumps(schema, indent=2)}")
         
         # Add tool_choice="auto" when tools are provided (unless already specified)
+        # This encourages the model to use tools when appropriate
         if 'tools' in params and params['tools'] and 'tool_choice' not in params:
-            # For Gemini models, use tool_choice to encourage tool usage
-            if self._is_gemini_model():
-                try:
-                    import litellm
-                    # Check if model supports function calling before setting tool_choice
-                    if litellm.supports_function_calling(model=self.model):
-                        params['tool_choice'] = 'auto'
-                except Exception as e:
-                    # If check fails, still set tool_choice for known Gemini models
-                    logging.debug(f"Could not verify function calling support: {e}. Setting tool_choice anyway.")
+            try:
+                import litellm
+                # Check if model supports function calling before setting tool_choice
+                if litellm.supports_function_calling(model=self.model):
                     params['tool_choice'] = 'auto'
+                    logging.debug(f"Set tool_choice='auto' for model {self.model}")
+            except Exception as e:
+                # If check fails, still set tool_choice to encourage tool usage
+                logging.debug(f"Could not verify function calling support: {e}. Setting tool_choice='auto' anyway.")
+                params['tool_choice'] = 'auto'
         
         # Add web_search_options if web_search is enabled and model supports it
         if self.web_search and self._supports_web_search():

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

@@ -10,15 +10,52 @@ import logging
 import time
 import json
 import asyncio
-from typing import Any, Dict, List, Optional, Union, AsyncIterator, Iterator, Callable, Tuple
-from openai import OpenAI, AsyncOpenAI
-from openai.types.chat import ChatCompletionChunk
+from typing import Any, Dict, List, Optional, Union, AsyncIterator, Iterator, Callable, Tuple, TYPE_CHECKING
 from pydantic import BaseModel
 from dataclasses import dataclass
-from rich.console import Console
-from rich.live import Live
 import inspect
 
+# Lazy imports for optional dependencies
+_openai_module = None
+_rich_console = None
+_rich_live = None
+
+def _get_openai():
+    """Lazy import openai module."""
+    global _openai_module
+    if _openai_module is None:
+        try:
+            import openai as _openai
+            _openai_module = _openai
+        except ImportError:
+            raise ImportError(
+                "openai is required for OpenAI client. "
+                "Install with: pip install openai"
+            )
+    return _openai_module
+
+def _get_openai_classes():
+    """Get OpenAI and AsyncOpenAI classes lazily."""
+    openai = _get_openai()
+    return openai.OpenAI, openai.AsyncOpenAI
+
+def _get_rich_console():
+    """Lazy import rich Console."""
+    global _rich_console
+    if _rich_console is None:
+        from rich.console import Console
+        _rich_console = Console
+    return _rich_console
+
+def _get_rich_live():
+    """Lazy import rich Live."""
+    global _rich_live
+    if _rich_live is None:
+        from rich.live import Live
+        _rich_live = Live
+    return _rich_live
+
+
 # Import display_tool_call for callback support (lazy import to avoid circular imports)
 _display_tool_call = None
 def _get_display_tool_call():
@@ -258,21 +295,23 @@ class OpenAIClient:
     def console(self):
         """Lazily initialize Rich Console only when needed."""
         if self._console is None:
-            from rich.console import Console
+            Console = _get_rich_console()
             self._console = Console()
         return self._console
     
     @property
-    def sync_client(self) -> OpenAI:
+    def sync_client(self):
         """Get the synchronous OpenAI client (lazy initialization)."""
         if self._sync_client is None:
+            OpenAI, _ = _get_openai_classes()
             self._sync_client = OpenAI(api_key=self.api_key, base_url=self.base_url)
         return self._sync_client
     
     @property
-    def async_client(self) -> AsyncOpenAI:
+    def async_client(self):
         """Get the asynchronous OpenAI client (lazy initialization)."""
         if self._async_client is None:
+            _, AsyncOpenAI = _get_openai_classes()
             self._async_client = AsyncOpenAI(api_key=self.api_key, base_url=self.base_url)
         return self._async_client
     
@@ -579,9 +618,11 @@ class OpenAIClient:
         temperature: float = 1.0,
         tools: Optional[List[Dict]] = None,
         start_time: Optional[float] = None,
-        console: Optional[Console] = None,
+        console: Optional[Any] = None,
         display_fn: Optional[Callable] = None,
         reasoning_steps: bool = False,
+        stream_callback: Optional[Callable] = None,
+        emit_events: bool = False,
         **kwargs
     ) -> Optional[ChatCompletion]:
         """
@@ -596,11 +637,18 @@ 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 +656,15 @@ 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 +672,82 @@ 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:
+                Live = _get_rich_live()
                 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 +758,61 @@ 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
@@ -666,9 +828,11 @@ class OpenAIClient:
         temperature: float = 1.0,
         tools: Optional[List[Dict]] = None,
         start_time: Optional[float] = None,
-        console: Optional[Console] = None,
+        console: Optional[Any] = None,
         display_fn: Optional[Callable] = None,
         reasoning_steps: bool = False,
+        stream_callback: Optional[Callable] = None,
+        emit_events: bool = False,
         **kwargs
     ) -> Optional[ChatCompletion]:
         """
@@ -683,11 +847,18 @@ 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 +866,19 @@ 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 +886,82 @@ 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:
+                Live = _get_rich_live()
                 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 +975,50 @@ 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
@@ -749,13 +1030,13 @@ 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,
         tool_choice: Optional[Union[str, Dict[str, Any]]] = None,
         **kwargs
-    ) -> Union[Any, Iterator[ChatCompletionChunk]]:
+    ) -> Union[Any, Iterator[Any]]:
         """
         Create a chat completion using the synchronous client.
         
@@ -794,13 +1075,13 @@ 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,
         tool_choice: Optional[Union[str, Dict[str, Any]]] = None,
         **kwargs
-    ) -> Union[Any, AsyncIterator[ChatCompletionChunk]]:
+    ) -> Union[Any, AsyncIterator[Any]]:
         """
         Create a chat completion using the asynchronous client.
         
@@ -839,12 +1120,12 @@ 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,
         stream: bool = True,
-        console: Optional[Console] = None,
+        console: Optional[Any] = None,
         display_fn: Optional[Callable] = None,
         reasoning_steps: bool = False,
         verbose: bool = True,
@@ -886,6 +1167,10 @@ class OpenAIClient:
         iteration_count = 0
         
         while iteration_count < max_iterations:
+            # Trigger LLM callback for status/trace output
+            from ..main import execute_sync_callback
+            execute_sync_callback('llm_start', model=model, agent_name=None)
+            
             if stream:
                 # Process as streaming response with formatted tools
                 final_response = self.process_stream_response(
@@ -904,7 +1189,10 @@ 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
+                        Live = _get_rich_live()
+                        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 +1210,11 @@ 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))
                             
@@ -954,6 +1247,35 @@ class OpenAIClient:
             if not final_response:
                 return None
             
+            # Trigger llm_end callback with metrics for debug output
+            llm_end_time = time.perf_counter()
+            llm_latency_ms = (llm_end_time - start_time) * 1000
+            
+            # Extract usage info if available
+            usage = getattr(final_response, 'usage', None)
+            tokens_in = getattr(usage, 'prompt_tokens', 0) if usage else 0
+            tokens_out = getattr(usage, 'completion_tokens', 0) if usage else 0
+            
+            # Calculate cost if litellm available
+            cost = None
+            try:
+                import litellm
+                if hasattr(final_response, 'model_dump'):
+                    cost = litellm.completion_cost(completion_response=final_response.model_dump())
+                elif isinstance(final_response, dict):
+                    cost = litellm.completion_cost(completion_response=final_response)
+            except Exception:
+                pass  # Cost calculation is optional
+            
+            execute_sync_callback(
+                'llm_end',
+                model=model,
+                tokens_in=tokens_in,
+                tokens_out=tokens_out,
+                cost=cost,
+                latency_ms=llm_latency_ms
+            )
+            
             # Check for tool calls
             tool_calls = getattr(final_response.choices[0].message, 'tool_calls', None)
             
@@ -989,17 +1311,19 @@ class OpenAIClient:
                     
                     # Always trigger callback for tool call tracking (even when verbose=False)
                     display_tool_call_fn = _get_display_tool_call()
-                    display_tool_call_fn(f"Calling function: {function_name}", console=console if verbose else None)
-                    
-                    if verbose and console:
-                        console.print(f"[dim]Arguments:[/dim] {arguments}")
                     
                     # Execute the tool
                     tool_result = execute_tool_fn(function_name, arguments)
                     results_str = json.dumps(tool_result) if tool_result else "Function returned an empty output"
                     
-                    # Trigger callback with result
-                    display_tool_call_fn(f"Function {function_name} returned: {results_str[:200]}{'...' if len(results_str) > 200 else ''}", console=console if verbose else None)
+                    # Trigger callback with structured parameters for status output
+                    display_tool_call_fn(
+                        f"Calling function: {function_name}",
+                        console=console if verbose else None,
+                        tool_name=function_name,
+                        tool_input=arguments,
+                        tool_output=results_str[:200] if results_str else None
+                    )
                     
                     messages.append({
                         "role": "tool",
@@ -1020,12 +1344,12 @@ 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,
         stream: bool = True,
-        console: Optional[Console] = None,
+        console: Optional[Any] = None,
         display_fn: Optional[Callable] = None,
         reasoning_steps: bool = False,
         verbose: bool = True,
@@ -1079,6 +1403,7 @@ class OpenAIClient:
                 if display_fn and console:
                     # When verbose (display_fn provided), use streaming for better UX
                     try:
+                        Live = _get_rich_live()
                         with Live(display_fn("", start_time), console=console, refresh_per_second=4, transient=True) as live:
                             # Use streaming when display_fn is provided for progressive display
                             response_stream = await self.acreate_completion(
@@ -1204,7 +1529,7 @@ 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 +1681,7 @@ 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 +1715,7 @@ 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 +243,40 @@ 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
 
+    # Task is inline (less visual weight), Response keeps panel (it's the main content)
+    # Format: 📝 Task message
+    task_prefix = "[bold #FF9B9B]📝[/]"
     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(f"{task_prefix} {message}\n")
+        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(f"{task_prefix} [bold {PRAISON_COLORS['task_text']}]{message}[/]\n")
+        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 +314,11 @@ 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 +326,8 @@ 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():
@@ -251,19 +339,41 @@ def display_tool_call(message: str, console=None, tool_name: str = None, tool_in
     # Execute synchronous callbacks (always, even when console is None)
     execute_sync_callback('tool_call', message=message, tool_name=tool_name, tool_input=tool_input, tool_output=tool_output)
     
-    # Only print panel if console is provided (verbose mode)
+    # Only print 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 clean inline format - no panels, just prefixed lines
+        if tool_name:
+            # Format: ▸ tool_name(args) → result [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 inline entry
+            status_icon = "[green]✓[/]" if success else "[red]✗[/]"
+            time_str = f"[dim][{elapsed_time:.1f}s][/]" if elapsed_time else ""
+            
+            # Base tool call line
+            tool_line = f"[bold #86A789]▸[/] [#B4D4FF]{tool_name}[/]({args_str})"
+            
+            # Add output inline with arrow if available
             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) > 80:
+                    output_str = output_str[:77] + "..."
+                tool_line += f" [dim]→[/] [italic]{output_str}[/]"
+            
+            tool_line += f" {time_str} {status_icon}"
+            console.print(tool_line)
         else:
-            # Legacy format
-            console.print(Panel.fit(Text(message, style="bold cyan"), title="Tool Call", border_style="green"))
+            # Legacy format - simple inline
+            console.print(f"[bold #86A789]▸[/] {message}")
 
 def display_error(message: str, console=None):
     if not message or not message.strip():
@@ -275,7 +385,12 @@ 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 +408,73 @@ 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/__init__.py

@@ -45,6 +45,15 @@ def __getattr__(name):
     elif name == "python_type_to_json_schema":
         from .mcp_utils import python_type_to_json_schema
         return python_type_to_json_schema
+    elif name == "fix_array_schemas":
+        from .mcp_schema_utils import fix_array_schemas
+        return fix_array_schemas
+    elif name == "ThreadLocalEventLoop":
+        from .mcp_schema_utils import ThreadLocalEventLoop
+        return ThreadLocalEventLoop
+    elif name == "get_thread_local_event_loop":
+        from .mcp_schema_utils import get_thread_local_event_loop
+        return get_thread_local_event_loop
     raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
 
 __all__ = [
@@ -57,6 +66,10 @@ __all__ = [
     "function_to_mcp_schema",
     "get_tool_metadata",
     "python_type_to_json_schema",
+    # Schema utilities (lazy loaded)
+    "fix_array_schemas",
+    "ThreadLocalEventLoop",
+    "get_thread_local_event_loop",
     # Transport components (lazy loaded)
     "WebSocketMCPClient",
     "SessionManager",

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")
         )
         
@@ -507,6 +507,26 @@ class MCP:
         """
         return iter(self._tools)
     
+    def get_tools(self) -> List[Callable]:
+        """
+        Get the list of tool functions from this MCP instance.
+        
+        This method provides explicit access to the tools list, which is useful
+        when you need to inspect or manipulate the tools programmatically.
+        
+        Returns:
+            List[Callable]: List of tool functions that can be called
+            
+        Example:
+            ```python
+            mcp = MCP("npx -y @modelcontextprotocol/server-time")
+            tools = mcp.get_tools()
+            for tool in tools:
+                print(f"Tool: {tool.__name__}")
+            ```
+        """
+        return self._tools
+    
     def _fix_array_schemas(self, schema):
         """
         Fix array schemas by adding missing 'items' attribute required by OpenAI.
@@ -548,7 +568,7 @@ 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/mcp/mcp_http_stream.py

@@ -30,20 +30,19 @@ except ImportError:
 
 logger = logging.getLogger("mcp-http-stream")
 
-# Global event loop for async operations
-_event_loop = None
+# Import shared utilities for thread-safe event loop and schema fixing
+from .mcp_schema_utils import ThreadLocalEventLoop, fix_array_schemas
+
+# Thread-local event loop for async operations (thread-safe)
+_event_loop_manager = ThreadLocalEventLoop()
 
 # Global registry of active clients for cleanup
 _active_clients = weakref.WeakSet()
 _cleanup_registered = False
 
 def get_event_loop():
-    """Get or create a global event loop."""
-    global _event_loop
-    if _event_loop is None or _event_loop.is_closed():
-        _event_loop = asyncio.new_event_loop()
-        asyncio.set_event_loop(_event_loop)
-    return _event_loop
+    """Get or create a thread-local event loop."""
+    return _event_loop_manager.get_loop()
 
 
 def _cleanup_all_clients():
@@ -148,47 +147,10 @@ class HTTPStreamMCPTool:
             logger.error(f"Error in _async_call for {self.name}: {e}")
             raise
     
-    def _fix_array_schemas(self, schema):
-        """
-        Fix array schemas by adding missing 'items' attribute required by OpenAI.
-        
-        This ensures compatibility with OpenAI's function calling format which
-        requires array types to specify the type of items they contain.
-        
-        Args:
-            schema: The schema dictionary to fix
-            
-        Returns:
-            dict: The fixed schema
-        """
-        if not isinstance(schema, dict):
-            return schema
-            
-        # Create a copy to avoid modifying the original
-        fixed_schema = schema.copy()
-        
-        # Fix array types at the current level
-        if fixed_schema.get("type") == "array" and "items" not in fixed_schema:
-            # Add a default items schema for arrays without it
-            fixed_schema["items"] = {"type": "string"}
-            
-        # Recursively fix nested schemas
-        if "properties" in fixed_schema:
-            fixed_properties = {}
-            for prop_name, prop_schema in fixed_schema["properties"].items():
-                fixed_properties[prop_name] = self._fix_array_schemas(prop_schema)
-            fixed_schema["properties"] = fixed_properties
-            
-        # Fix items schema if it exists
-        if "items" in fixed_schema:
-            fixed_schema["items"] = self._fix_array_schemas(fixed_schema["items"])
-            
-        return fixed_schema
-    
     def to_openai_tool(self):
         """Convert the tool to OpenAI format."""
-        # Fix array schemas to include 'items' attribute
-        fixed_schema = self._fix_array_schemas(self.input_schema)
+        # Fix array schemas to include 'items' attribute (using shared utility)
+        fixed_schema = fix_array_schemas(self.input_schema)
         
         return {
             "type": "function",

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

@@ -0,0 +1,189 @@
+"""
+MCP Schema Utilities.
+
+This module provides shared schema manipulation utilities for MCP transports.
+These utilities ensure compatibility with OpenAI's function calling format
+and other LLM providers.
+
+DRY: This module centralizes schema fixing logic that was previously
+duplicated across mcp.py, mcp_sse.py, mcp_http_stream.py, and mcp_websocket.py.
+"""
+
+import threading
+from typing import Any, Dict, Optional
+
+
+def fix_array_schemas(schema: Any) -> Any:
+    """
+    Fix array schemas by adding missing 'items' attribute required by OpenAI.
+    
+    This ensures compatibility with OpenAI's function calling format which
+    requires array types to specify the type of items they contain.
+    
+    Args:
+        schema: The schema dictionary to fix
+        
+    Returns:
+        dict: The fixed schema with 'items' added to array types
+        
+    Example:
+        >>> schema = {"type": "array"}
+        >>> fixed = fix_array_schemas(schema)
+        >>> fixed
+        {'type': 'array', 'items': {'type': 'string'}}
+    """
+    if not isinstance(schema, dict):
+        return schema
+        
+    # Create a copy to avoid modifying the original
+    fixed_schema = schema.copy()
+    
+    # Fix array types at the current level
+    if fixed_schema.get("type") == "array" and "items" not in fixed_schema:
+        # Add a default items schema for arrays without it
+        fixed_schema["items"] = {"type": "string"}
+        
+    # Recursively fix nested schemas
+    if "properties" in fixed_schema:
+        fixed_properties = {}
+        for prop_name, prop_schema in fixed_schema["properties"].items():
+            fixed_properties[prop_name] = fix_array_schemas(prop_schema)
+        fixed_schema["properties"] = fixed_properties
+        
+    # Fix items schema if it exists
+    if "items" in fixed_schema:
+        fixed_schema["items"] = fix_array_schemas(fixed_schema["items"])
+        
+    # Fix additionalProperties if it's a schema
+    if isinstance(fixed_schema.get("additionalProperties"), dict):
+        fixed_schema["additionalProperties"] = fix_array_schemas(
+            fixed_schema["additionalProperties"]
+        )
+    
+    # Fix anyOf/oneOf/allOf schemas
+    for key in ("anyOf", "oneOf", "allOf"):
+        if key in fixed_schema and isinstance(fixed_schema[key], list):
+            fixed_schema[key] = [fix_array_schemas(s) for s in fixed_schema[key]]
+        
+    return fixed_schema
+
+
+class ThreadLocalEventLoop:
+    """
+    Thread-local event loop storage for MCP transports.
+    
+    This class provides thread-safe event loop management, ensuring that
+    multiple MCP instances don't share the same event loop, which could
+    cause race conditions.
+    
+    Usage:
+        event_loop_manager = ThreadLocalEventLoop()
+        loop = event_loop_manager.get_loop()
+        # Use loop for async operations
+        event_loop_manager.cleanup()
+    """
+    
+    def __init__(self):
+        """Initialize thread-local storage."""
+        self._local = threading.local()
+        self._lock = threading.Lock()
+    
+    def get_loop(self):
+        """
+        Get or create an event loop for the current thread.
+        
+        Returns:
+            asyncio.AbstractEventLoop: Event loop for current thread
+        """
+        import asyncio
+        
+        # Check if we have a loop for this thread
+        loop = getattr(self._local, 'loop', None)
+        
+        if loop is None or loop.is_closed():
+            with self._lock:
+                # Double-check after acquiring lock
+                loop = getattr(self._local, 'loop', None)
+                if loop is None or loop.is_closed():
+                    loop = asyncio.new_event_loop()
+                    self._local.loop = loop
+        
+        return loop
+    
+    def set_loop(self, loop):
+        """
+        Set the event loop for the current thread.
+        
+        Args:
+            loop: The event loop to set
+        """
+        self._local.loop = loop
+    
+    def cleanup(self):
+        """Clean up the event loop for the current thread."""
+        loop = getattr(self._local, 'loop', None)
+        if loop is not None and not loop.is_closed():
+            try:
+                loop.close()
+            except Exception:
+                pass
+        self._local.loop = None
+
+
+# Singleton instance for backward compatibility with existing code
+# that uses get_event_loop() function pattern
+_default_event_loop_manager = ThreadLocalEventLoop()
+
+
+def get_thread_local_event_loop():
+    """
+    Get or create a thread-local event loop.
+    
+    This is a drop-in replacement for the global get_event_loop() functions
+    in the transport modules, providing thread safety.
+    
+    Returns:
+        asyncio.AbstractEventLoop: Event loop for current thread
+    """
+    return _default_event_loop_manager.get_loop()
+
+
+def json_type_to_python_type(json_type: str) -> type:
+    """
+    Convert JSON Schema type to Python type annotation.
+    
+    Args:
+        json_type: JSON Schema type string
+        
+    Returns:
+        Python type corresponding to the JSON type
+    """
+    type_map = {
+        'string': str,
+        'integer': int,
+        'number': float,
+        'boolean': bool,
+        'array': list,
+        'object': dict,
+        'null': type(None),
+    }
+    return type_map.get(json_type, str)
+
+
+def extract_schema_properties(input_schema: Optional[Dict[str, Any]]) -> Dict[str, Any]:
+    """
+    Extract properties and required fields from an input schema.
+    
+    Args:
+        input_schema: JSON Schema for tool input
+        
+    Returns:
+        Dictionary with 'properties' and 'required' keys
+    """
+    if not input_schema:
+        return {'properties': {}, 'required': []}
+    
+    return {
+        'properties': input_schema.get('properties', {}),
+        'required': input_schema.get('required', [])
+    }

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

@@ -22,16 +22,15 @@ except ImportError:
 
 logger = logging.getLogger("mcp-sse")
 
-# Global event loop for async operations
-_event_loop = None
+# Import shared utilities for thread-safe event loop and schema fixing
+from .mcp_schema_utils import ThreadLocalEventLoop, fix_array_schemas
+
+# Thread-local event loop for async operations (thread-safe)
+_event_loop_manager = ThreadLocalEventLoop()
 
 def get_event_loop():
-    """Get or create a global event loop."""
-    global _event_loop
-    if _event_loop is None or _event_loop.is_closed():
-        _event_loop = asyncio.new_event_loop()
-        asyncio.set_event_loop(_event_loop)
-    return _event_loop
+    """Get or create a thread-local event loop."""
+    return _event_loop_manager.get_loop()
 
 
 class SSEMCPTool:
@@ -111,47 +110,10 @@ class SSEMCPTool:
             logger.error(f"Error in _async_call for {self.name}: {e}")
             raise
     
-    def _fix_array_schemas(self, schema):
-        """
-        Fix array schemas by adding missing 'items' attribute required by OpenAI.
-        
-        This ensures compatibility with OpenAI's function calling format which
-        requires array types to specify the type of items they contain.
-        
-        Args:
-            schema: The schema dictionary to fix
-            
-        Returns:
-            dict: The fixed schema
-        """
-        if not isinstance(schema, dict):
-            return schema
-            
-        # Create a copy to avoid modifying the original
-        fixed_schema = schema.copy()
-        
-        # Fix array types at the current level
-        if fixed_schema.get("type") == "array" and "items" not in fixed_schema:
-            # Add a default items schema for arrays without it
-            fixed_schema["items"] = {"type": "string"}
-            
-        # Recursively fix nested schemas
-        if "properties" in fixed_schema:
-            fixed_properties = {}
-            for prop_name, prop_schema in fixed_schema["properties"].items():
-                fixed_properties[prop_name] = self._fix_array_schemas(prop_schema)
-            fixed_schema["properties"] = fixed_properties
-            
-        # Fix items schema if it exists
-        if "items" in fixed_schema:
-            fixed_schema["items"] = self._fix_array_schemas(fixed_schema["items"])
-            
-        return fixed_schema
-    
     def to_openai_tool(self):
         """Convert the tool to OpenAI format."""
-        # Fix array schemas to include 'items' attribute
-        fixed_schema = self._fix_array_schemas(self.input_schema)
+        # Fix array schemas to include 'items' attribute (using shared utility)
+        fixed_schema = fix_array_schemas(self.input_schema)
         
         return {
             "type": "function",

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

@@ -26,6 +26,9 @@ from typing import Any, Dict, Optional, List
 
 logger = logging.getLogger("mcp-websocket")
 
+# Import shared utilities for thread-safe event loop and schema fixing
+from .mcp_schema_utils import ThreadLocalEventLoop, fix_array_schemas
+
 
 def is_websocket_url(url: str) -> bool:
     """
@@ -58,18 +61,13 @@ def calculate_backoff(attempt: int, base_delay: float = 1.0, max_delay: float =
     return min(delay, max_delay)
 
 
-# Global event loop for WebSocket operations (similar to other MCP transports)
-_event_loop = None
-_loop_lock = threading.Lock()
+# Thread-local event loop for WebSocket operations (thread-safe)
+_event_loop_manager = ThreadLocalEventLoop()
 
 
 def get_event_loop():
-    """Get or create a global event loop for WebSocket operations."""
-    global _event_loop
-    with _loop_lock:
-        if _event_loop is None or _event_loop.is_closed():
-            _event_loop = asyncio.new_event_loop()
-        return _event_loop
+    """Get or create a thread-local event loop for WebSocket operations."""
+    return _event_loop_manager.get_loop()
 
 
 class WebSocketTransport:
@@ -359,36 +357,10 @@ class WebSocketMCPTool:
             logger.error(f"Error in _async_call for {self.name}: {e}")
             raise
     
-    def _fix_array_schemas(self, schema: Dict[str, Any]) -> Dict[str, Any]:
-        """
-        Fix array schemas by adding missing 'items' attribute.
-        
-        OpenAI requires array types to specify item types.
-        """
-        if not isinstance(schema, dict):
-            return schema
-        
-        fixed_schema = schema.copy()
-        
-        # Fix array types at current level
-        if fixed_schema.get("type") == "array" and "items" not in fixed_schema:
-            fixed_schema["items"] = {"type": "string"}
-        
-        # Recursively fix nested schemas
-        if "properties" in fixed_schema:
-            fixed_properties = {}
-            for prop_name, prop_schema in fixed_schema["properties"].items():
-                fixed_properties[prop_name] = self._fix_array_schemas(prop_schema)
-            fixed_schema["properties"] = fixed_properties
-        
-        if "items" in fixed_schema:
-            fixed_schema["items"] = self._fix_array_schemas(fixed_schema["items"])
-        
-        return fixed_schema
-    
     def to_openai_tool(self) -> Dict[str, Any]:
         """Convert the tool to OpenAI function calling format."""
-        fixed_schema = self._fix_array_schemas(self.input_schema)
+        # Fix array schemas to include 'items' attribute (using shared utility)
+        fixed_schema = fix_array_schemas(self.input_schema)
         
         return {
             "type": "function",

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

@@ -18,6 +18,11 @@ This module provides memory management capabilities including:
 - Docs management (like Cursor docs)
 - MCP config management (like Cursor .cursor/mcp/)
 
+Protocols:
+- MemoryProtocol: Minimal interface for memory implementations
+- AsyncMemoryProtocol: Async interface for memory
+- ResettableMemoryProtocol: Interface with reset methods
+
 Memory Providers:
 - FileMemory: Zero-dependency JSON file-based storage (default)
 - Memory: Full-featured with SQLite, ChromaDB, Mem0, MongoDB support
@@ -33,6 +38,13 @@ from .file_memory import FileMemory, create_memory
 from .rules_manager import RulesManager, Rule, create_rules_manager
 from .docs_manager import DocsManager, Doc
 from .mcp_config import MCPConfigManager, MCPConfig
+from .protocols import (
+    MemoryProtocol, 
+    AsyncMemoryProtocol, 
+    ResettableMemoryProtocol,
+    EntityMemoryProtocol,
+)
+
 
 # Lazy imports for optional modules to avoid dependency issues and improve startup time
 def __getattr__(name):
@@ -110,6 +122,31 @@ 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 +190,18 @@ __all__ = [
     # MCP config management
     "MCPConfigManager",
     "MCPConfig",
+    # Learn module
+    "LearnManager",
+    "PersonaStore",
+    "InsightStore",
+    "ThreadStore",
+    "PatternStore",
+    "DecisionStore",
+    "FeedbackStore",
+    "ImprovementStore",
+    # Protocols
+    "MemoryProtocol",
+    "AsyncMemoryProtocol",
+    "ResettableMemoryProtocol",
+    "EntityMemoryProtocol",
 ] 

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

@@ -18,7 +18,7 @@ Storage Structure:
 
 import json
 import time
-import fcntl
+import sys
 import hashlib
 import logging
 from pathlib import Path
@@ -26,9 +26,17 @@ from datetime import datetime, timedelta
 from typing import Any, Dict, List, Optional
 from dataclasses import dataclass, asdict, field
 
+# fcntl is Unix-only; on Windows, skip file locking (acceptable for single-process usage)
+if sys.platform != 'win32':
+    import fcntl
+    _HAS_FCNTL = True
+else:
+    _HAS_FCNTL = False
+
 logger = logging.getLogger(__name__)
 
 
+
 @dataclass
 class MemoryItem:
     """A single memory item."""
@@ -178,35 +186,41 @@ class FileMemory:
     # -------------------------------------------------------------------------
     
     def _read_json(self, filepath: Path, default: Any = None) -> Any:
-        """Read JSON file with file locking."""
+        """Read JSON file with file locking (Unix only)."""
         if not filepath.exists():
             return default if default is not None else []
         
         try:
             with open(filepath, 'r', encoding='utf-8') as f:
-                fcntl.flock(f.fileno(), fcntl.LOCK_SH)
+                if _HAS_FCNTL:
+                    fcntl.flock(f.fileno(), fcntl.LOCK_SH)
                 try:
                     data = json.load(f)
                 finally:
-                    fcntl.flock(f.fileno(), fcntl.LOCK_UN)
+                    if _HAS_FCNTL:
+                        fcntl.flock(f.fileno(), fcntl.LOCK_UN)
                 return data
         except (json.JSONDecodeError, IOError) as e:
             self._log(f"Error reading {filepath}: {e}", logging.WARNING)
             return default if default is not None else []
+
     
     def _write_json(self, filepath: Path, data: Any) -> bool:
-        """Write JSON file with file locking."""
+        """Write JSON file with file locking (Unix only)."""
         try:
             with open(filepath, 'w', encoding='utf-8') as f:
-                fcntl.flock(f.fileno(), fcntl.LOCK_EX)
+                if _HAS_FCNTL:
+                    fcntl.flock(f.fileno(), fcntl.LOCK_EX)
                 try:
                     json.dump(data, f, indent=2, ensure_ascii=False)
                 finally:
-                    fcntl.flock(f.fileno(), fcntl.LOCK_UN)
+                    if _HAS_FCNTL:
+                        fcntl.flock(f.fileno(), fcntl.LOCK_UN)
             return True
         except IOError as e:
             self._log(f"Error writing {filepath}: {e}", logging.ERROR)
             return False
+
     
     def _load_config(self, overrides: Optional[Dict[str, Any]] = None) -> Dict[str, Any]:
         """Load or create configuration."""