Release v2.7.0

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>=2.6.8" \
+    "praisonai>=2.7.0" \
     "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>=2.6.8" \
+    "praisonai>=2.7.0" \
     "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>=2.6.8" \
+    "praisonai>=2.7.0" \
     "praisonai[ui]" \
     "praisonai[crewai]"
 

examples/performance/benchmark_example.py

@@ -0,0 +1,108 @@
+"""
+Performance Benchmark Example for PraisonAI.
+
+This example demonstrates how to measure and verify performance metrics
+for PraisonAI CLI and SDK imports.
+
+Usage:
+    python benchmark_example.py
+"""
+
+import time
+import tracemalloc
+import sys
+
+
+def measure_import_time(module_name: str) -> float:
+    """Measure import time for a module."""
+    # Clear from cache if already imported
+    if module_name in sys.modules:
+        del sys.modules[module_name]
+    
+    start = time.time()
+    __import__(module_name)
+    elapsed = time.time() - start
+    return elapsed * 1000  # Return milliseconds
+
+
+def measure_memory_usage(import_func) -> tuple:
+    """Measure memory usage during import."""
+    tracemalloc.start()
+    import_func()
+    current, peak = tracemalloc.get_traced_memory()
+    tracemalloc.stop()
+    return current / 1024 / 1024, peak / 1024 / 1024  # Return MB
+
+
+def main():
+    print("=" * 60)
+    print("PraisonAI Performance Benchmark")
+    print("=" * 60)
+    
+    # Test 1: CLI Import Time
+    print("\n1. CLI Import Time")
+    print("-" * 40)
+    
+    import os
+    os.environ['LOGLEVEL'] = 'WARNING'
+    
+    start = time.time()
+    from praisonai.cli.main import PraisonAI
+    cli_import_time = (time.time() - start) * 1000
+    
+    print(f"   CLI import time: {cli_import_time:.0f}ms")
+    print(f"   Status: {'✓ FAST' if cli_import_time < 500 else '✗ SLOW'}")
+    
+    # Test 2: Memory Usage
+    print("\n2. Memory Usage")
+    print("-" * 40)
+    
+    tracemalloc.start()
+    # Force re-import by clearing cache
+    for mod in list(sys.modules.keys()):
+        if 'praisonai' in mod:
+            del sys.modules[mod]
+    
+    from praisonai.cli.main import PraisonAI as PraisonAI2
+    current, peak = tracemalloc.get_traced_memory()
+    tracemalloc.stop()
+    
+    print(f"   Current memory: {current/1024/1024:.1f}MB")
+    print(f"   Peak memory: {peak/1024/1024:.1f}MB")
+    print(f"   Status: {'✓ LOW' if peak/1024/1024 < 50 else '✗ HIGH'}")
+    
+    # Test 3: Lazy Loading Verification
+    print("\n3. Lazy Loading Verification")
+    print("-" * 40)
+    
+    heavy_modules = ['litellm', 'instructor', 'praisonai.auto']
+    loaded = [m for m in heavy_modules if m in sys.modules]
+    
+    if loaded:
+        print(f"   Heavy modules loaded: {loaded}")
+        print("   Status: ✗ NOT LAZY")
+    else:
+        print("   Heavy modules NOT loaded at startup")
+        print("   Status: ✓ LAZY")
+    
+    # Summary
+    print("\n" + "=" * 60)
+    print("Summary")
+    print("=" * 60)
+    
+    all_pass = (
+        cli_import_time < 500 and
+        peak/1024/1024 < 50 and
+        len(loaded) == 0
+    )
+    
+    if all_pass:
+        print("✓ All performance benchmarks PASSED")
+    else:
+        print("✗ Some benchmarks FAILED")
+    
+    return 0 if all_pass else 1
+
+
+if __name__ == "__main__":
+    sys.exit(main())

src/praisonai-agents/benchmarks/import_time.py

@@ -0,0 +1,148 @@
+#!/usr/bin/env python3
+"""
+Import time benchmark for praisonaiagents.
+
+This benchmark measures the import time of the praisonaiagents package
+and verifies that heavy dependencies are not loaded at import time.
+
+Usage:
+    python benchmarks/import_time.py
+    
+CI Gate:
+    - PASS: import time < 200ms
+    - WARN: import time 200-300ms
+    - FAIL: import time > 300ms
+"""
+import sys
+import time
+import statistics
+
+
+def clear_modules():
+    """Clear all praisonai and litellm related modules from cache."""
+    to_remove = [m for m in sys.modules.keys() 
+                 if 'praison' in m or 'litellm' in m]
+    for mod in to_remove:
+        del sys.modules[mod]
+
+
+def measure_import_time(runs: int = 5) -> dict:
+    """
+    Measure import time over multiple runs.
+    
+    Returns:
+        dict with min, max, median, mean times in milliseconds
+    """
+    times = []
+    
+    for _ in range(runs):
+        clear_modules()
+        
+        start = time.perf_counter()
+        import praisonaiagents  # noqa: F401
+        end = time.perf_counter()
+        
+        times.append((end - start) * 1000)
+        
+        # Clean up for next run
+        if 'praisonaiagents' in sys.modules:
+            del sys.modules['praisonaiagents']
+    
+    times.sort()
+    
+    return {
+        'times': times,
+        'min': min(times),
+        'max': max(times),
+        'median': statistics.median(times),
+        'mean': statistics.mean(times),
+        'stdev': statistics.stdev(times) if len(times) > 1 else 0,
+    }
+
+
+def check_lazy_imports() -> dict:
+    """
+    Check that heavy dependencies are NOT loaded after importing praisonaiagents.
+    
+    Returns:
+        dict with module names and whether they are loaded
+    """
+    clear_modules()
+    
+    import praisonaiagents  # noqa: F401
+    
+    heavy_modules = [
+        'litellm',
+        'chromadb', 
+        'mem0',
+        'requests',
+    ]
+    
+    results = {}
+    for mod in heavy_modules:
+        results[mod] = mod in sys.modules
+    
+    return results
+
+
+def main():
+    """Run the benchmark and print results."""
+    print("=" * 60)
+    print("PraisonAI Agents Import Time Benchmark")
+    print("=" * 60)
+    
+    # Measure import time
+    print("\n[1/2] Measuring import time (5 runs)...")
+    results = measure_import_time(runs=5)
+    
+    print(f"\nImport times (ms): {[round(t, 1) for t in results['times']]}")
+    print(f"  Min:    {results['min']:.1f} ms")
+    print(f"  Max:    {results['max']:.1f} ms")
+    print(f"  Median: {results['median']:.1f} ms")
+    print(f"  Mean:   {results['mean']:.1f} ms")
+    print(f"  StdDev: {results['stdev']:.1f} ms")
+    
+    # Check lazy imports
+    print("\n[2/2] Checking lazy imports...")
+    lazy_results = check_lazy_imports()
+    
+    print("\nHeavy modules loaded at import time:")
+    all_lazy = True
+    for mod, loaded in lazy_results.items():
+        status = "LOADED (BAD)" if loaded else "NOT LOADED (GOOD)"
+        print(f"  {mod}: {status}")
+        if loaded:
+            all_lazy = False
+    
+    # Determine pass/fail
+    print("\n" + "=" * 60)
+    print("RESULTS")
+    print("=" * 60)
+    
+    median_time = results['median']
+    
+    if median_time < 200:
+        time_status = "PASS"
+        time_msg = f"Import time {median_time:.0f}ms < 200ms target"
+    elif median_time < 300:
+        time_status = "WARN"
+        time_msg = f"Import time {median_time:.0f}ms between 200-300ms"
+    else:
+        time_status = "FAIL"
+        time_msg = f"Import time {median_time:.0f}ms > 300ms limit"
+    
+    lazy_status = "PASS" if all_lazy else "FAIL"
+    lazy_msg = "All heavy deps lazy loaded" if all_lazy else "Some heavy deps loaded eagerly"
+    
+    print(f"\nImport Time: [{time_status}] {time_msg}")
+    print(f"Lazy Imports: [{lazy_status}] {lazy_msg}")
+    
+    overall = "PASS" if time_status != "FAIL" and lazy_status == "PASS" else "FAIL"
+    print(f"\nOverall: [{overall}]")
+    
+    # Exit code for CI
+    return 0 if overall == "PASS" else 1
+
+
+if __name__ == "__main__":
+    sys.exit(main())

src/praisonai-agents/benchmarks/memory_usage.py

@@ -0,0 +1,89 @@
+#!/usr/bin/env python3
+"""
+Memory usage benchmark for praisonaiagents.
+
+This benchmark measures the memory footprint of importing praisonaiagents
+and verifies it stays within acceptable limits.
+
+Usage:
+    python benchmarks/memory_usage.py
+    
+CI Gate:
+    - PASS: memory < 30MB
+    - WARN: memory 30-45MB
+    - FAIL: memory > 45MB
+"""
+import sys
+import tracemalloc
+
+
+def clear_modules():
+    """Clear all praisonai and litellm related modules from cache."""
+    to_remove = [m for m in sys.modules.keys() 
+                 if 'praison' in m or 'litellm' in m]
+    for mod in to_remove:
+        del sys.modules[mod]
+
+
+def measure_memory() -> dict:
+    """
+    Measure memory usage after importing praisonaiagents.
+    
+    Returns:
+        dict with current and peak memory in MB
+    """
+    clear_modules()
+    
+    tracemalloc.start()
+    
+    import praisonaiagents  # noqa: F401
+    
+    current, peak = tracemalloc.get_traced_memory()
+    tracemalloc.stop()
+    
+    return {
+        'current_bytes': current,
+        'peak_bytes': peak,
+        'current_mb': current / 1024 / 1024,
+        'peak_mb': peak / 1024 / 1024,
+    }
+
+
+def main():
+    """Run the benchmark and print results."""
+    print("=" * 60)
+    print("PraisonAI Agents Memory Usage Benchmark")
+    print("=" * 60)
+    
+    print("\nMeasuring memory usage...")
+    results = measure_memory()
+    
+    print(f"\nMemory Usage:")
+    print(f"  Current: {results['current_mb']:.1f} MB")
+    print(f"  Peak:    {results['peak_mb']:.1f} MB")
+    
+    # Determine pass/fail based on peak memory
+    peak_mb = results['peak_mb']
+    
+    print("\n" + "=" * 60)
+    print("RESULTS")
+    print("=" * 60)
+    
+    if peak_mb < 30:
+        status = "PASS"
+        msg = f"Memory {peak_mb:.1f}MB < 30MB target"
+    elif peak_mb < 45:
+        status = "WARN"
+        msg = f"Memory {peak_mb:.1f}MB between 30-45MB"
+    else:
+        status = "FAIL"
+        msg = f"Memory {peak_mb:.1f}MB > 45MB limit"
+    
+    print(f"\nMemory: [{status}] {msg}")
+    
+    # Exit code for CI
+    return 0 if status != "FAIL" else 1
+
+
+if __name__ == "__main__":
+    sys.exit(main())

src/praisonai-agents/examples/lite_agent_example.py

@@ -0,0 +1,123 @@
+#!/usr/bin/env python3
+"""
+Example: Using PraisonAI Agents Lite Package
+
+The lite package provides a minimal agent framework without heavy dependencies
+like litellm. It's ideal for users who want to bring their own LLM client.
+
+This example shows:
+1. Creating a LiteAgent with a custom LLM function
+2. Using the built-in OpenAI adapter
+3. Thread-safe chat history management
+"""
+import os
+
+
+def example_custom_llm():
+    """Example 1: Using a custom LLM function."""
+    from praisonaiagents.lite import LiteAgent
+    
+    # Define your own LLM function
+    def my_simple_llm(messages):
+        """A mock LLM that echoes back the last message."""
+        last_msg = messages[-1]["content"] if messages else "No message"
+        return f"Echo: {last_msg}"
+    
+    agent = LiteAgent(
+        name="EchoAgent",
+        llm_fn=my_simple_llm,
+        instructions="You are a simple echo agent."
+    )
+    
+    response = agent.chat("Hello, world!")
+    print(f"Response: {response}")
+    print(f"Chat history length: {len(agent.chat_history)}")
+
+
+def example_openai_adapter():
+    """Example 2: Using the built-in OpenAI adapter."""
+    from praisonaiagents.lite import LiteAgent, create_openai_llm_fn
+    
+    # Check for API key
+    if not os.environ.get("OPENAI_API_KEY"):
+        print("Skipping OpenAI example - OPENAI_API_KEY not set")
+        return
+    
+    # Create LLM function using OpenAI SDK directly (no litellm)
+    llm_fn = create_openai_llm_fn(model="gpt-4o-mini")
+    
+    agent = LiteAgent(
+        name="OpenAIAgent",
+        llm_fn=llm_fn,
+        instructions="You are a helpful assistant. Keep responses brief."
+    )
+    
+    response = agent.chat("What is 2+2?")
+    print(f"OpenAI Response: {response}")
+
+
+def example_with_tools():
+    """Example 3: Using tools with LiteAgent."""
+    from praisonaiagents.lite import LiteAgent, tool
+    
+    @tool
+    def add_numbers(a: int, b: int) -> int:
+        """Add two numbers together."""
+        return a + b
+    
+    def mock_llm(messages):
+        return "The sum is 8"
+    
+    agent = LiteAgent(
+        name="MathAgent",
+        llm_fn=mock_llm,
+        tools=[add_numbers]
+    )
+    
+    # Execute tool directly
+    result = agent.execute_tool("add_numbers", a=5, b=3)
+    print(f"Tool result: {result.output}")
+    print(f"Tool success: {result.success}")
+
+
+def example_anthropic_adapter():
+    """Example 4: Using the built-in Anthropic adapter."""
+    from praisonaiagents.lite import LiteAgent, create_anthropic_llm_fn
+    
+    # Check for API key
+    if not os.environ.get("ANTHROPIC_API_KEY"):
+        print("Skipping Anthropic example - ANTHROPIC_API_KEY not set")
+        return
+    
+    # Create LLM function using Anthropic SDK directly
+    llm_fn = create_anthropic_llm_fn(model="claude-3-5-sonnet-20241022")
+    
+    agent = LiteAgent(
+        name="ClaudeAgent",
+        llm_fn=llm_fn,
+        instructions="You are a helpful assistant. Keep responses brief."
+    )
+    
+    response = agent.chat("What is the capital of France?")
+    print(f"Claude Response: {response}")
+
+
+if __name__ == "__main__":
+    print("=" * 60)
+    print("PraisonAI Agents Lite Examples")
+    print("=" * 60)
+    
+    print("\n[1] Custom LLM Function:")
+    example_custom_llm()
+    
+    print("\n[2] OpenAI Adapter:")
+    example_openai_adapter()
+    
+    print("\n[3] Tools with LiteAgent:")
+    example_with_tools()
+    
+    print("\n[4] Anthropic Adapter:")
+    example_anthropic_adapter()
+    
+    print("\n" + "=" * 60)
+    print("Examples completed!")

src/praisonai-agents/praisonaiagents/__init__.py

@@ -1,5 +1,8 @@
 """
 Praison AI Agents - A package for hierarchical AI agent task execution
+
+This module uses lazy loading to minimize import time and memory usage.
+Heavy dependencies like litellm are only loaded when actually needed.
 """
 
 # Apply warning patch BEFORE any imports to intercept warnings at the source
@@ -11,51 +14,18 @@ from . import _logging
 # Configure root logger after logging is initialized
 _logging.configure_root_logger()
 
-# Now import everything else
-from .agent.agent import Agent
-from .agent.image_agent import ImageAgent
-from .agent.context_agent import ContextAgent, create_context_agent
-from .agent.deep_research_agent import (
-    DeepResearchAgent,
-    DeepResearchResponse,
-    Citation,
-    ReasoningStep,
-    WebSearchCall,
-    CodeExecutionStep,
-    MCPCall,
-    FileSearchCall,
-    Provider
-)
-from .agent.query_rewriter_agent import (
-    QueryRewriterAgent,
-    RewriteStrategy,
-    RewriteResult
-)
-from .agent.prompt_expander_agent import (
-    PromptExpanderAgent,
-    ExpandStrategy,
-    ExpandResult
-)
-from .agents.agents import PraisonAIAgents
-from .task.task import Task
+# Import configuration (lightweight, no heavy deps)
+from . import _config
+
+# Lightweight imports that don't trigger heavy dependency chains
 from .tools.tools import Tools
 from .tools.base import BaseTool, ToolResult, ToolValidationError, validate_tool
 from .tools.decorator import tool, FunctionTool
 from .tools.registry import get_registry, register_tool, get_tool, ToolRegistry
-from .agents.autoagents import AutoAgents
-# Knowledge is lazy-loaded to avoid importing heavy deps (chromadb, mem0) at startup
-# MCP support (optional)
-try:
-    from .mcp.mcp import MCP
-    _mcp_available = True
-except ImportError:
-    _mcp_available = False
-    MCP = None
-from .session import Session
-# Memory is lazy-loaded to avoid importing chromadb at startup
 from .db import db
 from .obs import obs
-# Workflows - import from dedicated workflows module
+
+# Workflows - lightweight module
 from .workflows import (
     Workflow, WorkflowStep, WorkflowContext, StepResult,
     Route, Parallel, Loop, Repeat,
@@ -64,80 +34,18 @@ from .workflows import (
 )
 from .guardrails import GuardrailResult, LLMGuardrail
 
-# Fast Context support (lazy loaded to avoid performance impact)
-def __getattr__(name):
-    """Lazy load heavy modules to avoid impacting package load time."""
-    # Knowledge module (imports chromadb, mem0)
-    if name == "Knowledge":
-        from praisonaiagents.knowledge.knowledge import Knowledge
-        return Knowledge
-    elif name == "Chunking":
-        from praisonaiagents.knowledge.chunking import Chunking
-        return Chunking
-    # FastContext support
-    elif name == "FastContext":
-        from praisonaiagents.context.fast import FastContext
-        return FastContext
-    elif name == "FastContextResult":
-        from praisonaiagents.context.fast import FastContextResult
-        return FastContextResult
-    elif name == "FileMatch":
-        from praisonaiagents.context.fast import FileMatch
-        return FileMatch
-    elif name == "LineRange":
-        from praisonaiagents.context.fast import LineRange
-        return LineRange
-    # Agent Skills support (lazy loaded for zero performance impact)
-    elif name == "SkillManager":
-        from praisonaiagents.skills import SkillManager
-        return SkillManager
-    elif name == "SkillProperties":
-        from praisonaiagents.skills import SkillProperties
-        return SkillProperties
-    elif name == "SkillMetadata":
-        from praisonaiagents.skills import SkillMetadata
-        return SkillMetadata
-    elif name == "SkillLoader":
-        from praisonaiagents.skills import SkillLoader
-        return SkillLoader
-    # Memory module (imports chromadb)
-    elif name == "Memory":
-        from praisonaiagents.memory.memory import Memory
-        return Memory
-    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
-
-# Planning mode support
-try:
-    from .planning import (
-        Plan,
-        PlanStep,
-        TodoList,
-        TodoItem,
-        PlanStorage,
-        PlanningAgent,
-        ApprovalCallback,
-        READ_ONLY_TOOLS,
-        RESTRICTED_TOOLS,
-    )
-    _planning_available = True
-except ImportError:
-    _planning_available = False
-    Plan = None
-    PlanStep = None
-    TodoList = None
-    TodoItem = None
-    PlanStorage = None
-    PlanningAgent = None
-    ApprovalCallback = None
-    READ_ONLY_TOOLS = []
-    RESTRICTED_TOOLS = []
+# Handoff - lightweight
 from .agent.handoff import Handoff, handoff, handoff_filters, RECOMMENDED_PROMPT_PREFIX, prompt_with_handoff_instructions
-# Flow display
+
+# Flow display (optional)
 try:
     from .flow_display import FlowDisplay, track_workflow
 except ImportError:
     FlowDisplay = None
     track_workflow = None
+
+# Main display utilities - these are used frequently so import directly
+# but they don't pull in litellm
 from .main import (
     TaskOutput,
     ReflectionOutput,
@@ -154,98 +62,296 @@ from .main import (
     async_display_callbacks,
 )
 
-# Telemetry support (lazy loaded)
-try:
-    from .telemetry import (
-        get_telemetry,
-        enable_telemetry,
-        disable_telemetry,
-        enable_performance_mode,
-        disable_performance_mode,
-        cleanup_telemetry_resources,
-        MinimalTelemetry,
-        TelemetryCollector
-    )
-    _telemetry_available = True
-except ImportError:
-    # Telemetry not available - provide stub functions
-    _telemetry_available = False
-    def get_telemetry():
-        return None
-    
-    def enable_telemetry(*args, **kwargs):
-        import logging
-        logging.warning(
-            "Telemetry not available. Install with: pip install praisonaiagents[telemetry]"
-        )
-        return None
-    
-    def disable_telemetry():
-        pass
-    
-    def enable_performance_mode():
-        pass
-    
-    def disable_performance_mode():
-        pass
-    
-    def cleanup_telemetry_resources():
-        pass
-    
-    MinimalTelemetry = None
-    TelemetryCollector = None
+# Module-level caches for lazy-loaded classes
+_lazy_cache = {}
 
-# Add Agents as an alias for PraisonAIAgents
-Agents = PraisonAIAgents
+# Flags for optional modules
+_mcp_available = None
+_planning_available = None
+_telemetry_available = None
+_agui_available = None
+_a2a_available = None
 
-# Enable PostHog telemetry by default with actual event posting  
-# PostHog events are posted by default unless explicitly disabled
-# Users can:
-#   - Disable completely: PRAISONAI_DISABLE_TELEMETRY=true (or DO_NOT_TRACK=true)
-#   - Enable performance mode: PRAISONAI_PERFORMANCE_MODE=true (minimal overhead, limited events)
-#   - Enable full telemetry: PRAISONAI_FULL_TELEMETRY=true (detailed tracking)
-#   - Legacy opt-in mode: PRAISONAI_AUTO_INSTRUMENT=true
-if _telemetry_available:
+
+def __getattr__(name):
+    """
+    Lazy load heavy modules to avoid impacting package load time.
+    
+    This function is called when an attribute is not found in the module.
+    It enables deferred loading of heavy dependencies like litellm.
+    """
+    global _mcp_available, _planning_available, _telemetry_available
+    global _agui_available, _a2a_available
+    
+    # Return cached values if available
+    if name in _lazy_cache:
+        return _lazy_cache[name]
+    
+    # Agent and related classes (triggers litellm import chain)
+    if name == "Agent":
+        from .agent.agent import Agent
+        _lazy_cache[name] = Agent
+        return Agent
+    elif name == "ImageAgent":
+        from .agent.image_agent import ImageAgent
+        _lazy_cache[name] = ImageAgent
+        return ImageAgent
+    elif name == "ContextAgent":
+        from .agent.context_agent import ContextAgent
+        _lazy_cache[name] = ContextAgent
+        return ContextAgent
+    elif name == "create_context_agent":
+        from .agent.context_agent import create_context_agent
+        _lazy_cache[name] = create_context_agent
+        return create_context_agent
+    elif name == "DeepResearchAgent":
+        from .agent.deep_research_agent import DeepResearchAgent
+        _lazy_cache[name] = DeepResearchAgent
+        return DeepResearchAgent
+    elif name == "DeepResearchResponse":
+        from .agent.deep_research_agent import DeepResearchResponse
+        _lazy_cache[name] = DeepResearchResponse
+        return DeepResearchResponse
+    elif name == "Citation":
+        from .agent.deep_research_agent import Citation
+        _lazy_cache[name] = Citation
+        return Citation
+    elif name == "ReasoningStep":
+        from .agent.deep_research_agent import ReasoningStep
+        _lazy_cache[name] = ReasoningStep
+        return ReasoningStep
+    elif name == "WebSearchCall":
+        from .agent.deep_research_agent import WebSearchCall
+        _lazy_cache[name] = WebSearchCall
+        return WebSearchCall
+    elif name == "CodeExecutionStep":
+        from .agent.deep_research_agent import CodeExecutionStep
+        _lazy_cache[name] = CodeExecutionStep
+        return CodeExecutionStep
+    elif name == "MCPCall":
+        from .agent.deep_research_agent import MCPCall
+        _lazy_cache[name] = MCPCall
+        return MCPCall
+    elif name == "FileSearchCall":
+        from .agent.deep_research_agent import FileSearchCall
+        _lazy_cache[name] = FileSearchCall
+        return FileSearchCall
+    elif name == "Provider":
+        from .agent.deep_research_agent import Provider
+        _lazy_cache[name] = Provider
+        return Provider
+    elif name == "QueryRewriterAgent":
+        from .agent.query_rewriter_agent import QueryRewriterAgent
+        _lazy_cache[name] = QueryRewriterAgent
+        return QueryRewriterAgent
+    elif name == "RewriteStrategy":
+        from .agent.query_rewriter_agent import RewriteStrategy
+        _lazy_cache[name] = RewriteStrategy
+        return RewriteStrategy
+    elif name == "RewriteResult":
+        from .agent.query_rewriter_agent import RewriteResult
+        _lazy_cache[name] = RewriteResult
+        return RewriteResult
+    elif name == "PromptExpanderAgent":
+        from .agent.prompt_expander_agent import PromptExpanderAgent
+        _lazy_cache[name] = PromptExpanderAgent
+        return PromptExpanderAgent
+    elif name == "ExpandStrategy":
+        from .agent.prompt_expander_agent import ExpandStrategy
+        _lazy_cache[name] = ExpandStrategy
+        return ExpandStrategy
+    elif name == "ExpandResult":
+        from .agent.prompt_expander_agent import ExpandResult
+        _lazy_cache[name] = ExpandResult
+        return ExpandResult
+    
+    # PraisonAIAgents and Agents alias
+    elif name == "PraisonAIAgents":
+        from .agents.agents import PraisonAIAgents
+        _lazy_cache[name] = PraisonAIAgents
+        _lazy_cache["Agents"] = PraisonAIAgents  # Also cache alias
+        return PraisonAIAgents
+    elif name == "Agents":
+        from .agents.agents import PraisonAIAgents
+        _lazy_cache["PraisonAIAgents"] = PraisonAIAgents
+        _lazy_cache[name] = PraisonAIAgents
+        return PraisonAIAgents
+    
+    # Task
+    elif name == "Task":
+        from .task.task import Task
+        _lazy_cache[name] = Task
+        return Task
+    
+    # AutoAgents
+    elif name == "AutoAgents":
+        from .agents.autoagents import AutoAgents
+        _lazy_cache[name] = AutoAgents
+        return AutoAgents
+    
+    # Session (imports requests)
+    elif name == "Session":
+        from .session import Session
+        _lazy_cache[name] = Session
+        return Session
+    
+    # MCP support (optional)
+    elif name == "MCP":
+        if _mcp_available is None:
+            try:
+                from .mcp.mcp import MCP as _MCP
+                _lazy_cache[name] = _MCP
+                return _MCP
+            except ImportError:
+                return None
+        return _lazy_cache.get(name)
+    
+    # Knowledge module (imports chromadb, mem0)
+    elif name == "Knowledge":
+        from praisonaiagents.knowledge.knowledge import Knowledge
+        _lazy_cache[name] = Knowledge
+        return Knowledge
+    elif name == "Chunking":
+        from praisonaiagents.knowledge.chunking import Chunking
+        _lazy_cache[name] = Chunking
+        return Chunking
+    
+    # FastContext support
+    elif name == "FastContext":
+        from praisonaiagents.context.fast import FastContext
+        _lazy_cache[name] = FastContext
+        return FastContext
+    elif name == "FastContextResult":
+        from praisonaiagents.context.fast import FastContextResult
+        _lazy_cache[name] = FastContextResult
+        return FastContextResult
+    elif name == "FileMatch":
+        from praisonaiagents.context.fast import FileMatch
+        _lazy_cache[name] = FileMatch
+        return FileMatch
+    elif name == "LineRange":
+        from praisonaiagents.context.fast import LineRange
+        _lazy_cache[name] = LineRange
+        return LineRange
+    
+    # Agent Skills support (lazy loaded for zero performance impact)
+    elif name == "SkillManager":
+        from praisonaiagents.skills import SkillManager
+        _lazy_cache[name] = SkillManager
+        return SkillManager
+    elif name == "SkillProperties":
+        from praisonaiagents.skills import SkillProperties
+        _lazy_cache[name] = SkillProperties
+        return SkillProperties
+    elif name == "SkillMetadata":
+        from praisonaiagents.skills import SkillMetadata
+        _lazy_cache[name] = SkillMetadata
+        return SkillMetadata
+    elif name == "SkillLoader":
+        from praisonaiagents.skills import SkillLoader
+        _lazy_cache[name] = SkillLoader
+        return SkillLoader
+    
+    # Memory module (imports chromadb)
+    elif name == "Memory":
+        from praisonaiagents.memory.memory import Memory
+        _lazy_cache[name] = Memory
+        return Memory
+    
+    # Planning mode support (lazy)
+    elif name in ("Plan", "PlanStep", "TodoList", "TodoItem", "PlanStorage", 
+                  "PlanningAgent", "ApprovalCallback", "READ_ONLY_TOOLS", 
+                  "RESTRICTED_TOOLS"):
+        try:
+            from . import planning as _planning
+            result = getattr(_planning, name)
+            _lazy_cache[name] = result
+            return result
+        except ImportError:
+            if name in ("READ_ONLY_TOOLS", "RESTRICTED_TOOLS"):
+                return []
+            return None
+    
+    # Telemetry support (lazy loaded)
+    elif name in ("get_telemetry", "enable_telemetry", "disable_telemetry",
+                  "enable_performance_mode", "disable_performance_mode",
+                  "cleanup_telemetry_resources", "MinimalTelemetry", "TelemetryCollector"):
+        try:
+            from . import telemetry as _telemetry
+            result = getattr(_telemetry, name)
+            _lazy_cache[name] = result
+            return result
+        except ImportError:
+            # Provide stub functions
+            if name == "get_telemetry":
+                return lambda: None
+            elif name == "enable_telemetry":
+                import logging
+                def _stub(*args, **kwargs):
+                    logging.warning(
+                        "Telemetry not available. Install with: pip install praisonaiagents[telemetry]"
+                    )
+                    return None
+                return _stub
+            elif name in ("disable_telemetry", "enable_performance_mode", 
+                         "disable_performance_mode", "cleanup_telemetry_resources"):
+                return lambda: None
+            return None
+    
+    # AG-UI support (optional, lazy loaded)
+    elif name == "AGUI":
+        try:
+            from praisonaiagents.ui.agui import AGUI
+            _lazy_cache[name] = AGUI
+            return AGUI
+        except ImportError:
+            return None
+    
+    # A2A support (optional, lazy loaded)
+    elif name == "A2A":
+        try:
+            from praisonaiagents.ui.a2a import A2A
+            _lazy_cache[name] = A2A
+            return A2A
+        except ImportError:
+            return None
+    
+    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
+
+
+# Initialize telemetry only if explicitly enabled via config
+def _init_telemetry():
+    """Initialize telemetry if enabled via environment variable."""
+    if not _config.TELEMETRY_ENABLED:
+        return
+    
     try:
-        import os
+        from .telemetry import get_telemetry
+        from .telemetry.integration import auto_instrument_all
         
-        # Check for explicit disable (respects DO_NOT_TRACK and other disable flags)
-        telemetry_disabled = any([
-            os.environ.get('PRAISONAI_TELEMETRY_DISABLED', '').lower() in ('true', '1', 'yes'),
-            os.environ.get('PRAISONAI_DISABLE_TELEMETRY', '').lower() in ('true', '1', 'yes'),
-            os.environ.get('DO_NOT_TRACK', '').lower() in ('true', '1', 'yes'),
-        ])
-        
-        # Check for performance mode (minimal overhead with limited events)
-        performance_mode = os.environ.get('PRAISONAI_PERFORMANCE_MODE', '').lower() in ('true', '1', 'yes')
-        
-        # Check for full telemetry mode (more detailed tracking)
-        full_telemetry = os.environ.get('PRAISONAI_FULL_TELEMETRY', '').lower() in ('true', '1', 'yes')
-        
-        # Legacy explicit auto-instrument option
-        explicit_auto_instrument = os.environ.get('PRAISONAI_AUTO_INSTRUMENT', '').lower() in ('true', '1', 'yes')
-        
-        # Enable PostHog by default unless explicitly disabled
-        if not telemetry_disabled:
-            _telemetry = get_telemetry()
-            if _telemetry and _telemetry.enabled:
-                from .telemetry.integration import auto_instrument_all
-                
-                # Default: PostHog telemetry is enabled and events are posted
-                # Performance mode can be explicitly enabled for minimal overhead
-                use_performance_mode = performance_mode and not (full_telemetry or explicit_auto_instrument)
-                auto_instrument_all(_telemetry, performance_mode=use_performance_mode)
-                
-                # Track package import for basic usage analytics
-                try:
-                    _telemetry.track_feature_usage("package_import")
-                except Exception:
-                    pass
+        _telemetry = get_telemetry()
+        if _telemetry and _telemetry.enabled:
+            use_performance_mode = _config.PERFORMANCE_MODE and not (
+                _config.FULL_TELEMETRY or _config.AUTO_INSTRUMENT
+            )
+            auto_instrument_all(_telemetry, performance_mode=use_performance_mode)
+            
+            # Track package import for basic usage analytics
+            try:
+                _telemetry.track_feature_usage("package_import")
+            except Exception:
+                pass
     except Exception:
         # Silently fail if there are any issues - never break user applications
         pass
 
+
+# Only initialize telemetry if explicitly enabled
+_init_telemetry()
+
+
 __all__ = [
+    # Core classes (lazy loaded)
     'Agent',
     'ImageAgent',
     'ContextAgent',
@@ -314,11 +420,8 @@ __all__ = [
     'RewriteResult',
     'PromptExpanderAgent',
     'ExpandStrategy',
-    'ExpandResult'
-]
-
-# Add workflow exports
-__all__.extend([
+    'ExpandResult',
+    # Workflows
     'Workflow',
     'WorkflowStep', 
     'WorkflowContext',
@@ -331,48 +434,33 @@ __all__.extend([
     'parallel',
     'loop',
     'repeat',
-    'Pipeline'
-])
-
-# Add MCP to __all__ if available
-if _mcp_available:
-    __all__.append('MCP')
-    
-# Add flow display if available
-if FlowDisplay is not None:
-    __all__.extend(['FlowDisplay', 'track_workflow'])
-
-# Add FastContext exports (lazy loaded)
-__all__.extend([
+    'Pipeline',
+    # MCP
+    'MCP',
+    # Flow display
+    'FlowDisplay',
+    'track_workflow',
+    # FastContext
     'FastContext',
     'FastContextResult',
     'FileMatch',
-    'LineRange'
-])
-
-# Add Agent Skills exports (lazy loaded)
-__all__.extend([
+    'LineRange',
+    # Agent Skills
     'SkillManager',
     'SkillProperties',
     'SkillMetadata',
-    'SkillLoader'
-])
-
-# AG-UI support (optional, lazy loaded)
-try:
-    from praisonaiagents.ui.agui import AGUI
-    __all__.append('AGUI')
-    _agui_available = True
-except ImportError:
-    _agui_available = False
-    AGUI = None
-
-# A2A support (optional, lazy loaded)
-try:
-    from praisonaiagents.ui.a2a import A2A
-    __all__.append('A2A')
-    _a2a_available = True
-except ImportError:
-    _a2a_available = False
-    A2A = None
-
+    'SkillLoader',
+    # Planning
+    'Plan',
+    'PlanStep',
+    'TodoList',
+    'TodoItem',
+    'PlanStorage',
+    'PlanningAgent',
+    'ApprovalCallback',
+    'READ_ONLY_TOOLS',
+    'RESTRICTED_TOOLS',
+    # UI
+    'AGUI',
+    'A2A',
+]

src/praisonai-agents/praisonaiagents/_config.py

@@ -0,0 +1,106 @@
+"""
+Configuration module for PraisonAI Agents.
+
+Provides feature flags and configuration options that can be controlled
+via environment variables for performance tuning and behavior control.
+
+Environment Variables:
+    PRAISONAI_LAZY_IMPORTS: Enable lazy imports (default: true)
+    PRAISONAI_TELEMETRY_ENABLED: Enable telemetry (default: false, opt-in)
+    PRAISONAI_TELEMETRY_DISABLED: Disable telemetry (takes precedence)
+    DO_NOT_TRACK: Standard opt-out flag (takes precedence)
+"""
+import os
+from typing import Optional
+
+
+def _str_to_bool(value: Optional[str], default: bool = False) -> bool:
+    """Convert string environment variable to boolean."""
+    if value is None:
+        return default
+    return value.lower() in ('true', '1', 'yes', 'on')
+
+
+# Feature Flags
+# -------------
+
+# Lazy imports: When True (default), heavy dependencies like litellm are
+# only imported when actually needed, reducing import time significantly.
+LAZY_IMPORTS = _str_to_bool(
+    os.environ.get('PRAISONAI_LAZY_IMPORTS'), 
+    default=True
+)
+
+# Telemetry: Opt-in by default (False). Users must explicitly enable.
+# Precedence (highest to lowest):
+#   1. DO_NOT_TRACK=true -> disabled
+#   2. PRAISONAI_TELEMETRY_DISABLED=true -> disabled  
+#   3. PRAISONAI_DISABLE_TELEMETRY=true -> disabled
+#   4. PRAISONAI_TELEMETRY_ENABLED=true -> enabled
+#   5. Default: disabled (opt-in model)
+def _get_telemetry_enabled() -> bool:
+    """Determine if telemetry should be enabled based on env vars."""
+    # Check disable flags first (highest precedence)
+    if _str_to_bool(os.environ.get('DO_NOT_TRACK')):
+        return False
+    if _str_to_bool(os.environ.get('PRAISONAI_TELEMETRY_DISABLED')):
+        return False
+    if _str_to_bool(os.environ.get('PRAISONAI_DISABLE_TELEMETRY')):
+        return False
+    
+    # Check explicit enable flag
+    if _str_to_bool(os.environ.get('PRAISONAI_TELEMETRY_ENABLED')):
+        return True
+    
+    # Default: disabled (opt-in model)
+    return False
+
+
+TELEMETRY_ENABLED = _get_telemetry_enabled()
+
+# Performance mode: Minimal telemetry overhead when enabled
+PERFORMANCE_MODE = _str_to_bool(
+    os.environ.get('PRAISONAI_PERFORMANCE_MODE'),
+    default=False
+)
+
+# Full telemetry mode: Detailed tracking when enabled
+FULL_TELEMETRY = _str_to_bool(
+    os.environ.get('PRAISONAI_FULL_TELEMETRY'),
+    default=False
+)
+
+# Legacy auto-instrument flag
+AUTO_INSTRUMENT = _str_to_bool(
+    os.environ.get('PRAISONAI_AUTO_INSTRUMENT'),
+    default=False
+)
+
+
+def missing_dependency_error(
+    package_name: str,
+    extra_name: Optional[str] = None,
+    feature_name: Optional[str] = None
+) -> ImportError:
+    """
+    Create a standardized ImportError for missing optional dependencies.
+    
+    Args:
+        package_name: The name of the missing package (e.g., 'litellm')
+        extra_name: The pip extra to install (e.g., 'llm' for [llm])
+        feature_name: Human-readable feature name (e.g., 'LLM support')
+    
+    Returns:
+        ImportError with helpful install instructions
+    """
+    if extra_name:
+        install_cmd = f"pip install 'praisonaiagents[{extra_name}]'"
+    else:
+        install_cmd = f"pip install {package_name}"
+    
+    feature_desc = f" for {feature_name}" if feature_name else ""
+    
+    return ImportError(
+        f"{package_name} is required{feature_desc} but not installed. "
+        f"Please install with: {install_cmd}"
+    )

src/praisonai-agents/praisonaiagents/_logging.py

@@ -84,8 +84,18 @@ def _configure_loggers():
 # ========================================================================
 # LITELLM CONFIGURATION
 # ========================================================================
-def _configure_litellm():
-    """Configure litellm after it's imported."""
+_litellm_configured = False
+
+def configure_litellm():
+    """Configure litellm after it's imported.
+    
+    This function should be called lazily when litellm is actually used,
+    not at module import time, to avoid importing litellm unnecessarily.
+    """
+    global _litellm_configured
+    if _litellm_configured:
+        return
+    
     try:
         import litellm
         litellm.telemetry = False
@@ -97,11 +107,17 @@ def _configure_litellm():
         
         if hasattr(litellm, 'set_verbose'):
             litellm.set_verbose = False
+        
+        _litellm_configured = True
             
     except (ImportError, AttributeError):
         pass
 
 
+# Alias for backward compatibility
+_configure_litellm = configure_litellm
+
+
 # ========================================================================
 # ROOT LOGGER CONFIGURATION
 # ========================================================================
@@ -124,11 +140,16 @@ def configure_root_logger():
 # INITIALIZATION
 # ========================================================================
 def initialize_logging():
-    """Initialize all logging configuration."""
+    """Initialize all logging configuration.
+    
+    Note: litellm configuration is NOT done here to avoid importing litellm
+    at package import time. Call configure_litellm() when litellm is needed.
+    """
     _configure_environment()
     _configure_loggers()
-    _configure_litellm()
+    # NOTE: _configure_litellm() is NOT called here to avoid importing litellm
+    # It will be called lazily when LLM class is instantiated
 
 
-# Auto-initialize on import
+# Auto-initialize on import (but NOT litellm)
 initialize_logging()

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

@@ -4,6 +4,7 @@ import json
 import copy
 import logging
 import asyncio
+import threading
 from typing import List, Optional, Any, Dict, Union, Literal, TYPE_CHECKING, Callable, Tuple, Generator
 from rich.console import Console
 from rich.live import Live
@@ -580,9 +581,9 @@ class Agent:
         self.embedder_config = embedder_config
         self.knowledge = knowledge
         self.use_system_prompt = use_system_prompt
-        # NOTE: chat_history is not thread-safe. If concurrent access is needed,
-        # consider using threading.Lock or other synchronization mechanisms
+        # Thread-safe chat_history with lock for concurrent access
         self.chat_history = []
+        self._history_lock = threading.Lock()
         self.markdown = markdown
         self.stream = stream
         self.metrics = metrics
@@ -643,11 +644,11 @@ Your Goal: {self.goal}
         self._guardrail_fn = None
         self._setup_guardrail()
         
-        # Cache for system prompts and formatted tools
-        # Note: In single-threaded usage (common case), these are safe
-        # For multi-threaded usage, consider using threading.Lock
+        # Cache for system prompts and formatted tools with thread-safe lock
+        # RLock allows re-entrant access from the same thread
         self._system_prompt_cache = {}
         self._formatted_tools_cache = {}
+        self._cache_lock = threading.RLock()
         # Limit cache size to prevent unbounded growth
         self._max_cache_size = 100
 

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

@@ -5,13 +5,26 @@ This module uses only stdlib (typing, dataclasses) to define the interface.
 No external dependencies are imported here.
 
 Implementations are provided by the wrapper layer (praisonai.db).
+
+Schema Versioning:
+- SCHEMA_VERSION tracks the protocol version
+- Adapters should implement get_schema_version() to report their version
+- Adapters should implement migrate_schema() for upgrades
+- Version format: MAJOR.MINOR (e.g., "1.0", "1.1", "2.0")
+- MAJOR changes are breaking, MINOR changes are backward compatible
 """
 
 from dataclasses import dataclass, field
-from typing import Any, Dict, List, Optional, Protocol, runtime_checkable
+from typing import Any, Dict, List, Optional, Protocol, runtime_checkable, Tuple
 import time
 
 
+# Current schema version for the DbAdapter protocol
+# Increment MINOR for backward-compatible additions
+# Increment MAJOR for breaking changes
+SCHEMA_VERSION = "1.0"
+
+
 @dataclass
 class DbToolCall:
     """A tool call to be persisted."""
@@ -365,6 +378,48 @@ class DbAdapter(Protocol):
     def close(self) -> None:
         """Close the database connection."""
         ...
+    
+    # --- Schema Versioning (Optional) ---
+    # These methods are optional for backward compatibility.
+    # Adapters that don't implement them will work but won't support migrations.
+    
+    def get_schema_version(self) -> str:
+        """
+        Get the current schema version of this adapter.
+        
+        Returns:
+            Schema version string (e.g., "1.0")
+            Default implementation returns "1.0"
+        """
+        ...
+    
+    def migrate_schema(
+        self,
+        from_version: str,
+        to_version: str,
+    ) -> bool:
+        """
+        Migrate the schema from one version to another.
+        
+        Args:
+            from_version: Current schema version
+            to_version: Target schema version
+            
+        Returns:
+            True if migration succeeded, False otherwise
+        """
+        ...
+    
+    def check_schema_compatibility(self) -> Tuple[bool, str]:
+        """
+        Check if the adapter's schema is compatible with the protocol.
+        
+        Returns:
+            Tuple of (is_compatible, message)
+            - is_compatible: True if schema is compatible
+            - message: Description of compatibility status or required action
+        """
+        ...
 
 
 @runtime_checkable

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

@@ -0,0 +1,327 @@
+"""
+PraisonAI Agents Lite - Minimal agent framework without heavy dependencies.
+
+This subpackage provides a lightweight version of PraisonAI Agents that:
+- Does NOT import litellm, chromadb, or other heavy dependencies at import time
+- Provides core Agent/Task/Tools abstractions
+- Allows users to bring their own LLM client (BYO-LLM)
+- Has minimal import time (<50ms) and memory footprint (<10MB)
+
+Usage:
+    from praisonaiagents.lite import LiteAgent, LiteTask, Tools
+    
+    # Create a lite agent with your own LLM function
+    def my_llm(messages):
+        # Your LLM implementation here
+        return "response"
+    
+    agent = LiteAgent(
+        name="Assistant",
+        llm_fn=my_llm,
+        tools=[my_tool]
+    )
+    
+    response = agent.chat("Hello!")
+
+For full functionality with litellm integration, use the main package:
+    from praisonaiagents import Agent
+"""
+
+from typing import Any, Callable, Dict, List, Optional
+from dataclasses import dataclass, field
+import threading
+
+
+@dataclass
+class LiteToolResult:
+    """Result from a tool execution."""
+    output: Any
+    success: bool = True
+    error: Optional[str] = None
+
+
+@dataclass
+class LiteTask:
+    """
+    Lightweight task definition without heavy dependencies.
+    
+    Attributes:
+        description: What the task should accomplish
+        expected_output: Description of expected output format
+        agent: Optional agent assigned to this task
+        context: Additional context for the task
+        result: Result after task execution
+    """
+    description: str
+    expected_output: str = ""
+    agent: Optional['LiteAgent'] = None
+    context: Dict[str, Any] = field(default_factory=dict)
+    result: Optional[str] = None
+    
+    def execute(self, input_text: Optional[str] = None) -> str:
+        """Execute the task using the assigned agent."""
+        if self.agent is None:
+            raise ValueError("No agent assigned to this task")
+        
+        prompt = input_text or self.description
+        if self.expected_output:
+            prompt += f"\n\nExpected output format: {self.expected_output}"
+        
+        self.result = self.agent.chat(prompt)
+        return self.result
+
+
+class LiteAgent:
+    """
+    Lightweight agent that works with any LLM function.
+    
+    This agent does NOT import litellm or other heavy dependencies.
+    Users provide their own LLM function for chat completions.
+    
+    Args:
+        name: Agent name
+        llm_fn: Function that takes messages list and returns response string.
+                Signature: (messages: List[Dict]) -> str
+        instructions: System instructions for the agent
+        tools: List of tool functions the agent can use
+        verbose: Enable verbose output
+    
+    Example:
+        def my_openai_llm(messages):
+            import openai
+            client = openai.OpenAI()
+            response = client.chat.completions.create(
+                model="gpt-4o-mini",
+                messages=messages
+            )
+            return response.choices[0].message.content
+        
+        agent = LiteAgent(
+            name="Assistant",
+            llm_fn=my_openai_llm,
+            instructions="You are a helpful assistant."
+        )
+        
+        response = agent.chat("Hello!")
+    """
+    
+    def __init__(
+        self,
+        name: str = "LiteAgent",
+        llm_fn: Optional[Callable[[List[Dict]], str]] = None,
+        instructions: str = "You are a helpful assistant.",
+        tools: Optional[List[Callable]] = None,
+        verbose: bool = False,
+    ):
+        self.name = name
+        self.llm_fn = llm_fn
+        self.instructions = instructions
+        self.tools = tools or []
+        self.verbose = verbose
+        
+        # Thread-safe chat history
+        self.chat_history: List[Dict[str, str]] = []
+        self._history_lock = threading.Lock()
+        
+        # Tool registry
+        self._tool_map: Dict[str, Callable] = {}
+        for tool_fn in self.tools:
+            tool_name = getattr(tool_fn, '__name__', str(tool_fn))
+            self._tool_map[tool_name] = tool_fn
+    
+    def _build_messages(self, user_message: str) -> List[Dict[str, str]]:
+        """Build messages list for LLM call."""
+        messages = [{"role": "system", "content": self.instructions}]
+        
+        with self._history_lock:
+            messages.extend(self.chat_history.copy())
+        
+        messages.append({"role": "user", "content": user_message})
+        return messages
+    
+    def _add_to_history(self, role: str, content: str):
+        """Thread-safe addition to chat history."""
+        with self._history_lock:
+            self.chat_history.append({"role": role, "content": content})
+    
+    def chat(self, message: str) -> str:
+        """
+        Send a message to the agent and get a response.
+        
+        Args:
+            message: User message
+            
+        Returns:
+            Agent's response string
+        """
+        if self.llm_fn is None:
+            raise ValueError(
+                "No LLM function provided. Please provide an llm_fn when creating the agent.\n"
+                "Example:\n"
+                "  def my_llm(messages):\n"
+                "      # Your LLM implementation\n"
+                "      return 'response'\n"
+                "  agent = LiteAgent(llm_fn=my_llm)"
+            )
+        
+        messages = self._build_messages(message)
+        
+        if self.verbose:
+            print(f"[{self.name}] Processing: {message[:50]}...")
+        
+        response = self.llm_fn(messages)
+        
+        # Add to history
+        self._add_to_history("user", message)
+        self._add_to_history("assistant", response)
+        
+        if self.verbose:
+            print(f"[{self.name}] Response: {response[:100]}...")
+        
+        return response
+    
+    def clear_history(self):
+        """Clear chat history."""
+        with self._history_lock:
+            self.chat_history.clear()
+    
+    def execute_tool(self, tool_name: str, **kwargs) -> LiteToolResult:
+        """
+        Execute a tool by name.
+        
+        Args:
+            tool_name: Name of the tool to execute
+            **kwargs: Arguments to pass to the tool
+            
+        Returns:
+            LiteToolResult with output or error
+        """
+        if tool_name not in self._tool_map:
+            return LiteToolResult(
+                output=None,
+                success=False,
+                error=f"Tool '{tool_name}' not found"
+            )
+        
+        try:
+            tool_fn = self._tool_map[tool_name]
+            result = tool_fn(**kwargs)
+            return LiteToolResult(output=result, success=True)
+        except Exception as e:
+            return LiteToolResult(
+                output=None,
+                success=False,
+                error=str(e)
+            )
+
+
+def create_openai_llm_fn(
+    model: str = "gpt-4o-mini",
+    api_key: Optional[str] = None,
+    **kwargs
+) -> Callable[[List[Dict]], str]:
+    """
+    Create an LLM function using the OpenAI SDK.
+    
+    This is a convenience function for users who want to use OpenAI
+    without the full litellm dependency.
+    
+    Args:
+        model: OpenAI model name
+        api_key: OpenAI API key (or use OPENAI_API_KEY env var)
+        **kwargs: Additional arguments to pass to OpenAI
+        
+    Returns:
+        LLM function compatible with LiteAgent
+        
+    Example:
+        llm_fn = create_openai_llm_fn(model="gpt-4o-mini")
+        agent = LiteAgent(llm_fn=llm_fn)
+    """
+    def llm_fn(messages: List[Dict]) -> str:
+        try:
+            import openai
+        except ImportError:
+            raise ImportError(
+                "openai package is required for create_openai_llm_fn. "
+                "Install with: pip install openai"
+            )
+        
+        client = openai.OpenAI(api_key=api_key)
+        response = client.chat.completions.create(
+            model=model,
+            messages=messages,
+            **kwargs
+        )
+        return response.choices[0].message.content or ""
+    
+    return llm_fn
+
+
+def create_anthropic_llm_fn(
+    model: str = "claude-3-5-sonnet-20241022",
+    api_key: Optional[str] = None,
+    max_tokens: int = 4096,
+    **kwargs
+) -> Callable[[List[Dict]], str]:
+    """
+    Create an LLM function using the Anthropic SDK.
+    
+    Args:
+        model: Anthropic model name
+        api_key: Anthropic API key (or use ANTHROPIC_API_KEY env var)
+        max_tokens: Maximum tokens in response
+        **kwargs: Additional arguments to pass to Anthropic
+        
+    Returns:
+        LLM function compatible with LiteAgent
+    """
+    def llm_fn(messages: List[Dict]) -> str:
+        try:
+            import anthropic
+        except ImportError:
+            raise ImportError(
+                "anthropic package is required for create_anthropic_llm_fn. "
+                "Install with: pip install anthropic"
+            )
+        
+        client = anthropic.Anthropic(api_key=api_key)
+        
+        # Extract system message
+        system = ""
+        chat_messages = []
+        for msg in messages:
+            if msg["role"] == "system":
+                system = msg["content"]
+            else:
+                chat_messages.append(msg)
+        
+        response = client.messages.create(
+            model=model,
+            max_tokens=max_tokens,
+            system=system,
+            messages=chat_messages,
+            **kwargs
+        )
+        return response.content[0].text
+    
+    return llm_fn
+
+
+# Re-export Tools from main package (lightweight, no heavy deps)
+from ..tools.tools import Tools
+from ..tools.base import BaseTool, ToolResult
+from ..tools.decorator import tool
+
+
+__all__ = [
+    "LiteAgent",
+    "LiteTask",
+    "LiteToolResult",
+    "Tools",
+    "BaseTool",
+    "ToolResult",
+    "tool",
+    "create_openai_llm_fn",
+    "create_anthropic_llm_fn",
+]

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

@@ -1,33 +1,102 @@
+"""
+LLM Module for PraisonAI Agents.
+
+This module provides language model integrations with lazy loading
+to minimize import time when LLM functionality is not immediately needed.
+"""
 import os
 
-# Ensure litellm telemetry is disabled before imports
+# Ensure litellm telemetry is disabled before any imports
 os.environ["LITELLM_TELEMETRY"] = "False"
 
-# Import modules
-from .llm import LLM, LLMContextLengthExceededException
-from .openai_client import (
-    OpenAIClient, 
-    get_openai_client,
-    ChatCompletionMessage,
-    Choice,
-    CompletionTokensDetails,
-    PromptTokensDetails,
-    CompletionUsage,
-    ChatCompletion,
-    ToolCall,
-    process_stream_chunks
-)
-from .model_capabilities import (
-    supports_structured_outputs,
-    supports_streaming_with_tools
-)
-from .model_router import (
-    ModelRouter,
-    ModelProfile, 
-    TaskComplexity,
-    create_routing_agent
-)
-from .rate_limiter import RateLimiter
+# Module-level cache for lazy-loaded classes
+_lazy_cache = {}
+
+
+def __getattr__(name):
+    """Lazy load LLM classes to avoid importing litellm at module load time."""
+    if name in _lazy_cache:
+        return _lazy_cache[name]
+    
+    if name == "LLM":
+        from .llm import LLM
+        _lazy_cache[name] = LLM
+        return LLM
+    elif name == "LLMContextLengthExceededException":
+        from .llm import LLMContextLengthExceededException
+        _lazy_cache[name] = LLMContextLengthExceededException
+        return LLMContextLengthExceededException
+    elif name == "OpenAIClient":
+        from .openai_client import OpenAIClient
+        _lazy_cache[name] = OpenAIClient
+        return OpenAIClient
+    elif name == "get_openai_client":
+        from .openai_client import get_openai_client
+        _lazy_cache[name] = get_openai_client
+        return get_openai_client
+    elif name == "ChatCompletionMessage":
+        from .openai_client import ChatCompletionMessage
+        _lazy_cache[name] = ChatCompletionMessage
+        return ChatCompletionMessage
+    elif name == "Choice":
+        from .openai_client import Choice
+        _lazy_cache[name] = Choice
+        return Choice
+    elif name == "CompletionTokensDetails":
+        from .openai_client import CompletionTokensDetails
+        _lazy_cache[name] = CompletionTokensDetails
+        return CompletionTokensDetails
+    elif name == "PromptTokensDetails":
+        from .openai_client import PromptTokensDetails
+        _lazy_cache[name] = PromptTokensDetails
+        return PromptTokensDetails
+    elif name == "CompletionUsage":
+        from .openai_client import CompletionUsage
+        _lazy_cache[name] = CompletionUsage
+        return CompletionUsage
+    elif name == "ChatCompletion":
+        from .openai_client import ChatCompletion
+        _lazy_cache[name] = ChatCompletion
+        return ChatCompletion
+    elif name == "ToolCall":
+        from .openai_client import ToolCall
+        _lazy_cache[name] = ToolCall
+        return ToolCall
+    elif name == "process_stream_chunks":
+        from .openai_client import process_stream_chunks
+        _lazy_cache[name] = process_stream_chunks
+        return process_stream_chunks
+    elif name == "supports_structured_outputs":
+        from .model_capabilities import supports_structured_outputs
+        _lazy_cache[name] = supports_structured_outputs
+        return supports_structured_outputs
+    elif name == "supports_streaming_with_tools":
+        from .model_capabilities import supports_streaming_with_tools
+        _lazy_cache[name] = supports_streaming_with_tools
+        return supports_streaming_with_tools
+    elif name == "ModelRouter":
+        from .model_router import ModelRouter
+        _lazy_cache[name] = ModelRouter
+        return ModelRouter
+    elif name == "ModelProfile":
+        from .model_router import ModelProfile
+        _lazy_cache[name] = ModelProfile
+        return ModelProfile
+    elif name == "TaskComplexity":
+        from .model_router import TaskComplexity
+        _lazy_cache[name] = TaskComplexity
+        return TaskComplexity
+    elif name == "create_routing_agent":
+        from .model_router import create_routing_agent
+        _lazy_cache[name] = create_routing_agent
+        return create_routing_agent
+    elif name == "RateLimiter":
+        from .rate_limiter import RateLimiter
+        _lazy_cache[name] = RateLimiter
+        return RateLimiter
+    
+    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
+
 
 __all__ = [
     "LLM", 
@@ -49,4 +118,4 @@ __all__ = [
     "TaskComplexity",
     "create_routing_agent",
     "RateLimiter"
-]
+]

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

@@ -128,6 +128,10 @@ Respond with ONLY a valid JSON tool call in this format:
     @classmethod
     def _configure_logging(cls):
         """Configure logging settings once for all LLM instances."""
+        # First, configure litellm via the centralized logging module
+        from .._logging import configure_litellm
+        configure_litellm()
+        
         try:
             import litellm
             # Disable telemetry

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

@@ -580,7 +580,60 @@ class MCP:
         
         return openai_tools
     
+    def __enter__(self):
+        """Context manager entry - return self for use in 'with' statements."""
+        return self
+    
+    def __exit__(self, exc_type, exc_val, exc_tb):
+        """Context manager exit - clean up resources."""
+        self.shutdown()
+        return False  # Don't suppress exceptions
+    
+    def shutdown(self):
+        """Explicitly shut down MCP resources.
+        
+        Call this method when done using the MCP instance to ensure
+        all background threads and connections are properly cleaned up.
+        """
+        # Shutdown stdio runner if present
+        if hasattr(self, 'runner') and self.runner is not None:
+            try:
+                self.runner.shutdown()
+            except Exception:
+                pass  # Best effort cleanup
+        
+        # Shutdown SSE client if present
+        if hasattr(self, 'sse_client') and self.sse_client is not None:
+            try:
+                if hasattr(self.sse_client, 'shutdown'):
+                    self.sse_client.shutdown()
+            except Exception:
+                pass
+        
+        # Shutdown HTTP stream client if present
+        if hasattr(self, 'http_stream_client') and self.http_stream_client is not None:
+            try:
+                if hasattr(self.http_stream_client, 'shutdown'):
+                    self.http_stream_client.shutdown()
+            except Exception:
+                pass
+        
+        # Shutdown WebSocket client if present
+        if hasattr(self, 'websocket_client') and self.websocket_client is not None:
+            try:
+                if hasattr(self.websocket_client, 'shutdown'):
+                    self.websocket_client.shutdown()
+            except Exception:
+                pass
+    
     def __del__(self):
-        """Clean up resources when the object is garbage collected."""
-        if hasattr(self, 'runner'):
-            self.runner.shutdown() 
+        """Clean up resources when the object is garbage collected.
+        
+        Note: __del__ is called during garbage collection and may not
+        always run. For reliable cleanup, use the context manager
+        pattern or call shutdown() explicitly.
+        """
+        try:
+            self.shutdown()
+        except Exception:
+            pass  # Best effort cleanup in __del__

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

@@ -27,13 +27,12 @@ Usage:
     )
     
     result = agents.start()
+
+This module uses lazy loading to minimize import time.
 """
 
-from .plan import Plan, PlanStep
-from .todo import TodoList, TodoItem
-from .storage import PlanStorage
-from .planner import PlanningAgent
-from .approval import ApprovalCallback
+# Module-level cache for lazy-loaded classes
+_lazy_cache = {}
 
 # Read-only tools that are allowed in plan mode
 READ_ONLY_TOOLS = [
@@ -90,6 +89,44 @@ RESEARCH_TOOLS = [
     "find_files",
 ]
 
+
+def __getattr__(name):
+    """Lazy load planning classes to avoid importing heavy dependencies at module load time."""
+    if name in _lazy_cache:
+        return _lazy_cache[name]
+    
+    if name == "Plan":
+        from .plan import Plan
+        _lazy_cache[name] = Plan
+        return Plan
+    elif name == "PlanStep":
+        from .plan import PlanStep
+        _lazy_cache[name] = PlanStep
+        return PlanStep
+    elif name == "TodoList":
+        from .todo import TodoList
+        _lazy_cache[name] = TodoList
+        return TodoList
+    elif name == "TodoItem":
+        from .todo import TodoItem
+        _lazy_cache[name] = TodoItem
+        return TodoItem
+    elif name == "PlanStorage":
+        from .storage import PlanStorage
+        _lazy_cache[name] = PlanStorage
+        return PlanStorage
+    elif name == "PlanningAgent":
+        from .planner import PlanningAgent
+        _lazy_cache[name] = PlanningAgent
+        return PlanningAgent
+    elif name == "ApprovalCallback":
+        from .approval import ApprovalCallback
+        _lazy_cache[name] = ApprovalCallback
+        return ApprovalCallback
+    
+    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
+
+
 __all__ = [
     "Plan",
     "PlanStep",

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

@@ -4,6 +4,7 @@ This module provides centralized tool management with support for:
 - Manual registration
 - Auto-discovery via entry_points
 - Tool lookup by name
+- Thread-safe operations for multi-agent scenarios
 
 Usage:
     from praisonaiagents.tools.registry import get_registry
@@ -14,6 +15,7 @@ Usage:
 """
 
 import logging
+import threading
 from typing import Callable, Dict, List, Optional, Union
 from importlib.metadata import entry_points
 
@@ -38,6 +40,7 @@ class ToolRegistry:
         self._tools: Dict[str, BaseTool] = {}
         self._functions: Dict[str, Callable] = {}  # For backward compat with plain functions
         self._discovered: bool = False
+        self._lock = threading.RLock()  # Thread-safe operations for multi-agent scenarios
     
     def register(
         self,
@@ -47,6 +50,8 @@ class ToolRegistry:
     ) -> None:
         """Register a tool with the registry.
         
+        Thread-safe: Uses lock for concurrent access in multi-agent scenarios.
+        
         Args:
             tool: BaseTool instance or callable function
             name: Override name (default: tool.name or function.__name__)
@@ -55,83 +60,93 @@ class ToolRegistry:
         Raises:
             ValueError: If tool with same name exists and overwrite=False
         """
-        # Handle BaseTool instances
-        if isinstance(tool, BaseTool):
-            tool_name = name or tool.name
-            if tool_name in self._tools and not overwrite:
-                logging.debug(f"Tool '{tool_name}' already registered, skipping")
+        with self._lock:
+            # Handle BaseTool instances
+            if isinstance(tool, BaseTool):
+                tool_name = name or tool.name
+                if tool_name in self._tools and not overwrite:
+                    logging.debug(f"Tool '{tool_name}' already registered, skipping")
+                    return
+                self._tools[tool_name] = tool
+                logging.debug(f"Registered tool: {tool_name}")
                 return
-            self._tools[tool_name] = tool
-            logging.debug(f"Registered tool: {tool_name}")
-            return
-        
-        # Handle plain callables
-        if callable(tool):
-            tool_name = name or getattr(tool, '__name__', str(id(tool)))
-            if tool_name in self._functions and not overwrite:
-                logging.debug(f"Function '{tool_name}' already registered, skipping")
+            
+            # Handle plain callables
+            if callable(tool):
+                tool_name = name or getattr(tool, '__name__', str(id(tool)))
+                if tool_name in self._functions and not overwrite:
+                    logging.debug(f"Function '{tool_name}' already registered, skipping")
+                    return
+                self._functions[tool_name] = tool
+                logging.debug(f"Registered function: {tool_name}")
                 return
-            self._functions[tool_name] = tool
-            logging.debug(f"Registered function: {tool_name}")
-            return
-        
-        raise TypeError(f"Cannot register {type(tool)}, expected BaseTool or callable")
+            
+            raise TypeError(f"Cannot register {type(tool)}, expected BaseTool or callable")
     
     def unregister(self, name: str) -> bool:
         """Remove a tool from the registry.
         
+        Thread-safe: Uses lock for concurrent access.
+        
         Args:
             name: Tool name to remove
             
         Returns:
             True if tool was removed, False if not found
         """
-        if name in self._tools:
-            del self._tools[name]
-            return True
-        if name in self._functions:
-            del self._functions[name]
-            return True
-        return False
+        with self._lock:
+            if name in self._tools:
+                del self._tools[name]
+                return True
+            if name in self._functions:
+                del self._functions[name]
+                return True
+            return False
     
     def get(self, name: str) -> Optional[Union[BaseTool, Callable]]:
         """Get a tool by name.
         
+        Thread-safe: Uses lock for concurrent access.
+        
         Args:
             name: Tool name
             
         Returns:
             BaseTool instance, callable, or None if not found
         """
-        # Check BaseTool registry first
-        if name in self._tools:
-            return self._tools[name]
-        
-        # Check functions registry
-        if name in self._functions:
-            return self._functions[name]
-        
-        # Try auto-discovery if not found
-        if not self._discovered:
-            self.discover_plugins()
+        with self._lock:
+            # Check BaseTool registry first
             if name in self._tools:
                 return self._tools[name]
-        
-        return None
+            
+            # Check functions registry
+            if name in self._functions:
+                return self._functions[name]
+            
+            # Try auto-discovery if not found
+            if not self._discovered:
+                self.discover_plugins()
+                if name in self._tools:
+                    return self._tools[name]
+            
+            return None
     
     def list_tools(self) -> List[str]:
-        """List all registered tool names."""
-        return list(self._tools.keys()) + list(self._functions.keys())
+        """List all registered tool names. Thread-safe."""
+        with self._lock:
+            return list(self._tools.keys()) + list(self._functions.keys())
     
     def list_base_tools(self) -> List[BaseTool]:
-        """List all registered BaseTool instances."""
-        return list(self._tools.values())
+        """List all registered BaseTool instances. Thread-safe."""
+        with self._lock:
+            return list(self._tools.values())
     
     def get_all(self) -> Dict[str, Union[BaseTool, Callable]]:
-        """Get all registered tools as a dict."""
-        result = dict(self._tools)
-        result.update(self._functions)
-        return result
+        """Get all registered tools as a dict. Thread-safe."""
+        with self._lock:
+            result = dict(self._tools)
+            result.update(self._functions)
+            return result
     
     def discover_plugins(self) -> int:
         """Discover and register tools from entry_points.
@@ -185,16 +200,19 @@ class ToolRegistry:
         return count
     
     def clear(self) -> None:
-        """Clear all registered tools."""
-        self._tools.clear()
-        self._functions.clear()
-        self._discovered = False
+        """Clear all registered tools. Thread-safe."""
+        with self._lock:
+            self._tools.clear()
+            self._functions.clear()
+            self._discovered = False
     
     def __contains__(self, name: str) -> bool:
-        return name in self._tools or name in self._functions
+        with self._lock:
+            return name in self._tools or name in self._functions
     
     def __len__(self) -> int:
-        return len(self._tools) + len(self._functions)
+        with self._lock:
+            return len(self._tools) + len(self._functions)
     
     def __repr__(self) -> str:
         return f"ToolRegistry(tools={len(self._tools)}, functions={len(self._functions)})"

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

@@ -6,8 +6,35 @@ via various frontend protocols.
 
 Available integrations:
 - agui: AG-UI (Agent-User Interface) protocol for CopilotKit and compatible frontends
+- a2a: Agent-to-Agent protocol
+- a2ui: Agent-to-UI protocol
+
+This module uses lazy loading to minimize import time.
 """
 
-from praisonaiagents.ui.agui import AGUI
+# Module-level cache for lazy-loaded classes
+_lazy_cache = {}
 
-__all__ = ["AGUI"]
+
+def __getattr__(name):
+    """Lazy load UI classes to avoid importing heavy dependencies at module load time."""
+    if name in _lazy_cache:
+        return _lazy_cache[name]
+    
+    if name == "AGUI":
+        from praisonaiagents.ui.agui import AGUI
+        _lazy_cache[name] = AGUI
+        return AGUI
+    elif name == "A2A":
+        from praisonaiagents.ui.a2a import A2A
+        _lazy_cache[name] = A2A
+        return A2A
+    elif name == "A2UI":
+        from praisonaiagents.ui.a2ui import A2UI
+        _lazy_cache[name] = A2UI
+        return A2UI
+    
+    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
+
+
+__all__ = ["AGUI", "A2A", "A2UI"]

src/praisonai-agents/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "praisonaiagents"
-version = "0.4.1"
+version = "0.5.0"
 description = "Praison AI agents for completing complex tasks with Self Reflection Agents"
 readme = "README.md"
 requires-python = ">=3.10"

src/praisonai-agents/tests/integration/test_real_api.py

@@ -0,0 +1,243 @@
+"""
+Real API key integration tests for praisonaiagents.
+
+These tests are gated by the RUN_REAL_KEY_TESTS environment variable.
+They require valid API keys to be set in the environment.
+
+Usage:
+    RUN_REAL_KEY_TESTS=1 pytest tests/integration/test_real_api.py -v
+
+Required environment variables:
+    - OPENAI_API_KEY: OpenAI API key
+    
+Optional:
+    - ANTHROPIC_API_KEY: Anthropic API key (for Claude tests)
+    - GOOGLE_API_KEY: Google API key (for Gemini tests)
+"""
+import os
+import sys
+import pytest
+
+# Gate all tests in this module
+pytestmark = pytest.mark.skipif(
+    os.environ.get('RUN_REAL_KEY_TESTS', '').lower() not in ('1', 'true', 'yes'),
+    reason="Real API key tests disabled. Set RUN_REAL_KEY_TESTS=1 to enable."
+)
+
+
+def clear_modules():
+    """Clear all praisonai and litellm related modules from cache."""
+    to_remove = [m for m in list(sys.modules.keys()) 
+                 if 'praison' in m or 'litellm' in m]
+    for mod in to_remove:
+        del sys.modules[mod]
+
+
+class TestAgentRealAPI:
+    """Test Agent with real API calls."""
+    
+    @pytest.fixture(autouse=True)
+    def setup(self):
+        """Set up test fixtures."""
+        clear_modules()
+        
+        # Verify API key is set (without printing it)
+        api_key = os.environ.get('OPENAI_API_KEY')
+        if not api_key:
+            pytest.skip("OPENAI_API_KEY not set")
+        
+        yield
+        clear_modules()
+    
+    def test_agent_simple_chat(self):
+        """Test basic Agent.chat() with real API."""
+        from praisonaiagents import Agent
+        
+        agent = Agent(
+            name="TestAgent",
+            instructions="You are a helpful assistant. Keep responses very brief.",
+            llm="gpt-4o-mini",
+            verbose=False
+        )
+        
+        response = agent.chat("Say 'hello' and nothing else.")
+        
+        assert response is not None
+        assert len(response) > 0
+        assert 'hello' in response.lower()
+        
+        print(f"[PASS] Agent response received (length: {len(response)})")
+    
+    def test_agent_with_tool(self):
+        """Test Agent with a simple tool."""
+        from praisonaiagents import Agent
+        
+        # Define a simple deterministic tool
+        def add_numbers(a: int, b: int) -> int:
+            """Add two numbers together."""
+            return a + b
+        
+        agent = Agent(
+            name="MathAgent",
+            instructions="You are a math assistant. Use the add_numbers tool when asked to add.",
+            llm="gpt-4o-mini",
+            tools=[add_numbers],
+            verbose=False
+        )
+        
+        response = agent.chat("What is 5 + 3? Use the add_numbers tool.")
+        
+        assert response is not None
+        assert '8' in response
+        
+        print(f"[PASS] Tool call successful, response contains '8'")
+    
+    def test_agent_chat_history(self):
+        """Test that chat history is maintained."""
+        from praisonaiagents import Agent
+        
+        agent = Agent(
+            name="HistoryAgent",
+            instructions="You are a helpful assistant. Remember what the user tells you.",
+            llm="gpt-4o-mini",
+            verbose=False
+        )
+        
+        # First message
+        agent.chat("My favorite color is blue.")
+        
+        # Second message referencing first
+        response = agent.chat("What is my favorite color?")
+        
+        assert response is not None
+        assert 'blue' in response.lower()
+        
+        print(f"[PASS] Chat history maintained correctly")
+
+
+class TestLiteAgentRealAPI:
+    """Test LiteAgent with real API using OpenAI SDK directly."""
+    
+    @pytest.fixture(autouse=True)
+    def setup(self):
+        """Set up test fixtures."""
+        clear_modules()
+        
+        api_key = os.environ.get('OPENAI_API_KEY')
+        if not api_key:
+            pytest.skip("OPENAI_API_KEY not set")
+        
+        yield
+        clear_modules()
+    
+    def test_lite_agent_with_openai(self):
+        """Test LiteAgent with OpenAI SDK."""
+        from praisonaiagents.lite import LiteAgent, create_openai_llm_fn
+        
+        # Create LLM function using OpenAI
+        llm_fn = create_openai_llm_fn(model="gpt-4o-mini")
+        
+        agent = LiteAgent(
+            name="LiteTestAgent",
+            llm_fn=llm_fn,
+            instructions="You are a helpful assistant. Keep responses very brief."
+        )
+        
+        response = agent.chat("Say 'hello' and nothing else.")
+        
+        assert response is not None
+        assert len(response) > 0
+        
+        print(f"[PASS] LiteAgent with OpenAI response received")
+    
+    def test_lite_agent_no_litellm_loaded(self):
+        """Verify LiteAgent doesn't load litellm."""
+        from praisonaiagents.lite import LiteAgent, create_openai_llm_fn
+        
+        llm_fn = create_openai_llm_fn(model="gpt-4o-mini")
+        agent = LiteAgent(llm_fn=llm_fn)
+        
+        response = agent.chat("Hi")
+        
+        assert response is not None
+        assert 'litellm' not in sys.modules, \
+            "LiteAgent should not load litellm"
+        
+        print(f"[PASS] LiteAgent works without litellm")
+
+
+class TestAnthropicAPI:
+    """Test with Anthropic API (optional)."""
+    
+    @pytest.fixture(autouse=True)
+    def setup(self):
+        """Set up test fixtures."""
+        clear_modules()
+        
+        api_key = os.environ.get('ANTHROPIC_API_KEY')
+        if not api_key:
+            pytest.skip("ANTHROPIC_API_KEY not set")
+        
+        yield
+        clear_modules()
+    
+    def test_agent_with_claude(self):
+        """Test Agent with Claude model."""
+        from praisonaiagents import Agent
+        
+        agent = Agent(
+            name="ClaudeAgent",
+            instructions="You are a helpful assistant. Keep responses very brief.",
+            llm="claude-3-5-sonnet-20241022",
+            verbose=False
+        )
+        
+        response = agent.chat("Say 'hello' and nothing else.")
+        
+        assert response is not None
+        assert len(response) > 0
+        
+        print(f"[PASS] Claude response received")
+
+
+class TestGoogleAPI:
+    """Test with Google API (optional)."""
+    
+    @pytest.fixture(autouse=True)
+    def setup(self):
+        """Set up test fixtures."""
+        clear_modules()
+        
+        api_key = os.environ.get('GOOGLE_API_KEY')
+        if not api_key:
+            pytest.skip("GOOGLE_API_KEY not set")
+        
+        yield
+        clear_modules()
+    
+    def test_agent_with_gemini(self):
+        """Test Agent with Gemini model."""
+        from praisonaiagents import Agent
+        
+        agent = Agent(
+            name="GeminiAgent",
+            instructions="You are a helpful assistant. Keep responses very brief.",
+            llm="gemini/gemini-2.0-flash",
+            verbose=False
+        )
+        
+        response = agent.chat("Say 'hello' and nothing else.")
+        
+        assert response is not None
+        assert len(response) > 0
+        
+        print(f"[PASS] Gemini response received")
+
+
+if __name__ == "__main__":
+    # Run with real API keys
+    if os.environ.get('RUN_REAL_KEY_TESTS', '').lower() not in ('1', 'true', 'yes'):
+        print("Set RUN_REAL_KEY_TESTS=1 to run these tests")
+        sys.exit(1)
+    
+    pytest.main([__file__, "-v", "-s"])

src/praisonai-agents/tests/test_http_stream_basic.py

@@ -5,11 +5,16 @@ Basic test to verify HTTP Stream implementation is working correctly.
 
 import sys
 import os
+import pytest
 
 # Add parent directory to path for imports
 sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
 
-from praisonaiagents.mcp import MCP
+# Skip entire module if mcp package is not installed
+try:
+    from praisonaiagents.mcp import MCP
+except ImportError:
+    pytest.skip("MCP package not installed", allow_module_level=True)
 
 def test_transport_selection():
     """Test that URLs are correctly routed to appropriate transports."""

src/praisonai-agents/tests/test_multiple_mcp_tools.py

@@ -18,8 +18,12 @@ from unittest.mock import Mock
 # Add the package to path for testing
 sys.path.insert(0, os.path.join(os.path.dirname(__file__), '..'))
 
-from praisonaiagents.mcp.mcp import MCP
-from praisonaiagents.agent.agent import Agent
+# Skip entire module if mcp package is not installed
+try:
+    from praisonaiagents.mcp.mcp import MCP
+    from praisonaiagents.agent.agent import Agent
+except ImportError:
+    pytest.skip("MCP package not installed", allow_module_level=True)
 
 
 # ============================================================================

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

@@ -0,0 +1,217 @@
+"""
+Tests for lazy import behavior in praisonaiagents.
+
+These tests verify that heavy dependencies like litellm are NOT loaded
+at package import time, ensuring fast startup.
+"""
+import sys
+import pytest
+
+
+def clear_modules():
+    """Clear all praisonai and litellm related modules from cache."""
+    to_remove = [m for m in list(sys.modules.keys()) 
+                 if 'praison' in m or 'litellm' in m]
+    for mod in to_remove:
+        del sys.modules[mod]
+
+
+class TestLazyImports:
+    """Test that heavy dependencies are lazily loaded."""
+    
+    def setup_method(self):
+        """Clear module cache before each test."""
+        clear_modules()
+    
+    def teardown_method(self):
+        """Clean up after each test."""
+        clear_modules()
+    
+    def test_litellm_not_loaded_on_import(self):
+        """litellm should NOT be loaded when importing praisonaiagents."""
+        import praisonaiagents  # noqa: F401
+        
+        assert 'litellm' not in sys.modules, \
+            "litellm should not be loaded at import time"
+    
+    def test_requests_not_loaded_on_import(self):
+        """requests should NOT be loaded when importing praisonaiagents."""
+        import praisonaiagents  # noqa: F401
+        
+        assert 'requests' not in sys.modules, \
+            "requests should not be loaded at import time"
+    
+    def test_chromadb_not_loaded_on_import(self):
+        """chromadb should NOT be loaded when importing praisonaiagents."""
+        import praisonaiagents  # noqa: F401
+        
+        assert 'chromadb' not in sys.modules, \
+            "chromadb should not be loaded at import time"
+    
+    def test_agent_available_via_getattr(self):
+        """Agent should be available via lazy loading."""
+        import praisonaiagents
+        
+        # Access Agent - this triggers lazy loading
+        Agent = praisonaiagents.Agent
+        
+        assert Agent is not None
+        assert hasattr(Agent, 'chat')
+    
+    def test_session_available_via_getattr(self):
+        """Session should be available via lazy loading."""
+        import praisonaiagents
+        
+        # Access Session - this triggers lazy loading
+        Session = praisonaiagents.Session
+        
+        assert Session is not None
+    
+    def test_knowledge_available_via_getattr(self):
+        """Knowledge should be available via lazy loading."""
+        import praisonaiagents
+        
+        # Access Knowledge - this triggers lazy loading
+        Knowledge = praisonaiagents.Knowledge
+        
+        assert Knowledge is not None
+    
+    def test_memory_available_via_getattr(self):
+        """Memory should be available via lazy loading."""
+        import praisonaiagents
+        
+        # Access Memory - this triggers lazy loading
+        Memory = praisonaiagents.Memory
+        
+        assert Memory is not None
+    
+    def test_tools_loaded_eagerly(self):
+        """Tools should be available immediately (lightweight)."""
+        import praisonaiagents
+        
+        # Tools is lightweight and should be available
+        assert hasattr(praisonaiagents, 'Tools')
+        assert praisonaiagents.Tools is not None
+    
+    def test_all_exports_accessible(self):
+        """All items in __all__ should be accessible."""
+        import praisonaiagents
+        
+        for name in praisonaiagents.__all__:
+            # Some items may be None if optional deps not installed
+            try:
+                attr = getattr(praisonaiagents, name)
+                # Just verify it's accessible, may be None
+            except AttributeError:
+                pytest.fail(f"'{name}' in __all__ but not accessible")
+
+
+class TestLitePackage:
+    """Test the lite subpackage."""
+    
+    def setup_method(self):
+        """Clear module cache before each test."""
+        clear_modules()
+    
+    def test_lite_import_fast(self):
+        """Lite package should import quickly without heavy deps."""
+        import time
+        
+        start = time.perf_counter()
+        from praisonaiagents.lite import LiteAgent  # noqa: F401
+        end = time.perf_counter()
+        
+        import_time_ms = (end - start) * 1000
+        
+        # Lite should import in under 100ms
+        assert import_time_ms < 100, \
+            f"Lite import took {import_time_ms:.0f}ms, expected < 100ms"
+    
+    def test_lite_agent_no_litellm(self):
+        """LiteAgent should not require litellm."""
+        from praisonaiagents.lite import LiteAgent
+        
+        # Create agent without LLM function
+        agent = LiteAgent(name="Test")
+        
+        assert agent.name == "Test"
+        assert 'litellm' not in sys.modules
+    
+    def test_lite_agent_with_custom_llm(self):
+        """LiteAgent should work with custom LLM function."""
+        from praisonaiagents.lite import LiteAgent
+        
+        # Create a mock LLM function
+        def mock_llm(messages):
+            return "Hello from mock LLM!"
+        
+        agent = LiteAgent(
+            name="TestAgent",
+            llm_fn=mock_llm,
+            instructions="You are a test agent."
+        )
+        
+        response = agent.chat("Hi")
+        
+        assert response == "Hello from mock LLM!"
+        assert len(agent.chat_history) == 2  # user + assistant
+
+
+class TestTelemetryConfig:
+    """Test telemetry configuration."""
+    
+    def setup_method(self):
+        """Clear module cache and env vars before each test."""
+        clear_modules()
+        import os
+        # Save original env vars
+        self._orig_env = {}
+        for key in ['PRAISONAI_TELEMETRY_ENABLED', 'PRAISONAI_TELEMETRY_DISABLED', 
+                    'DO_NOT_TRACK', 'PRAISONAI_DISABLE_TELEMETRY']:
+            self._orig_env[key] = os.environ.get(key)
+            if key in os.environ:
+                del os.environ[key]
+    
+    def teardown_method(self):
+        """Restore env vars after each test."""
+        import os
+        for key, value in self._orig_env.items():
+            if value is None:
+                if key in os.environ:
+                    del os.environ[key]
+            else:
+                os.environ[key] = value
+        clear_modules()
+    
+    def test_telemetry_disabled_by_default(self):
+        """Telemetry should be disabled by default (opt-in model)."""
+        from praisonaiagents._config import TELEMETRY_ENABLED
+        
+        assert TELEMETRY_ENABLED is False, \
+            "Telemetry should be disabled by default"
+    
+    def test_telemetry_enabled_via_env(self):
+        """Telemetry can be enabled via environment variable."""
+        import os
+        os.environ['PRAISONAI_TELEMETRY_ENABLED'] = 'true'
+        
+        # Re-import to pick up new env var
+        clear_modules()
+        from praisonaiagents._config import _get_telemetry_enabled
+        
+        assert _get_telemetry_enabled() is True
+    
+    def test_do_not_track_takes_precedence(self):
+        """DO_NOT_TRACK should disable telemetry even if enabled."""
+        import os
+        os.environ['PRAISONAI_TELEMETRY_ENABLED'] = 'true'
+        os.environ['DO_NOT_TRACK'] = 'true'
+        
+        clear_modules()
+        from praisonaiagents._config import _get_telemetry_enabled
+        
+        assert _get_telemetry_enabled() is False
+
+
+if __name__ == "__main__":
+    pytest.main([__file__, "-v"])

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

@@ -0,0 +1,180 @@
+"""
+Tests for thread safety in praisonaiagents.
+
+These tests verify that Agent chat_history and caches are thread-safe
+when accessed concurrently.
+"""
+import sys
+import threading
+import time
+import pytest
+
+
+def clear_modules():
+    """Clear all praisonai and litellm related modules from cache."""
+    to_remove = [m for m in list(sys.modules.keys()) 
+                 if 'praison' in m or 'litellm' in m]
+    for mod in to_remove:
+        del sys.modules[mod]
+
+
+class TestLiteAgentThreadSafety:
+    """Test thread safety of LiteAgent (no external deps)."""
+    
+    def test_concurrent_chat_history_access(self):
+        """Multiple threads should safely access chat_history."""
+        from praisonaiagents.lite import LiteAgent
+        
+        # Create agent with mock LLM
+        call_count = [0]
+        lock = threading.Lock()
+        
+        def mock_llm(messages):
+            with lock:
+                call_count[0] += 1
+            time.sleep(0.01)  # Simulate some work
+            return f"Response {call_count[0]}"
+        
+        agent = LiteAgent(llm_fn=mock_llm)
+        
+        errors = []
+        
+        def worker(thread_id):
+            try:
+                for i in range(5):
+                    response = agent.chat(f"Message from thread {thread_id}, iteration {i}")
+                    assert response is not None
+            except Exception as e:
+                errors.append(f"Thread {thread_id}: {e}")
+        
+        # Start multiple threads
+        threads = []
+        for i in range(4):
+            t = threading.Thread(target=worker, args=(i,))
+            threads.append(t)
+            t.start()
+        
+        # Wait for all threads
+        for t in threads:
+            t.join()
+        
+        # Check for errors
+        assert len(errors) == 0, f"Thread errors: {errors}"
+        
+        # Verify history was updated (should have entries from all threads)
+        assert len(agent.chat_history) > 0
+    
+    def test_concurrent_clear_history(self):
+        """Clearing history while other threads are chatting should be safe."""
+        from praisonaiagents.lite import LiteAgent
+        
+        def mock_llm(messages):
+            return "Response"
+        
+        agent = LiteAgent(llm_fn=mock_llm)
+        
+        errors = []
+        stop_flag = threading.Event()
+        
+        def chat_worker():
+            try:
+                while not stop_flag.is_set():
+                    agent.chat("Hello")
+                    time.sleep(0.001)
+            except Exception as e:
+                errors.append(f"Chat error: {e}")
+        
+        def clear_worker():
+            try:
+                for _ in range(10):
+                    time.sleep(0.01)
+                    agent.clear_history()
+            except Exception as e:
+                errors.append(f"Clear error: {e}")
+        
+        # Start chat threads
+        chat_threads = []
+        for _ in range(2):
+            t = threading.Thread(target=chat_worker)
+            chat_threads.append(t)
+            t.start()
+        
+        # Start clear thread
+        clear_thread = threading.Thread(target=clear_worker)
+        clear_thread.start()
+        
+        # Wait for clear thread to finish
+        clear_thread.join()
+        
+        # Stop chat threads
+        stop_flag.set()
+        for t in chat_threads:
+            t.join()
+        
+        assert len(errors) == 0, f"Thread errors: {errors}"
+
+
+class TestAgentThreadSafety:
+    """Test thread safety of main Agent class.
+    
+    Note: These tests use a mock LLM to avoid network calls.
+    """
+    
+    def setup_method(self):
+        """Set up test fixtures."""
+        clear_modules()
+    
+    def test_agent_has_history_lock(self):
+        """Agent should have a history lock for thread safety."""
+        from praisonaiagents import Agent
+        
+        agent = Agent(
+            name="Test",
+            instructions="Test agent",
+            llm="gpt-4o-mini"  # Won't actually call
+        )
+        
+        assert hasattr(agent, '_history_lock')
+        assert isinstance(agent._history_lock, type(threading.Lock()))
+    
+    def test_agent_has_cache_lock(self):
+        """Agent should have a cache lock for thread safety."""
+        from praisonaiagents import Agent
+        
+        agent = Agent(
+            name="Test",
+            instructions="Test agent",
+            llm="gpt-4o-mini"
+        )
+        
+        assert hasattr(agent, '_cache_lock')
+        # RLock is a different type than Lock
+        assert agent._cache_lock is not None
+
+
+class TestMCPCleanup:
+    """Test MCP cleanup and context manager support."""
+    
+    def test_mcp_has_context_manager(self):
+        """MCP should support context manager protocol."""
+        try:
+            from praisonaiagents.mcp.mcp import MCP
+        except ImportError:
+            pytest.skip("MCP package not installed")
+        
+        assert hasattr(MCP, '__enter__')
+        assert hasattr(MCP, '__exit__')
+        assert hasattr(MCP, 'shutdown')
+    
+    def test_mcp_shutdown_method(self):
+        """MCP should have a shutdown method."""
+        try:
+            from praisonaiagents.mcp.mcp import MCP
+        except ImportError:
+            pytest.skip("MCP package not installed")
+        
+        assert callable(getattr(MCP, 'shutdown', None))
+
+
+if __name__ == "__main__":
+    pytest.main([__file__, "-v"])

src/praisonai-agents/uv.lock

@@ -2422,7 +2422,7 @@ wheels = [
 
 [[package]]
 name = "praisonaiagents"
-version = "0.4.1"
+version = "0.5.0"
 source = { editable = "." }
 dependencies = [
     { name = "aiohttp" },

src/praisonai/praisonai.rb

@@ -3,8 +3,8 @@ class Praisonai < Formula
   
     desc "AI tools for various AI applications"
     homepage "https://github.com/MervinPraison/PraisonAI"
-    url "https://github.com/MervinPraison/PraisonAI/archive/refs/tags/v2.6.8.tar.gz"
-    sha256 `curl -sL https://github.com/MervinPraison/PraisonAI/archive/refs/tags/v2.6.8.tar.gz | shasum -a 256`.split.first
+    url "https://github.com/MervinPraison/PraisonAI/archive/refs/tags/v2.7.0.tar.gz"
+    sha256 `curl -sL https://github.com/MervinPraison/PraisonAI/archive/refs/tags/v2.7.0.tar.gz | shasum -a 256`.split.first
     license "MIT"
   
     depends_on "[email protected]"

src/praisonai/praisonai/cli/main.py

@@ -61,11 +61,23 @@ import subprocess
 import logging
 import importlib
 
-from praisonai.auto import AutoGenerator
-from praisonai.agents_generator import AgentsGenerator
+# Lazy imports for performance - these are imported when needed, not at module load
+# from praisonai.auto import AutoGenerator  # Lazy: imported in auto/init commands
+# from praisonai.agents_generator import AgentsGenerator  # Lazy: imported when running agents
 from praisonai.inbuilt_tools import *
 from praisonai.inc.config import generate_config
 
+# Lazy import helpers for heavy modules
+def _get_auto_generator():
+    """Lazy import AutoGenerator to avoid loading heavy deps at CLI startup."""
+    from praisonai.auto import AutoGenerator
+    return AutoGenerator
+
+def _get_agents_generator():
+    """Lazy import AgentsGenerator to avoid loading heavy deps at CLI startup."""
+    from praisonai.agents_generator import AgentsGenerator
+    return AgentsGenerator
+
 # Optional module imports with availability checks
 CHAINLIT_AVAILABLE = False
 GRADIO_AVAILABLE = False
@@ -564,8 +576,10 @@ class PraisonAI:
             self.topic = temp_topic
 
             self.agent_file = "test.yaml"
+            AutoGenerator = _get_auto_generator()
             generator = AutoGenerator(topic=self.topic, framework=self.framework, agent_file=self.agent_file)
             self.agent_file = generator.generate(merge=getattr(args, 'merge', False))
+            AgentsGenerator = _get_agents_generator()
             agents_generator = AgentsGenerator(self.agent_file, self.framework, self.config_list)
             result = agents_generator.generate_crew_and_kickoff()
             print(result)
@@ -577,6 +591,7 @@ class PraisonAI:
             self.topic = temp_topic
 
             self.agent_file = "agents.yaml"
+            AutoGenerator = _get_auto_generator()
             generator = AutoGenerator(topic=self.topic, framework=self.framework, agent_file=self.agent_file)
             self.agent_file = generator.generate(merge=getattr(args, 'merge', False))
             print(f"File {self.agent_file} created successfully")
@@ -589,6 +604,7 @@ class PraisonAI:
                 self.create_chainlit_interface()
             else:
                 # Modify code to allow default UI
+                AgentsGenerator = _get_agents_generator()
                 agents_generator = AgentsGenerator(
                     self.agent_file,
                     self.framework,
@@ -626,6 +642,7 @@ class PraisonAI:
             except Exception:
                 pass  # Continue even if flow display fails
             
+            AgentsGenerator = _get_agents_generator()
             agents_generator = AgentsGenerator(
                 self.agent_file,
                 self.framework,
@@ -4284,8 +4301,10 @@ Now, {final_instruction.lower()}:"""
             def generate_crew_and_kickoff_interface(auto_args, framework):
                 self.framework = framework
                 self.agent_file = "test.yaml"
+                AutoGenerator = _get_auto_generator()
                 generator = AutoGenerator(topic=auto_args, framework=self.framework)
                 self.agent_file = generator.generate()
+                AgentsGenerator = _get_agents_generator()
                 agents_generator = AgentsGenerator(self.agent_file, self.framework, self.config_list)
                 result = agents_generator.generate_crew_and_kickoff()
                 return result

src/praisonai/praisonai/deploy.py

@@ -57,7 +57,7 @@ class CloudDeployer:
             file.write("FROM python:3.11-slim\n")
             file.write("WORKDIR /app\n")
             file.write("COPY . .\n")
-            file.write("RUN pip install flask praisonai==2.6.8 gunicorn markdown\n")
+            file.write("RUN pip install flask praisonai==2.7.0 gunicorn markdown\n")
             file.write("EXPOSE 8080\n")
             file.write('CMD ["gunicorn", "-b", "0.0.0.0:8080", "api:app"]\n')
             

src/praisonai/praisonai/inc/models.py

@@ -3,7 +3,8 @@ import os
 import logging
 from urllib.parse import urlparse
 logger = logging.getLogger(__name__)
-logging.basicConfig(level=os.environ.get('LOGLEVEL', 'INFO').upper(), format='%(asctime)s - %(levelname)s - %(message)s')
+_loglevel = os.environ.get('LOGLEVEL', 'INFO').strip().upper() or 'INFO'
+logging.basicConfig(level=_loglevel, format='%(asctime)s - %(levelname)s - %(message)s')
 
 # Constants
 LOCAL_SERVER_API_KEY_PLACEHOLDER = "not-needed"

src/praisonai/praisonai/templates/discovery.py

@@ -3,12 +3,14 @@ Template Discovery
 
 Discovers templates from multiple directories with precedence support.
 Supports custom user directories, built-in templates, and package templates.
+Includes caching to avoid repeated filesystem scans.
 """
 
 import os
+import time
 from pathlib import Path
 from typing import Dict, List, Optional, Tuple
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 
 
 @dataclass
@@ -25,6 +27,14 @@ class DiscoveredTemplate:
     author: Optional[str] = None
 
 
+@dataclass
+class _CacheEntry:
+    """Cache entry for template discovery results."""
+    templates: Dict[str, 'DiscoveredTemplate']
+    timestamp: float
+    mtimes: Dict[str, float] = field(default_factory=dict)  # path -> mtime for invalidation
+
+
 class TemplateDiscovery:
     """
     Discovers templates from multiple directories with precedence.
@@ -36,6 +46,12 @@ class TemplateDiscovery:
     4. Built-in package templates (agent_recipes)
     
     Templates in higher priority directories override those in lower priority.
+    
+    Caching:
+    - Results are cached to avoid repeated filesystem scans
+    - Cache is invalidated when directory mtimes change
+    - Use refresh=True to force a fresh scan
+    - Default TTL is 300 seconds (5 minutes)
     """
     
     # Default search paths (in priority order)
@@ -47,11 +63,18 @@ class TemplateDiscovery:
     
     TEMPLATE_FILE = "TEMPLATE.yaml"
     
+    # Cache TTL in seconds (5 minutes default)
+    CACHE_TTL = 300
+    
+    # Class-level cache shared across instances with same search paths
+    _cache: Dict[str, _CacheEntry] = {}
+    
     def __init__(
         self,
         custom_dirs: Optional[List[str]] = None,
         include_package: bool = True,
-        include_defaults: bool = True
+        include_defaults: bool = True,
+        cache_ttl: Optional[int] = None
     ):
         """
         Initialize template discovery.
@@ -60,6 +83,7 @@ class TemplateDiscovery:
             custom_dirs: Additional custom directories to search (highest priority)
             include_package: Whether to include package templates (agent_recipes)
             include_defaults: Whether to include default search paths
+            cache_ttl: Cache time-to-live in seconds (default: 300)
         """
         self.search_paths: List[Tuple[Path, str, int]] = []
         
@@ -78,6 +102,40 @@ class TemplateDiscovery:
         
         self.include_package = include_package
         self._package_path: Optional[Path] = None
+        self._cache_ttl = cache_ttl if cache_ttl is not None else self.CACHE_TTL
+        
+        # Generate cache key based on search paths configuration
+        self._cache_key = self._generate_cache_key()
+    
+    def _generate_cache_key(self) -> str:
+        """Generate a unique cache key based on search paths."""
+        paths_str = "|".join(
+            f"{p}:{s}:{pr}" for p, s, pr in self.search_paths
+        )
+        return f"{paths_str}|pkg={self.include_package}"
+    
+    def _get_dir_mtime(self, path: Path) -> float:
+        """Get directory modification time, or 0 if not accessible."""
+        try:
+            if path.exists():
+                return path.stat().st_mtime
+        except (OSError, PermissionError):
+            pass
+        return 0.0
+    
+    def _is_cache_valid(self, entry: _CacheEntry) -> bool:
+        """Check if cache entry is still valid."""
+        # Check TTL
+        if time.time() - entry.timestamp > self._cache_ttl:
+            return False
+        
+        # Check if any directory mtimes have changed
+        for path_str, cached_mtime in entry.mtimes.items():
+            current_mtime = self._get_dir_mtime(Path(path_str))
+            if current_mtime != cached_mtime:
+                return False
+        
+        return True
     
     @property
     def package_templates_path(self) -> Optional[Path]:
@@ -97,15 +155,28 @@ class TemplateDiscovery:
                 pass
         return self._package_path
     
-    def discover_all(self) -> Dict[str, DiscoveredTemplate]:
+    def discover_all(self, refresh: bool = False) -> Dict[str, DiscoveredTemplate]:
         """
         Discover all templates from all search paths.
         
+        Results are cached to avoid repeated filesystem scans.
+        Cache is invalidated when directory mtimes change or TTL expires.
+        
+        Args:
+            refresh: If True, bypass cache and force a fresh scan
+        
         Returns:
             Dict mapping template name to DiscoveredTemplate.
             When duplicates exist, higher priority wins.
         """
+        # Check cache first (unless refresh requested)
+        if not refresh and self._cache_key in self._cache:
+            entry = self._cache[self._cache_key]
+            if self._is_cache_valid(entry):
+                return entry.templates.copy()
+        
         templates: Dict[str, DiscoveredTemplate] = {}
+        mtimes: Dict[str, float] = {}
         
         # Scan search paths in reverse priority order (lowest first)
         # so that higher priority overwrites lower
@@ -120,11 +191,21 @@ class TemplateDiscovery:
         all_paths.sort(key=lambda x: x[2], reverse=True)
         
         for dir_path, source, priority in all_paths:
+            # Record mtime for cache invalidation
+            mtimes[str(dir_path)] = self._get_dir_mtime(dir_path)
+            
             if dir_path.exists() and dir_path.is_dir():
                 discovered = self._scan_directory(dir_path, source, priority)
                 # Higher priority overwrites lower
                 templates.update(discovered)
         
+        # Store in cache
+        self._cache[self._cache_key] = _CacheEntry(
+            templates=templates.copy(),
+            timestamp=time.time(),
+            mtimes=mtimes
+        )
+        
         return templates
     
     def _scan_directory(
@@ -191,7 +272,7 @@ class TemplateDiscovery:
         except Exception:
             pass  # Metadata loading is optional
     
-    def find_template(self, name: str) -> Optional[DiscoveredTemplate]:
+    def find_template(self, name: str, refresh: bool = False) -> Optional[DiscoveredTemplate]:
         """
         Find a template by name.
         
@@ -200,11 +281,12 @@ class TemplateDiscovery:
         
         Args:
             name: Template name to find
+            refresh: If True, bypass cache and force a fresh scan
             
         Returns:
             DiscoveredTemplate if found, None otherwise
         """
-        all_templates = self.discover_all()
+        all_templates = self.discover_all(refresh=refresh)
         return all_templates.get(name)
     
     def resolve_template_path(self, name: str) -> Optional[Path]:
@@ -222,24 +304,36 @@ class TemplateDiscovery:
     
     def list_templates(
         self,
-        source_filter: Optional[str] = None
+        source_filter: Optional[str] = None,
+        refresh: bool = False
     ) -> List[DiscoveredTemplate]:
         """
         List all discovered templates.
         
         Args:
             source_filter: Optional filter by source ('custom', 'package', etc.)
+            refresh: If True, bypass cache and force a fresh scan
             
         Returns:
             List of DiscoveredTemplate objects
         """
-        templates = self.discover_all()
+        templates = self.discover_all(refresh=refresh)
         
         if source_filter:
             return [t for t in templates.values() if t.source == source_filter]
         
         return list(templates.values())
     
+    def clear_cache(self) -> None:
+        """Clear the template discovery cache."""
+        if self._cache_key in self._cache:
+            del self._cache[self._cache_key]
+    
+    @classmethod
+    def clear_all_caches(cls) -> None:
+        """Clear all template discovery caches."""
+        cls._cache.clear()
+    
     def get_search_paths(self) -> List[Tuple[str, str, bool]]:
         """
         Get all search paths with their status.

src/praisonai/praisonai/version.py

@@ -1 +1 @@
-__version__ = "2.6.8"
+__version__ = "2.7.0"

src/praisonai/pyproject.toml

@@ -12,7 +12,7 @@ dependencies = [
     "rich>=13.7",
     "markdown>=3.5",
     "pyparsing>=3.0.0",
-    "praisonaiagents>=0.4.1",
+    "praisonaiagents>=0.5.0",
     "python-dotenv>=0.19.0",
     "instructor>=1.3.3",
     "PyYAML>=6.0",

src/praisonai/tests/integration/test_real_key_smoke.py

@@ -0,0 +1,230 @@
+"""
+Real API Key Smoke Tests for PraisonAI.
+
+These tests validate end-to-end behavior using real API keys.
+They are gated by environment variables and use minimal tokens.
+
+Run with:
+    RUN_REAL_KEY_TESTS=1 pytest tests/integration/test_real_key_smoke.py -v
+
+Required environment variables:
+    - OPENAI_API_KEY: For OpenAI provider tests
+    - RUN_REAL_KEY_TESTS=1: Gate to enable these tests
+"""
+
+import os
+import sys
+import subprocess
+import pytest
+import time
+
+# Gate: Skip all tests if RUN_REAL_KEY_TESTS is not set
+pytestmark = pytest.mark.skipif(
+    os.environ.get("RUN_REAL_KEY_TESTS") != "1",
+    reason="Real key tests disabled. Set RUN_REAL_KEY_TESTS=1 to enable."
+)
+
+# Paths
+PRAISONAI_PATH = os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+PRAISONAI_AGENTS_PATH = os.path.join(os.path.dirname(PRAISONAI_PATH), "praisonai-agents")
+PYTHONPATH = f"{PRAISONAI_PATH}:{PRAISONAI_AGENTS_PATH}"
+
+
+def run_python_code(code: str, timeout: int = 30) -> subprocess.CompletedProcess:
+    """Run Python code in a subprocess with proper PYTHONPATH."""
+    env = os.environ.copy()
+    env['PYTHONPATH'] = PYTHONPATH
+    return subprocess.run(
+        [sys.executable, '-c', code],
+        capture_output=True,
+        text=True,
+        cwd=PRAISONAI_PATH,
+        env=env,
+        timeout=timeout
+    )
+
+
+class TestSDKRealKey:
+    """Test SDK functionality with real API keys."""
+    
+    def test_openai_api_key_present(self):
+        """Verify OPENAI_API_KEY is set."""
+        api_key = os.environ.get("OPENAI_API_KEY")
+        assert api_key is not None, "OPENAI_API_KEY not set"
+        assert api_key.startswith("sk-"), "OPENAI_API_KEY format invalid"
+        # Never print the key
+        print("OPENAI_API_KEY: [REDACTED]")
+    
+    def test_agent_minimal_chat(self):
+        """Test minimal Agent.chat() with real OpenAI API."""
+        code = '''
+import os
+from praisonaiagents import Agent
+
+# Minimal agent with low token usage
+agent = Agent(
+    instructions="Reply with exactly one word: OK",
+    llm="gpt-4o-mini",
+    verbose=False
+)
+
+# Minimal prompt
+response = agent.chat("Say OK")
+print(f"Response received: {len(response)} chars")
+print("SUCCESS" if response else "FAIL")
+'''
+        result = run_python_code(code, timeout=60)
+        assert result.returncode == 0, f"Agent chat failed: {result.stderr}"
+        assert "SUCCESS" in result.stdout, f"Agent chat did not succeed: {result.stdout}"
+        print(f"Agent chat test passed. Output length: {result.stdout}")
+    
+    def test_agent_with_tool(self):
+        """Test Agent with a simple tool."""
+        code = '''
+import os
+from praisonaiagents import Agent
+
+def get_time():
+    """Returns current time."""
+    import datetime
+    return datetime.datetime.now().strftime("%H:%M")
+
+agent = Agent(
+    instructions="You have access to get_time tool. Use it and report the time briefly.",
+    llm="gpt-4o-mini",
+    tools=[get_time],
+    verbose=False
+)
+
+response = agent.chat("What time is it?")
+print(f"Response: {len(response)} chars")
+print("SUCCESS" if response and ":" in response else "PARTIAL")
+'''
+        result = run_python_code(code, timeout=60)
+        assert result.returncode == 0, f"Tool test failed: {result.stderr}"
+        assert "SUCCESS" in result.stdout or "PARTIAL" in result.stdout, f"Tool test output: {result.stdout}"
+
+
+class TestCLIRealKey:
+    """Test CLI functionality with real API keys."""
+    
+    def test_cli_import(self):
+        """Test CLI module imports correctly."""
+        code = '''
+import os
+os.environ['LOGLEVEL'] = 'WARNING'
+from praisonai.cli.main import PraisonAI
+print("CLI_IMPORT_OK")
+'''
+        result = run_python_code(code, timeout=30)
+        assert result.returncode == 0, f"CLI import failed: {result.stderr}"
+        assert "CLI_IMPORT_OK" in result.stdout
+    
+    def test_cli_praisonai_class(self):
+        """Test PraisonAI class instantiation."""
+        code = '''
+import os
+os.environ['LOGLEVEL'] = 'WARNING'
+from praisonai.cli.main import PraisonAI
+pai = PraisonAI(agent_file="agents.yaml", framework="praisonai")
+print("CLASS_OK")
+'''
+        result = run_python_code(code, timeout=30)
+        assert result.returncode == 0, f"CLI class failed: {result.stderr}"
+        assert "CLASS_OK" in result.stdout
+
+
+class TestPerformanceWithRealKey:
+    """Test that performance optimizations work with real usage."""
+    
+    def test_cli_startup_still_fast(self):
+        """Verify CLI startup is still fast after changes."""
+        code = '''
+import time
+import os
+os.environ['LOGLEVEL'] = 'WARNING'
+t = time.time()
+from praisonai.cli.main import PraisonAI
+elapsed = time.time() - t
+print(f"ELAPSED:{elapsed:.3f}")
+if elapsed < 1.0:
+    print("FAST")
+else:
+    print("SLOW")
+'''
+        result = run_python_code(code, timeout=30)
+        assert result.returncode == 0, f"Startup test failed: {result.stderr}"
+        assert "FAST" in result.stdout, f"CLI startup too slow: {result.stdout}"
+    
+    def test_lazy_loading_preserved(self):
+        """Verify heavy modules not loaded at startup."""
+        code = '''
+import sys
+import os
+os.environ['LOGLEVEL'] = 'WARNING'
+from praisonai.cli.main import PraisonAI
+heavy = ['litellm', 'instructor', 'praisonai.auto']
+loaded = [m for m in heavy if m in sys.modules]
+if loaded:
+    print(f"LOADED:{loaded}")
+else:
+    print("LAZY_OK")
+'''
+        result = run_python_code(code, timeout=30)
+        assert result.returncode == 0, f"Lazy loading test failed: {result.stderr}"
+        assert "LAZY_OK" in result.stdout, f"Heavy modules loaded: {result.stdout}"
+
+
+class TestMultiProvider:
+    """Test multiple LLM providers if keys available."""
+    
+    @pytest.mark.skipif(
+        not os.environ.get("ANTHROPIC_API_KEY"),
+        reason="ANTHROPIC_API_KEY not set"
+    )
+    def test_anthropic_agent(self):
+        """Test Agent with Anthropic Claude."""
+        code = '''
+import os
+from praisonaiagents import Agent
+
+agent = Agent(
+    instructions="Reply with exactly: ANTHROPIC_OK",
+    llm="claude-3-haiku-20240307",
+    verbose=False
+)
+response = agent.chat("Respond")
+print("SUCCESS" if "ANTHROPIC_OK" in response or response else "FAIL")
+'''
+        result = run_python_code(code, timeout=60)
+        # May fail if Anthropic not configured, that's OK
+        print(f"Anthropic test: {result.stdout}")
+    
+    @pytest.mark.skipif(
+        not os.environ.get("GOOGLE_API_KEY"),
+        reason="GOOGLE_API_KEY not set"
+    )
+    def test_google_agent(self):
+        """Test Agent with Google Gemini."""
+        code = '''
+import os
+from praisonaiagents import Agent
+
+agent = Agent(
+    instructions="Reply with exactly: GOOGLE_OK",
+    llm="gemini/gemini-1.5-flash",
+    verbose=False
+)
+response = agent.chat("Respond")
+print("SUCCESS" if response else "FAIL")
+'''
+        result = run_python_code(code, timeout=60)
+        print(f"Google test: {result.stdout}")
+
+
+if __name__ == "__main__":
+    # Allow running directly
+    if os.environ.get("RUN_REAL_KEY_TESTS") != "1":
+        print("Set RUN_REAL_KEY_TESTS=1 to run these tests")
+        sys.exit(1)
+    pytest.main([__file__, "-v"])

src/praisonai/tests/unit/test_performance_benchmarks.py

@@ -0,0 +1,357 @@
+"""
+Performance benchmark tests for PraisonAI.
+
+These tests verify that import times, CLI startup, and memory usage
+stay within acceptable bounds. They serve as regression guards.
+
+Run with: pytest tests/unit/test_performance_benchmarks.py -v
+"""
+
+import subprocess
+import sys
+import os
+import pytest
+
+# Get the paths for praisonai and praisonaiagents
+PRAISONAI_PATH = os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+PRAISONAI_AGENTS_PATH = os.path.join(os.path.dirname(PRAISONAI_PATH), "praisonai-agents")
+
+# Build PYTHONPATH that includes both packages
+PYTHONPATH = f"{PRAISONAI_PATH}:{PRAISONAI_AGENTS_PATH}"
+
+# Thresholds (can be adjusted based on CI environment)
+PRAISONAI_WRAPPER_IMPORT_TIME_MS = 500  # Should be very fast (lazy)
+CLI_STARTUP_TIME_S = 1.0  # Target < 1s for CLI cold start
+CLI_MEMORY_MB = 50  # Target < 50MB for CLI cold start (without heavy deps)
+
+
+def run_python_code(code: str, cwd: str = None) -> subprocess.CompletedProcess:
+    """Run Python code in a subprocess with proper PYTHONPATH."""
+    env = os.environ.copy()
+    env['PYTHONPATH'] = PYTHONPATH
+    return subprocess.run(
+        [sys.executable, '-c', code],
+        capture_output=True,
+        text=True,
+        cwd=cwd or PRAISONAI_PATH,
+        env=env
+    )
+
+
+class TestImportTimeBenchmarks:
+    """Test import time performance."""
+    
+    def test_praisonai_wrapper_import_time(self):
+        """Test that praisonai wrapper imports quickly (lazy loading)."""
+        code = '''
+import time
+t = time.time()
+import praisonai
+elapsed_ms = (time.time() - t) * 1000
+print(f"{elapsed_ms:.0f}")
+'''
+        result = run_python_code(code)
+        assert result.returncode == 0, f"Import failed: {result.stderr}"
+        elapsed_ms = float(result.stdout.strip())
+        assert elapsed_ms < PRAISONAI_WRAPPER_IMPORT_TIME_MS, \
+            f"praisonai import took {elapsed_ms:.0f}ms, expected < {PRAISONAI_WRAPPER_IMPORT_TIME_MS}ms"
+    
+    def test_cli_main_import_time(self):
+        """Test that CLI main module imports quickly (lazy loading)."""
+        code = '''
+import time
+import os
+os.environ['LOGLEVEL'] = 'WARNING'
+t = time.time()
+from praisonai.cli.main import PraisonAI
+elapsed_ms = (time.time() - t) * 1000
+print(f"{elapsed_ms:.0f}")
+'''
+        result = run_python_code(code)
+        assert result.returncode == 0, f"Import failed: {result.stderr}"
+        elapsed_ms = float(result.stdout.strip())
+        # CLI should be fast due to lazy loading
+        assert elapsed_ms < CLI_STARTUP_TIME_S * 1000, \
+            f"CLI import took {elapsed_ms:.0f}ms, expected < {CLI_STARTUP_TIME_S * 1000}ms"
+
+
+class TestMemoryBenchmarks:
+    """Test memory usage."""
+    
+    def test_cli_cold_start_memory(self):
+        """Test that CLI cold start uses reasonable memory."""
+        code = '''
+import tracemalloc
+import os
+os.environ['LOGLEVEL'] = 'WARNING'
+tracemalloc.start()
+from praisonai.cli.main import PraisonAI
+current, peak = tracemalloc.get_traced_memory()
+print(f"{peak / 1024 / 1024:.1f}")
+'''
+        result = run_python_code(code)
+        assert result.returncode == 0, f"Memory test failed: {result.stderr}"
+        peak_mb = float(result.stdout.strip())
+        assert peak_mb < CLI_MEMORY_MB, \
+            f"CLI cold start used {peak_mb:.1f}MB, expected < {CLI_MEMORY_MB}MB"
+
+
+class TestLazyLoadingVerification:
+    """Verify that lazy loading is working correctly."""
+    
+    def test_auto_generator_not_imported_at_cli_startup(self):
+        """Verify AutoGenerator is not imported when CLI starts."""
+        code = '''
+import sys
+import os
+os.environ['LOGLEVEL'] = 'WARNING'
+from praisonai.cli.main import PraisonAI
+# Check if heavy modules are NOT loaded
+heavy_modules = ['praisonai.auto', 'praisonai.agents_generator', 'litellm', 'instructor']
+loaded = [m for m in heavy_modules if m in sys.modules]
+if loaded:
+    print(f"FAIL: {loaded}")
+else:
+    print("OK")
+'''
+        result = run_python_code(code)
+        assert result.returncode == 0, f"Test failed: {result.stderr}"
+        output = result.stdout.strip()
+        assert output == "OK", f"Heavy modules loaded at startup: {output}"
+    
+    def test_praisonaiagents_not_imported_at_cli_startup(self):
+        """Verify praisonaiagents is not imported when CLI starts."""
+        code = '''
+import sys
+import os
+os.environ['LOGLEVEL'] = 'WARNING'
+from praisonai.cli.main import PraisonAI
+# Check if praisonaiagents is NOT loaded
+if 'praisonaiagents' in sys.modules:
+    print("FAIL: praisonaiagents loaded")
+else:
+    print("OK")
+'''
+        result = run_python_code(code)
+        assert result.returncode == 0, f"Test failed: {result.stderr}"
+        output = result.stdout.strip()
+        assert output == "OK", f"praisonaiagents loaded at startup: {output}"
+
+
+class TestThreadSafetyToolRegistry:
+    """Test thread safety of tool registry."""
+    
+    def test_concurrent_tool_registration(self):
+        """Test that concurrent tool registration is thread-safe."""
+        code = '''
+import threading
+import time
+from praisonaiagents.tools.registry import ToolRegistry
+
+registry = ToolRegistry()
+errors = []
+registered_count = [0]
+lock = threading.Lock()
+
+def register_tools(thread_id, count):
+    try:
+        for i in range(count):
+            def tool_func():
+                pass
+            tool_func.__name__ = f"tool_{thread_id}_{i}"
+            registry.register(tool_func)
+            with lock:
+                registered_count[0] += 1
+    except Exception as e:
+        errors.append(str(e))
+
+# Create multiple threads that register tools concurrently
+threads = []
+num_threads = 10
+tools_per_thread = 50
+
+for i in range(num_threads):
+    t = threading.Thread(target=register_tools, args=(i, tools_per_thread))
+    threads.append(t)
+
+# Start all threads
+for t in threads:
+    t.start()
+
+# Wait for all threads to complete
+for t in threads:
+    t.join()
+
+# Verify results
+expected_total = num_threads * tools_per_thread
+actual_total = len(registry)
+
+if errors:
+    print(f"FAIL: Errors occurred: {errors}")
+elif actual_total != expected_total:
+    print(f"FAIL: Expected {expected_total} tools, got {actual_total}")
+else:
+    print("OK")
+'''
+        result = run_python_code(code)
+        assert result.returncode == 0, f"Test failed: {result.stderr}"
+        output = result.stdout.strip()
+        assert output == "OK", f"Thread safety test failed: {output}"
+    
+    def test_concurrent_tool_access(self):
+        """Test that concurrent tool access is thread-safe."""
+        code = '''
+import threading
+from praisonaiagents.tools.registry import ToolRegistry
+
+registry = ToolRegistry()
+errors = []
+
+# Pre-register some tools
+for i in range(100):
+    def tool_func():
+        pass
+    tool_func.__name__ = f"tool_{i}"
+    registry.register(tool_func)
+
+def access_tools(thread_id, iterations):
+    try:
+        for _ in range(iterations):
+            # Mix of read operations
+            registry.list_tools()
+            registry.get(f"tool_{thread_id % 100}")
+            len(registry)
+            f"tool_{thread_id % 100}" in registry
+    except Exception as e:
+        errors.append(str(e))
+
+# Create multiple threads that access tools concurrently
+threads = []
+num_threads = 20
+iterations = 100
+
+for i in range(num_threads):
+    t = threading.Thread(target=access_tools, args=(i, iterations))
+    threads.append(t)
+
+for t in threads:
+    t.start()
+
+for t in threads:
+    t.join()
+
+if errors:
+    print(f"FAIL: Errors occurred: {errors}")
+else:
+    print("OK")
+'''
+        result = run_python_code(code)
+        assert result.returncode == 0, f"Test failed: {result.stderr}"
+        output = result.stdout.strip()
+        assert output == "OK", f"Thread safety test failed: {output}"
+
+
+class TestTemplateCaching:
+    """Test template discovery caching."""
+    
+    def test_template_discovery_caching(self):
+        """Test that template discovery results are cached."""
+        code = '''
+import time
+from praisonai.templates.discovery import TemplateDiscovery
+
+# Create discovery instance
+discovery = TemplateDiscovery(include_package=False)
+
+# First call - should scan filesystem
+t1 = time.time()
+result1 = discovery.discover_all()
+time1 = time.time() - t1
+
+# Second call - should use cache
+t2 = time.time()
+result2 = discovery.discover_all()
+time2 = time.time() - t2
+
+# Third call with refresh - should scan again
+t3 = time.time()
+result3 = discovery.discover_all(refresh=True)
+time3 = time.time() - t3
+
+# Cache hit should be significantly faster
+# (at least 2x faster, typically 10x+)
+if time2 > 0 and time1 / time2 < 1.5:
+    # If times are very similar, cache might not be working
+    # But if both are very fast (<1ms), it's fine
+    if time1 > 0.001 and time2 > 0.001:
+        print(f"WARN: Cache may not be working. First: {time1*1000:.2f}ms, Second: {time2*1000:.2f}ms")
+    else:
+        print("OK")
+else:
+    print("OK")
+'''
+        result = run_python_code(code)
+        assert result.returncode == 0, f"Test failed: {result.stderr}"
+        output = result.stdout.strip()
+        assert "OK" in output or "WARN" in output, f"Cache test failed: {output}"
+    
+    def test_template_cache_clear(self):
+        """Test that template cache can be cleared."""
+        code = '''
+from praisonai.templates.discovery import TemplateDiscovery
+
+discovery = TemplateDiscovery(include_package=False)
+
+# Populate cache
+discovery.discover_all()
+
+# Clear cache
+discovery.clear_cache()
+
+# Verify cache is cleared by checking internal state
+if discovery._cache_key in TemplateDiscovery._cache:
+    print("FAIL: Cache not cleared")
+else:
+    print("OK")
+'''
+        result = run_python_code(code)
+        assert result.returncode == 0, f"Test failed: {result.stderr}"
+        output = result.stdout.strip()
+        assert output == "OK", f"Cache clear test failed: {output}"
+
+
+class TestSchemaVersioning:
+    """Test DB adapter schema versioning."""
+    
+    def test_schema_version_constant_exists(self):
+        """Test that SCHEMA_VERSION constant is defined."""
+        code = '''
+from praisonaiagents.db.protocol import SCHEMA_VERSION
+if SCHEMA_VERSION and isinstance(SCHEMA_VERSION, str):
+    print("OK")
+else:
+    print(f"FAIL: Invalid SCHEMA_VERSION: {SCHEMA_VERSION}")
+'''
+        result = run_python_code(code)
+        assert result.returncode == 0, f"Test failed: {result.stderr}"
+        output = result.stdout.strip()
+        assert output == "OK", f"Schema version test failed: {output}"
+    
+    def test_schema_version_format(self):
+        """Test that SCHEMA_VERSION follows MAJOR.MINOR format."""
+        code = r'''
+import re
+from praisonaiagents.db.protocol import SCHEMA_VERSION
+if re.match(r"^\d+\.\d+$", SCHEMA_VERSION):
+    print("OK")
+else:
+    print(f"FAIL: Invalid format: {SCHEMA_VERSION}")
+'''
+        result = run_python_code(code)
+        assert result.returncode == 0, f"Test failed: {result.stderr}"
+        output = result.stdout.strip()
+        assert output == "OK", f"Schema version format test failed: {output}"
+
+
+if __name__ == "__main__":
+    pytest.main([__file__, "-v"])

src/praisonai/uv.lock

@@ -4117,7 +4117,7 @@ requires-dist = [
     { name = "praisonai-tools", marker = "extra == 'autogen'", specifier = ">=0.1.0" },
     { name = "praisonai-tools", marker = "extra == 'autogen-v4'", specifier = ">=0.1.0" },
     { name = "praisonai-tools", marker = "extra == 'crewai'", specifier = ">=0.1.0" },
-    { name = "praisonaiagents", specifier = ">=0.4.1" },
+    { name = "praisonaiagents", specifier = ">=0.5.0" },
     { name = "pyautogen", marker = "extra == 'autogen'", specifier = "==0.2.29" },
     { name = "pydantic", marker = "extra == 'chat'", specifier = "<=2.10.1" },
     { name = "pydantic", marker = "extra == 'code'", specifier = "<=2.10.1" },
@@ -4159,7 +4159,7 @@ wheels = [
 
 [[package]]
 name = "praisonaiagents"
-version = "0.4.1"
+version = "0.5.0"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "aiohttp" },
@@ -4168,9 +4168,9 @@ dependencies = [
     { name = "pydantic" },
     { name = "rich" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/de/e8/5cf1306785b5e47b278363f58dee68ea381d26acc87ac9fd86b9d7abab0a/praisonaiagents-0.4.1.tar.gz", hash = "sha256:61cf37b3392c8857855176cdae6a758a88b110d103c61e226a4da7ab2ad8d30c", size = 766683, upload-time = "2025-12-28T07:27:21.999Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/c9/8d/70dc885f0091ef9383490184cdc5f401a3fd18ffe5875f1ba504ad1c7c56/praisonaiagents-0.5.0.tar.gz", hash = "sha256:3e9869c4af2d680b0679bee6e7c395b88597cab1fc8faedffd5009e4397f514e", size = 773203, upload-time = "2025-12-29T01:27:47.612Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/0d/26/b1593bf8d360793cc7e82020f7bef080734290370ddbb54658c6d9a0d8e1/praisonaiagents-0.4.1-py3-none-any.whl", hash = "sha256:0cdb9c8514cf042340c093103d9051a4d25dfb6d8126d05757b552f21b8b7085", size = 713637, upload-time = "2025-12-28T07:27:19.944Z" },
+    { url = "https://files.pythonhosted.org/packages/18/2a/53d830c8e481a38bf21dd723188e98ece83d317a723238ca16f37365c3ae/praisonaiagents-0.5.0-py3-none-any.whl", hash = "sha256:103da992be67b01141a9eea4d842a627ed6c027f29bb193d3443b72c27d14546", size = 720892, upload-time = "2025-12-29T01:27:45.518Z" },
 ]
 
 [[package]]