Release v2.3.80

- Check for @pytest.mark.integration marker
- Check if test file is in integration/ directory
- This prevents conftest from overriding real API keys with 'test-key'

.github/workflows/docker-publish.yml

@@ -1,45 +1,54 @@
-name: Daily Publish Docker
+name: Publish Docker
 
 on:
-  push:
-    branches:
-      - main
-  schedule:
-    - cron: '0 0 * * *'
+  release:
+    types: [published]
+  workflow_dispatch:
+    inputs:
+      tag:
+        description: 'Docker image tag (default: latest)'
+        required: false
+        default: 'latest'
 
 jobs:
-  check-and-build:
+  build-and-push:
     runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      packages: write
     
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v2
-
-      - name: Check last commit time
-        id: check_commit
-        run: |
-          last_commit_time=$(git log -1 --format=%ct)
-          current_time=$(date +%s)
-          time_diff=$((current_time - last_commit_time))
-          if [ $time_diff -ge 120 ] && [ $time_diff -lt 240 ]; then
-            echo "should_run=true" >> $GITHUB_OUTPUT
-          else
-            echo "should_run=false" >> $GITHUB_OUTPUT
-          fi
+        uses: actions/checkout@v4
 
       - name: Set up Docker Buildx
-        if: steps.check_commit.outputs.should_run == 'true'
-        uses: docker/setup-buildx-action@v1
+        uses: docker/setup-buildx-action@v3
 
-      - name: Log in to the GitHub Container Registry
-        if: steps.check_commit.outputs.should_run == 'true'
-        run: echo ${{ secrets.GH_TOKEN }} | docker login ghcr.io -u USERNAME --password-stdin
+      - name: Log in to GitHub Container Registry
+        uses: docker/login-action@v3
+        with:
+          registry: ghcr.io
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Extract metadata for Docker
+        id: meta
+        uses: docker/metadata-action@v5
+        with:
+          images: ghcr.io/mervinpraison/praisonai
+          tags: |
+            type=raw,value=latest,enable=${{ github.event_name == 'release' }}
+            type=semver,pattern={{version}},enable=${{ github.event_name == 'release' }}
+            type=semver,pattern={{major}}.{{minor}},enable=${{ github.event_name == 'release' }}
+            type=raw,value=${{ github.event.inputs.tag || 'latest' }},enable=${{ github.event_name == 'workflow_dispatch' }}
 
       - name: Build and push Docker image
-        if: steps.check_commit.outputs.should_run == 'true'
-        uses: docker/build-push-action@v2
+        uses: docker/build-push-action@v5
         with:
           context: .
           file: ./docker/Dockerfile
           push: true
-          tags: ghcr.io/mervinpraison/praisonai:latest
+          tags: ${{ steps.meta.outputs.tags }}
+          labels: ${{ steps.meta.outputs.labels }}
+          cache-from: type=gha
+          cache-to: type=gha,mode=max

.gitignore

@@ -77,4 +77,7 @@ agents/praisonaiagents/praisonaiagents.egg-info
 .qodo
 CopilotKit*
 .gemini
-issue_details.json
+issue_details.json
+
+# MCP Registry tokens
+.mcpregistry_*

docker/Dockerfile

@@ -14,9 +14,10 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
 RUN mkdir -p /root/.praison
 
 # Install Python packages (using latest versions)
+# Note: praisonai is installed from PyPI - remove version pin to use latest available
 RUN pip install --no-cache-dir \
     flask \
-    "praisonai>=2.3.26" \
+    praisonai \
     "praisonai[api]" \
     gunicorn \
     markdown

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>=2.3.26" \
+    "praisonai>=2.3.80" \
     "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>=2.3.26" \
+    "praisonai>=2.3.80" \
     "praisonai[ui]" \
     "praisonai[chat]" \
     "praisonai[realtime]" \

docker/Dockerfile.serve

@@ -0,0 +1,25 @@
+# Dockerfile for PraisonAI Serve API
+# Deploy to Railway, Render, Fly.io, or any container platform
+#
+# Build: docker build -f Dockerfile.serve -t praisonai-serve .
+# Run:   docker run -p 8005:8005 -e OPENAI_API_KEY=sk-xxx praisonai-serve
+
+FROM python:3.11-slim
+
+WORKDIR /app
+
+# Install dependencies
+RUN pip install --no-cache-dir praisonai praisonaiagents
+
+# Copy agents configuration
+COPY agents.yaml .
+
+# Expose the API port
+EXPOSE 8005
+
+# Health check
+HEALTHCHECK --interval=30s --timeout=10s --start-period=5s --retries=3 \
+    CMD curl -f http://localhost:8005/health || exit 1
+
+# Start the API server
+CMD ["praisonai", "serve", "agents.yaml", "--host", "0.0.0.0", "--port", "8005"]

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>=2.3.26" \
+    "praisonai>=2.3.80" \
     "praisonai[ui]" \
     "praisonai[crewai]"
 

examples/README.md

@@ -0,0 +1,49 @@
+# PraisonAI Examples
+
+This folder contains examples for PraisonAI. For detailed documentation, visit [docs.praison.ai](https://docs.praison.ai).
+
+## Structure
+
+```
+examples/
+├── python/           # Python examples
+│   ├── agents/       # Agent examples (single, multi, router, etc.)
+│   ├── workflows/    # Workflow patterns (routing, parallel, loop)
+│   ├── tools/        # Custom tools examples
+│   ├── mcp/          # MCP protocol examples
+│   ├── memory/       # Memory and sessions
+│   ├── code/         # Code editing and external CLI tools
+│   └── ...
+├── yaml/             # YAML workflow examples
+└── cookbooks/        # Complete use-case examples
+```
+
+## Quick Links
+
+| Category | Examples | Docs |
+|----------|----------|------|
+| **Agents** | [python/agents/](python/agents/) | [📖](https://docs.praison.ai/concepts/agents) |
+| **Workflows** | [python/workflows/](python/workflows/) | [📖](https://docs.praison.ai/features/workflows) |
+| **Model Router** | [python/agents/router-agent-cost-optimization.py](python/agents/router-agent-cost-optimization.py) | [📖](https://docs.praison.ai/features/model-router) |
+| **MCP** | [python/mcp/](python/mcp/) | [📖](https://docs.praison.ai/mcp) |
+| **Memory** | [python/memory/](python/memory/) | [📖](https://docs.praison.ai/concepts/memory) |
+| **Tools** | [python/tools/](python/tools/) | [📖](https://docs.praison.ai/tools) |
+| **Code** | [python/code/](python/code/) | [📖](https://docs.praison.ai/code) |
+| **YAML** | [yaml/](yaml/) | [📖](https://docs.praison.ai/features/yaml-workflows) |
+
+## Running Examples
+
+```bash
+# Install PraisonAI
+pip install praisonai
+
+# Set API key
+export OPENAI_API_KEY=your_key_here
+
+# Run an example
+python examples/python/agents/single-agent.py
+```
+
+## CLI Commands
+
+See the main [README.md](../README.md#-cli--no-code-interface) for all CLI commands.

examples/compare/cli_compare_advanced.py

@@ -0,0 +1,85 @@
+"""
+CLI Compare Example - Advanced Usage
+
+This example demonstrates advanced comparison features including
+custom mode configurations and result analysis.
+"""
+
+from praisonai.cli.features.compare import (
+    CompareHandler,
+    CompareResult,
+    ModeResult,
+    get_mode_config,
+    save_compare_result,
+    format_comparison_table,
+)
+
+# Create custom mode configurations
+custom_modes = {
+    "fast": {"llm": "gpt-4o-mini"},
+    "accurate": {"llm": "gpt-4o", "planning": True},
+    "research": {"research": True, "web_search": True},
+}
+
+# Create handler with verbose output
+handler = CompareHandler(verbose=True)
+
+# Example: Manually create comparison results for analysis
+# (useful for testing without API calls)
+mock_result = CompareResult(
+    query="What are the latest AI trends?",
+    comparisons=[
+        ModeResult(
+            mode="basic",
+            output="AI trends include LLMs, multimodal models, and agents.",
+            execution_time_ms=1234.5,
+            model_used="gpt-4o-mini",
+            tools_used=[]
+        ),
+        ModeResult(
+            mode="tools",
+            output="Based on recent search results, AI trends in 2024 include...",
+            execution_time_ms=2567.8,
+            model_used="gpt-4o-mini",
+            tools_used=["internet_search"]
+        ),
+        ModeResult(
+            mode="research",
+            output="Comprehensive research shows that AI trends...",
+            execution_time_ms=5432.1,
+            model_used="gpt-4o-mini",
+            tools_used=["internet_search", "web_scraper"]
+        ),
+    ]
+)
+
+# Get summary statistics
+summary = mock_result.get_summary()
+print("Comparison Summary:")
+print(f"  Fastest mode: {summary['fastest']} ({summary['fastest_time_ms']:.1f}ms)")
+print(f"  Slowest mode: {summary['slowest']} ({summary['slowest_time_ms']:.1f}ms)")
+
+# Format as table
+print("\n" + format_comparison_table(mock_result))
+
+# Convert to JSON for storage
+print("\nJSON Output:")
+print(mock_result.to_json())
+
+# Save to file
+# save_compare_result(mock_result, "advanced_comparison.json")
+
+print("\n--- CLI Usage Examples ---")
+print("""
+# Basic comparison
+praisonai "What is AI?" --compare "basic,tools"
+
+# Compare with specific model
+praisonai "Explain quantum computing" --compare "basic,planning,research" --model gpt-4o
+
+# Save results to file
+praisonai "Latest tech news" --compare "basic,tools,web_search" --compare-output results.json
+
+# Verbose output
+praisonai "Write a poem" --compare "basic,planning" --verbose
+""")

examples/compare/cli_compare_basic.py

@@ -0,0 +1,45 @@
+"""
+CLI Compare Example - Basic Usage
+
+This example demonstrates how to compare different CLI modes programmatically.
+"""
+
+from praisonai.cli.features.compare import (
+    CompareHandler,
+    get_mode_config,
+    list_available_modes,
+    parse_modes,
+)
+
+# List all available comparison modes
+print("Available modes:")
+for mode in list_available_modes():
+    config = get_mode_config(mode)
+    print(f"  - {mode}: {config}")
+
+# Parse modes from comma-separated string
+modes = parse_modes("basic,tools,planning")
+print(f"\nParsed modes: {modes}")
+
+# Create compare handler
+handler = CompareHandler(verbose=True)
+
+# Compare modes (this will run actual agent calls)
+# Uncomment to run:
+# result = handler.compare(
+#     query="What is artificial intelligence?",
+#     modes=["basic", "tools"],
+#     model="gpt-4o-mini"
+# )
+# handler.print_result(result)
+
+# Save results to file
+# result = handler.execute(
+#     query="Explain machine learning",
+#     modes_str="basic,planning",
+#     output_path="comparison_results.json"
+# )
+
+print("\nTo run comparison via CLI:")
+print('  praisonai "What is AI?" --compare "basic,tools,planning"')
+print('  praisonai "What is AI?" --compare "basic,research" --compare-output results.json')

examples/cookbooks/yaml/aws_bedrock_agents.yaml

@@ -1,10 +1,10 @@
 framework: crewai
 topic: "Technology Research and Content Creation using AWS Bedrock"
 
-roles:
+agents:  # Canonical: use 'agents' instead of 'roles'
   research_specialist:
     role: "Research Specialist"
-    backstory: |
+    instructions:  # Canonical |
       You are an experienced technology researcher with deep expertise in analyzing 
       emerging technologies and market trends. You excel at gathering comprehensive 
       information from various sources and synthesizing it into actionable insights.
@@ -35,7 +35,7 @@ roles:
 
   content_creator:
     role: "Technical Content Creator"
-    backstory: |
+    instructions:  # Canonical |
       You are a skilled technical writer and content creator who specializes in 
       translating complex technology concepts into engaging, accessible content 
       for various audiences including developers, business leaders, and end users.
@@ -64,7 +64,7 @@ roles:
 
   quality_reviewer:
     role: "Content Quality Reviewer"
-    backstory: |
+    instructions:  # Canonical |
       You are a meticulous editor and quality assurance specialist with extensive 
       experience in technical content review. You ensure accuracy, clarity, and 
       consistency while maintaining the intended tone and message.

examples/cookbooks/yaml/data_collector_scheduled.yaml

@@ -0,0 +1,23 @@
+# Example: Data Collection Agent with 30-minute Schedule
+# Collects and analyzes data every 30 minutes
+
+framework: praisonai
+
+agents:
+  - name: "Data Collector"
+    role: "Data Analysis Specialist"
+    goal: "Collect and analyze market data at regular intervals"
+    instructions: |
+      You are a data analysis specialist focused on collecting and analyzing
+      market trends and metrics. Provide actionable insights from the data.
+    tools:
+      - search_tool
+    verbose: false
+
+task: "Collect latest market data and identify key trends"
+
+# Schedule: Every 30 minutes with immediate execution
+schedule:
+  interval: "*/30m"
+  max_retries: 5
+  run_immediately: true

examples/cookbooks/yaml/news_monitor_scheduled.yaml

@@ -0,0 +1,53 @@
+# Example: 24/7 AI News Monitor with Scheduler
+# This agent runs continuously at regular intervals to monitor AI news
+
+framework: praisonai
+
+agents:
+  - name: "AI News Monitor"
+    role: "Technology News Analyst"
+    goal: "Monitor and summarize the latest AI developments"
+    instructions: |
+      You are an expert technology news analyst specializing in AI and machine learning.
+      Your job is to search for the latest AI news, identify the most important developments,
+      and provide clear, concise summaries with sources.
+      
+      Focus on:
+      - Major AI product launches
+      - Research breakthroughs
+      - Industry trends
+      - Policy and regulation updates
+    tools:
+      - search_tool
+    verbose: true
+
+# Task to execute on each scheduled run
+task: |
+  Search for the latest AI news from the past 24 hours.
+  Provide the top 3 most important stories with:
+  1. Headline
+  2. Brief summary (2-3 sentences)
+  3. Source URL
+  4. Why it's significant
+
+# Schedule configuration
+schedule:
+  # Interval options:
+  # - "hourly" - Every hour
+  # - "daily" - Every 24 hours
+  # - "*/30m" - Every 30 minutes
+  # - "*/6h" - Every 6 hours
+  # - "3600" - Custom seconds
+  interval: "hourly"
+  
+  # Maximum retry attempts if execution fails
+  max_retries: 3
+  
+  # Run immediately when scheduler starts (before first scheduled interval)
+  run_immediately: true
+
+# Optional: Variables for customization
+variables:
+  news_category: "AI and Machine Learning"
+  region: "Global"
+  time_range: "24 hours"

examples/cookbooks/yaml/secondary_market_research_agents.yaml

@@ -8,10 +8,10 @@ variables:
   industry: "Electric Vehicles"
   research_focus: "comprehensive market analysis"
 
-roles:
+agents:  # Canonical: use 'agents' instead of 'roles'
   market_overview_specialist:
     role: "Market Overview Specialist" 
-    backstory: |
+    instructions:  # Canonical |
       You are a seasoned market research analyst with 10+ years of experience 
       in analyzing market dynamics, trends, and growth patterns across various 
       industries. You excel at synthesizing complex market data into clear, 
@@ -64,7 +64,7 @@ roles:
 
   competitive_intelligence_analyst:
     role: "Competitive Intelligence Analyst"
-    backstory: |
+    instructions:  # Canonical |
       You are a competitive intelligence expert specializing in analyzing 
       competitive landscapes, market positioning, and strategic intelligence. 
       You have deep expertise in competitor analysis, market share assessment, 
@@ -119,7 +119,7 @@ roles:
 
   financial_performance_analyst:
     role: "Financial Performance Analyst"
-    backstory: |
+    instructions:  # Canonical |
       You are a financial analyst with expertise in corporate finance, 
       financial statement analysis, and industry benchmarking. You specialize 
       in analyzing financial performance metrics, trends, and comparative 
@@ -173,7 +173,7 @@ roles:
 
   growth_opportunities_researcher:
     role: "Growth Opportunities Researcher"
-    backstory: |
+    instructions:  # Canonical |
       You are a strategic business analyst specializing in identifying growth 
       opportunities, market expansion strategies, and emerging business models. 
       You excel at spotting trends, gaps, and untapped potential in markets 
@@ -228,7 +228,7 @@ roles:
 
   risk_assessment_specialist:
     role: "Risk Assessment Specialist"
-    backstory: |
+    instructions:  # Canonical |
       You are a risk management expert with extensive experience in identifying, 
       analyzing, and assessing business risks across various industries. You 
       specialize in market risks, competitive threats, regulatory challenges, 
@@ -287,7 +287,7 @@ roles:
 
   research_report_synthesizer:
     role: "Research Report Synthesizer"
-    backstory: |
+    instructions:  # Canonical |
       You are an expert business report writer and analyst with extensive 
       experience in synthesizing complex research findings into clear, 
       actionable business intelligence reports. You excel at creating 

examples/eval/accuracy_example.py

@@ -0,0 +1,69 @@
+"""
+Accuracy Evaluation Example
+
+This example demonstrates how to evaluate agent output accuracy
+by comparing against expected outputs using LLM-as-judge.
+"""
+
+import os
+from praisonaiagents import Agent
+from praisonaiagents.eval import AccuracyEvaluator
+
+# Check if we have an API key
+has_api_key = os.getenv("OPENAI_API_KEY") is not None
+
+if has_api_key:
+    # Create a simple math tutor agent
+    agent = Agent(
+        instructions="You are a math tutor. Answer math questions concisely with just the number."
+    )
+
+    # Create accuracy evaluator
+    evaluator = AccuracyEvaluator(
+        agent=agent,
+        input_text="What is 2 + 2?",
+        expected_output="4",
+        num_iterations=3,  # Run 3 times for statistical significance
+        verbose=True
+    )
+
+    # Run evaluation
+    result = evaluator.run(print_summary=True)
+
+    # Check results
+    print(f"\nAverage Score: {result.avg_score}/10")
+    print(f"Passed: {result.passed}")
+else:
+    print("⚠️  No OPENAI_API_KEY found. Using mock evaluation...")
+    
+    # Create evaluator without agent for mock testing
+    evaluator = AccuracyEvaluator(
+        func=lambda x: "4",  # Mock function
+        input_text="What is 2 + 2?",
+        expected_output="4",
+        num_iterations=1,
+        verbose=True
+    )
+    
+    # Test with mock output
+    try:
+        result = evaluator.run(print_summary=True)
+        print(f"\nMock Average Score: {result.avg_score}/10")
+        print(f"Mock Passed: {result.passed}")
+    except Exception as e:
+        print(f"Mock evaluation failed: {e}")
+        print("This is expected without an API key")
+
+# You can also evaluate pre-generated outputs
+print("\n--- Testing pre-generated output evaluation ---")
+output = "The answer is 4"
+
+if has_api_key:
+    try:
+        result2 = evaluator.evaluate_output(output, print_summary=True)
+        print(f"Pre-generated output score: {result2.avg_score}/10")
+    except Exception as e:
+        print(f"Pre-generated evaluation failed: {e}")
+else:
+    print("⚠️  Skipping pre-generated evaluation (no API key)")
+    print("To run full evaluation, set OPENAI_API_KEY environment variable")

examples/eval/batch_example.py

@@ -0,0 +1,105 @@
+"""
+Batch Evaluation Example
+
+This example demonstrates how to run batch evaluations
+from a JSON test file.
+"""
+
+import os
+import json
+import tempfile
+from praisonaiagents.eval import AccuracyEvaluator
+
+# Check if we have an API key
+has_api_key = os.getenv("OPENAI_API_KEY") is not None
+
+# Create a test file with multiple test cases
+test_cases = [
+    {
+        "input": "What is 2 + 2?",
+        "expected": "4"
+    },
+    {
+        "input": "What is the capital of France?",
+        "expected": "Paris"
+    },
+    {
+        "input": "What color is the sky?",
+        "expected": "Blue"
+    }
+]
+
+# Save test cases to a temporary file
+with tempfile.NamedTemporaryFile(mode='w', suffix='.json', delete=False) as f:
+    json.dump(test_cases, f)
+    test_file = f.name
+
+print(f"Test file created: {test_file}")
+print(f"Running {len(test_cases)} test cases...")
+
+if has_api_key:
+    # You can use the CLI for batch evaluation:
+    # praisonai eval batch --agent agents.yaml --test-file tests.json --batch-type accuracy
+
+    # Or programmatically:
+    from praisonaiagents import Agent
+
+    agent = Agent(
+        instructions="You are a helpful assistant. Answer questions concisely."
+    )
+
+    results = []
+    for i, test_case in enumerate(test_cases):
+        print(f"\nTest {i + 1}: {test_case['input']}")
+        
+        evaluator = AccuracyEvaluator(
+            agent=agent,
+            input_text=test_case["input"],
+            expected_output=test_case["expected"],
+            num_iterations=1
+        )
+        
+        result = evaluator.run()
+        results.append({
+            "input": test_case["input"],
+            "expected": test_case["expected"],
+            "score": result.avg_score,
+            "passed": result.passed
+        })
+        
+        print(f"  Score: {result.avg_score}/10")
+        print(f"  Passed: {result.passed}")
+
+    # Summary
+    passed = sum(1 for r in results if r["passed"])
+    print(f"\n{'='*50}")
+    print(f"Summary: {passed}/{len(results)} tests passed")
+    print(f"Pass Rate: {passed/len(results):.1%}")
+else:
+    print("⚠️  No OPENAI_API_KEY found. Running mock batch evaluation...")
+    
+    # Mock evaluation without API key
+    results = []
+    for i, test_case in enumerate(test_cases):
+        print(f"\nTest {i + 1}: {test_case['input']}")
+        
+        # Mock result (all would fail without API key)
+        mock_score = 0.0
+        mock_passed = False
+        
+        results.append({
+            "input": test_case["input"],
+            "expected": test_case["expected"],
+            "score": mock_score,
+            "passed": mock_passed
+        })
+        
+        print(f"  Score: {mock_score}/10 (mock)")
+        print(f"  Passed: {mock_passed} (expected without API key)")
+
+    # Summary
+    passed = sum(1 for r in results if r["passed"])
+    print(f"\n{'='*50}")
+    print(f"Summary: {passed}/{len(results)} tests passed (mock evaluation)")
+    print(f"Pass Rate: {passed/len(results):.1%}")
+    print("\nTo run real batch evaluation, set OPENAI_API_KEY environment variable")

examples/eval/criteria_example.py

@@ -0,0 +1,82 @@
+"""
+Criteria Evaluation Example
+
+This example demonstrates how to evaluate agent outputs
+against custom criteria using LLM-as-judge.
+"""
+
+import os
+from praisonaiagents import Agent
+from praisonaiagents.eval import CriteriaEvaluator
+
+# Check if we have an API key
+has_api_key = os.getenv("OPENAI_API_KEY") is not None
+
+if has_api_key:
+    print("--- Testing Criteria Evaluation with Agent ---")
+    # Create a customer service agent
+    agent = Agent(
+        instructions="You are a friendly customer service agent. Be helpful and empathetic."
+    )
+
+    # Create criteria evaluator with numeric scoring
+    evaluator = CriteriaEvaluator(
+        criteria="Response is helpful, empathetic, and provides a clear solution",
+        agent=agent,
+        input_text="I'm frustrated because my order hasn't arrived yet.",
+        scoring_type="numeric",  # Score 1-10
+        threshold=7.0,           # Pass if score >= 7
+        num_iterations=2,
+        verbose=True
+    )
+
+    # Run evaluation
+    result = evaluator.run(print_summary=True)
+
+    print("\nNumeric Scoring Results:")
+    print(f"  Average Score: {result.avg_score}/10")
+    print(f"  Pass Rate: {result.pass_rate:.1%}")
+    print(f"  Passed: {result.passed}")
+
+    # Binary scoring example (pass/fail)
+    print("\n--- Testing Binary Scoring ---")
+    binary_evaluator = CriteriaEvaluator(
+        criteria="Response does not contain any offensive language",
+        agent=agent,
+        input_text="Tell me a joke",
+        scoring_type="binary",  # Pass or Fail
+        verbose=True
+    )
+
+    binary_result = binary_evaluator.run(print_summary=True)
+    
+    # With on_fail callback
+    print("\n--- Testing Failure Callback ---")
+    def handle_failure(score):
+        print(f"ALERT: Evaluation failed with score {score.score}")
+        print(f"Reasoning: {score.reasoning}")
+
+    callback_evaluator = CriteriaEvaluator(
+        criteria="Response is professional and helpful",
+        agent=agent,
+        input_text="Help me",
+        on_fail=handle_failure,
+        threshold=8.0
+    )
+
+    callback_evaluator.run()
+else:
+    print("⚠️  No OPENAI_API_KEY found. Skipping agent-based criteria evaluation...")
+
+# Test pre-generated output evaluation (doesn't need agent)
+print("\n--- Testing Pre-generated Output Evaluation ---")
+if has_api_key:
+    output = "I understand your frustration. Let me check on your order right away."
+    try:
+        result2 = evaluator.evaluate_output(output, print_summary=True)
+        print(f"Pre-generated output score: {result2.avg_score}/10")
+    except Exception as e:
+        print(f"Pre-generated evaluation failed: {e}")
+else:
+    print("⚠️  Skipping pre-generated evaluation (no API key)")
+    print("To run full evaluation, set OPENAI_API_KEY environment variable")

examples/eval/performance_example.py

@@ -0,0 +1,59 @@
+"""
+Performance Evaluation Example
+
+This example demonstrates how to benchmark agent performance
+by measuring runtime and memory usage across multiple iterations.
+"""
+
+import os
+from praisonaiagents import Agent
+from praisonaiagents.eval import PerformanceEvaluator
+
+# Check if we have an API key
+has_api_key = os.getenv("OPENAI_API_KEY") is not None
+
+if has_api_key:
+    print("--- Testing Agent Performance ---")
+    # Create a simple agent
+    agent = Agent(
+        instructions="You are a helpful assistant. Answer questions briefly."
+    )
+
+    # Create performance evaluator
+    evaluator = PerformanceEvaluator(
+        agent=agent,
+        input_text="What is the capital of France?",
+        num_iterations=10,  # Run 10 benchmark iterations
+        warmup_runs=2,      # 2 warmup runs before measurement
+        track_memory=True,  # Track memory usage
+        verbose=True
+    )
+
+    # Run evaluation
+    result = evaluator.run(print_summary=True)
+
+    # Access detailed metrics
+    print("\nAgent Performance Results:")
+    print(f"  Average Time: {result.avg_run_time:.4f}s")
+    print(f"  Min Time: {result.min_run_time:.4f}s")
+    print(f"  Max Time: {result.max_run_time:.4f}s")
+    print(f"  Median Time: {result.median_run_time:.4f}s")
+    print(f"  P95 Time: {result.p95_run_time:.4f}s")
+    print(f"  Avg Memory: {result.avg_memory:.2f} MB")
+else:
+    print("⚠️  No OPENAI_API_KEY found. Skipping agent performance test...")
+
+# You can also benchmark any function
+def my_function():
+    import time
+    time.sleep(0.1)
+    return "done"
+
+func_evaluator = PerformanceEvaluator(
+    func=my_function,
+    num_iterations=5,
+    warmup_runs=1,
+    track_memory=False
+)
+
+func_result = func_evaluator.run(print_summary=True)

examples/eval/reliability_example.py

@@ -0,0 +1,72 @@
+"""
+Reliability Evaluation Example
+
+This example demonstrates how to verify that agents call
+the expected tools during execution.
+"""
+
+import os
+from praisonaiagents import Agent
+from praisonaiagents.eval import ReliabilityEvaluator
+
+
+def search_web(query: str) -> str:
+    """Search the web for information."""
+    return f"Search results for: {query}"
+
+
+def calculate(expression: str) -> str:
+    """Calculate a math expression."""
+    return str(eval(expression))
+
+
+# Check if we have an API key
+has_api_key = os.getenv("OPENAI_API_KEY") is not None
+
+if has_api_key:
+    print("--- Testing Agent Tool Call Reliability ---")
+    # Create an agent with tools
+    agent = Agent(
+        instructions="You are a helpful assistant with access to search and calculator tools.",
+        tools=[search_web, calculate]
+    )
+
+    # Create reliability evaluator
+    evaluator = ReliabilityEvaluator(
+        agent=agent,
+        input_text="Search for the weather in Paris and calculate 25 * 4",
+        expected_tools=["search_web", "calculate"],  # Tools that should be called
+        forbidden_tools=["delete_file"],  # Tools that should NOT be called
+        verbose=True
+    )
+
+    # Run evaluation
+    result = evaluator.run(print_summary=True)
+
+    # Check results
+    print("\nAgent Reliability Results:")
+    print(f"  Passed: {result.passed}")
+    print(f"  Pass Rate: {result.pass_rate:.1%}")
+else:
+    print("⚠️  No OPENAI_API_KEY found. Skipping agent reliability test...")
+    print("Agent would fail to call tools without API key (expected behavior)")
+
+# You can also evaluate pre-recorded tool calls (doesn't need API key)
+print("\n--- Testing Pre-recorded Tool Call Evaluation ---")
+# Create a mock agent for tool call evaluation
+mock_agent = Agent(instructions="Mock agent")
+evaluator = ReliabilityEvaluator(
+    agent=mock_agent,
+    expected_tools=["search_web", "calculate"],  # Tools that should be called
+    forbidden_tools=["delete_file"],  # Tools that should NOT be called
+    verbose=True
+)
+
+result2 = evaluator.evaluate_tool_calls(
+    actual_tools=["search_web", "calculate"],
+    print_summary=True
+)
+
+print("\nPre-recorded Tool Call Results:")
+print(f"  Passed: {result2.passed}")
+print(f"  Pass Rate: {result2.pass_rate:.1%}")

examples/python/a2a/a2a-server.py

@@ -0,0 +1,53 @@
+"""
+PraisonAI A2A Server Example
+
+Expose a PraisonAI Agent as an A2A (Agent2Agent) Server.
+This enables other AI agents to discover and communicate with your agent.
+
+Run:
+    uvicorn a2a-server:app --reload
+
+Endpoints:
+    GET /.well-known/agent.json  - Agent Card for discovery
+    GET /status                   - Server status
+"""
+
+from praisonaiagents import Agent, A2A
+from fastapi import FastAPI
+
+# Create an agent with tools
+def search_web(query: str) -> str:
+    """Search the web for information."""
+    return f"Search results for: {query}"
+
+def calculate(expression: str) -> str:
+    """Calculate a mathematical expression."""
+    try:
+        return f"Result: {eval(expression)}"
+    except Exception:
+        return "Invalid expression"
+
+agent = Agent(
+    name="Research Assistant",
+    role="Research Analyst",
+    goal="Help users research topics and answer questions",
+    tools=[search_web, calculate]
+)
+
+# Expose as A2A Server
+a2a = A2A(
+    agent=agent,
+    url="http://localhost:8000/a2a",
+    version="1.0.0"
+)
+
+# Create FastAPI app
+app = FastAPI(
+    title="PraisonAI A2A Server",
+    description="A2A-compatible agent server"
+)
+app.include_router(a2a.get_router())
+
+if __name__ == "__main__":
+    import uvicorn
+    uvicorn.run(app, host="0.0.0.0", port=8000)

examples/python/agents/autoagents_workflow_patterns.py

@@ -0,0 +1,132 @@
+"""
+AutoAgents Workflow Patterns Example
+
+This example demonstrates the new workflow auto-generation features:
+- Dynamic agent count based on task complexity
+- Multiple workflow patterns (sequential, parallel, routing, orchestrator-workers, evaluator-optimizer)
+- Pattern recommendation based on task keywords
+
+Documentation: https://docs.praison.ai/features/autoagents
+CLI Reference: https://docs.praison.ai/nocode/auto
+"""
+
+from praisonai.auto import AutoGenerator, WorkflowAutoGenerator
+import os
+import tempfile
+
+# Ensure API key is set
+if not os.environ.get("OPENAI_API_KEY"):
+    print("Please set OPENAI_API_KEY environment variable")
+    exit(1)
+
+print("=" * 60)
+print("AutoAgents Workflow Patterns Demo")
+print("=" * 60)
+
+# =============================================================================
+# Example 1: Dynamic Agent Count (Simple vs Complex Tasks)
+# =============================================================================
+print("\n📊 Example 1: Dynamic Agent Count")
+print("-" * 40)
+
+# Simple task - should create 1-2 agents
+with tempfile.NamedTemporaryFile(mode='w', suffix='.yaml', delete=False) as f:
+    simple_file = f.name
+
+simple_generator = AutoGenerator(
+    topic="Write a haiku about spring",
+    agent_file=simple_file,
+    framework="praisonai"
+)
+
+print("Simple task: 'Write a haiku about spring'")
+result = simple_generator.generate()
+print(f"Generated: {result}")
+
+# Complex task - should create 3-4 agents
+with tempfile.NamedTemporaryFile(mode='w', suffix='.yaml', delete=False) as f:
+    complex_file = f.name
+
+complex_generator = AutoGenerator(
+    topic="Research AI trends, analyze market data, write a comprehensive report with executive summary",
+    agent_file=complex_file,
+    framework="praisonai"
+)
+
+print("\nComplex task: 'Research AI trends, analyze market data...'")
+result = complex_generator.generate()
+print(f"Generated: {result}")
+
+# =============================================================================
+# Example 2: Pattern Recommendation
+# =============================================================================
+print("\n🎯 Example 2: Pattern Recommendation")
+print("-" * 40)
+
+test_tasks = [
+    ("Write a blog post", "sequential"),
+    ("Research from multiple sources simultaneously", "parallel"),
+    ("Classify and route customer requests", "routing"),
+    ("Comprehensive analysis and break down the problem", "orchestrator-workers"),
+    ("Refine and improve content quality", "evaluator-optimizer"),
+]
+
+for task, expected in test_tasks:
+    generator = WorkflowAutoGenerator(topic=task)
+    recommended = generator.recommend_pattern()
+    status = "✅" if recommended == expected else "⚠️"
+    print(f"{status} Task: '{task[:40]}...'")
+    print(f"   Recommended: {recommended}")
+
+# =============================================================================
+# Example 3: Generate Workflow with Specific Pattern
+# =============================================================================
+print("\n🔄 Example 3: Generate Workflows with Patterns")
+print("-" * 40)
+
+patterns = ["sequential", "parallel", "orchestrator-workers", "evaluator-optimizer"]
+
+for pattern in patterns:
+    with tempfile.NamedTemporaryFile(mode='w', suffix='.yaml', delete=False) as f:
+        workflow_file = f.name
+    
+    generator = WorkflowAutoGenerator(
+        topic="Create a market analysis report",
+        workflow_file=workflow_file
+    )
+    
+    print(f"\n📋 Pattern: {pattern}")
+    result = generator.generate(pattern=pattern)
+    print(f"   Generated: {result}")
+    
+    # Read and show structure
+    import yaml
+    with open(result, 'r') as f:
+        data = yaml.safe_load(f)
+    print(f"   Agents: {list(data.get('agents', {}).keys())}")
+    print(f"   Steps: {len(data.get('steps', []))}")
+
+# =============================================================================
+# Example 4: CLI Usage (for reference)
+# =============================================================================
+print("\n💻 Example 4: CLI Commands (for reference)")
+print("-" * 40)
+print("""
+# Auto-generate agents (dynamic count)
+praisonai --auto "Write a haiku about spring"
+
+# Auto-generate workflow with pattern
+praisonai workflow auto "Research and write a report" --pattern sequential
+praisonai workflow auto "Research from multiple sources" --pattern parallel
+praisonai workflow auto "Comprehensive market analysis" --pattern orchestrator-workers
+praisonai workflow auto "Refine content quality" --pattern evaluator-optimizer
+
+# With output file
+praisonai workflow auto "Customer support routing" --pattern routing --output support.yaml
+""")
+
+print("\n" + "=" * 60)
+print("✅ AutoAgents Workflow Patterns Demo Complete!")
+print("=" * 60)
+print("\nDocumentation: https://docs.praison.ai/features/autoagents")
+print("CLI Reference: https://docs.praison.ai/nocode/auto")

examples/python/cli/autonomy_modes_example.py

@@ -0,0 +1,38 @@
+"""
+Autonomy Modes Example for PraisonAI CLI.
+
+Control AI autonomy: suggest, auto_edit, full_auto.
+Docs: https://docs.praison.ai/cli/autonomy-modes
+"""
+
+from praisonai.cli.features import AutonomyModeHandler
+from praisonai.cli.features.autonomy_mode import ActionRequest, ActionType
+
+# Initialize with suggest mode (safest)
+handler = AutonomyModeHandler()
+manager = handler.initialize(mode="suggest")
+
+print(f"Current mode: {handler.get_mode()}")
+print("Available modes: suggest, auto_edit, full_auto")
+
+# Create an action request
+action = ActionRequest(
+    action_type=ActionType.FILE_WRITE,
+    description="Edit src/main.py to add logging",
+    details={"file": "src/main.py"}
+)
+
+# In suggest mode, this would require approval
+# For demo, we'll just show the action
+print(f"\nAction: {action.description}")
+print(f"Type: {action.action_type.value}")
+print("Would require approval in 'suggest' mode")
+
+# Change to auto_edit mode
+handler.set_mode("auto_edit")
+print(f"\nChanged to: {handler.get_mode()}")
+print("File edits now auto-approved, shell commands still require approval")
+
+# Get statistics
+stats = handler.get_stats()
+print(f"\nStats: {stats}")

examples/python/cli/cost_tracking_example.py

@@ -0,0 +1,34 @@
+"""
+Cost Tracking Example for PraisonAI CLI.
+
+Real-time token usage and cost monitoring.
+Docs: https://docs.praison.ai/cli/cost-tracking
+"""
+
+from praisonai.cli.features import CostTrackerHandler
+
+# Initialize tracker
+handler = CostTrackerHandler()
+tracker = handler.initialize(session_id="demo-session")
+
+# Track some requests
+handler.track_request("gpt-4o", input_tokens=1000, output_tokens=500)
+handler.track_request("gpt-4o-mini", input_tokens=2000, output_tokens=800)
+handler.track_request("claude-3-5-sonnet", input_tokens=1500, output_tokens=600)
+
+# Get totals
+print(f"Total tokens: {handler.get_tokens():,}")
+print(f"Total cost: ${handler.get_cost():.4f}")
+
+# Get detailed summary
+summary = handler.get_summary()
+print("\n=== Session Summary ===")
+print(f"Session ID: {summary['session_id']}")
+print(f"Total requests: {summary['total_requests']}")
+print(f"Input tokens: {summary['total_input_tokens']:,}")
+print(f"Output tokens: {summary['total_output_tokens']:,}")
+print(f"Total cost: ${summary['total_cost']:.4f}")
+print(f"Avg cost/request: ${summary['avg_cost_per_request']:.4f}")
+
+# Display formatted (uses Rich if available)
+handler.display_summary()

examples/python/cli/git_integration_example.py

@@ -0,0 +1,45 @@
+"""
+Git Integration Example for PraisonAI CLI.
+
+Auto-commit with AI messages, diff viewing, undo.
+Docs: https://docs.praison.ai/cli/git-integration
+"""
+
+from praisonai.cli.features import GitIntegrationHandler
+
+# Initialize
+handler = GitIntegrationHandler(verbose=True)
+git = handler.initialize(repo_path=".")
+
+# Check if it's a git repo
+if not git.is_repo:
+    print("Not a git repository!")
+    exit(1)
+
+# Show status
+print("=== Git Status ===")
+status = handler.show_status()
+print(f"Branch: {status.branch}")
+print(f"Staged: {len(status.staged_files)}")
+print(f"Modified: {len(status.modified_files)}")
+print(f"Untracked: {len(status.untracked_files)}")
+
+# Show diff (if any changes)
+if status.has_changes:
+    print("\n=== Diff ===")
+    diff = handler.show_diff()
+    if diff:
+        print(diff[:300] + "..." if len(diff) > 300 else diff)
+
+# Show recent commits
+print("\n=== Recent Commits ===")
+commits = handler.show_log(count=5)
+for c in commits:
+    print(f"  {c.short_hash} {c.message[:50]}")
+
+# To commit with AI message:
+# commit = handler.commit()  # Auto-generates message
+# commit = handler.commit(message="Custom message")
+
+# To undo last commit:
+# handler.undo(soft=True)  # Keeps changes staged

examples/python/cli/interactive_tui_example.py

@@ -0,0 +1,49 @@
+"""
+Interactive TUI Example for PraisonAI CLI.
+
+Rich terminal interface with completions and history.
+Docs: https://docs.praison.ai/cli/interactive-tui
+"""
+
+from praisonai.cli.features import InteractiveTUIHandler
+from praisonai.cli.features.interactive_tui import InteractiveConfig
+
+# Custom configuration
+config = InteractiveConfig(
+    prompt="ai> ",
+    multiline=True,
+    enable_completions=True,
+    show_status_bar=True
+)
+
+# Define callbacks
+def on_input(text):
+    """Handle regular input."""
+    return f"You said: {text}"
+
+def on_command(cmd):
+    """Handle slash commands."""
+    if cmd == "/exit":
+        return {"type": "exit"}
+    return {"type": "command", "message": f"Executed: {cmd}"}
+
+# Initialize
+handler = InteractiveTUIHandler()
+session = handler.initialize(
+    config=config,
+    on_input=on_input,
+    on_command=on_command
+)
+
+# Add commands for completion
+session.add_commands(["help", "exit", "cost", "model", "plan"])
+
+# Add symbols from codebase
+session.add_symbols(["MyClass", "my_function"])
+
+print("Interactive TUI initialized!")
+print("Commands available: /help, /exit, /cost, /model, /plan")
+print("Run handler.run() to start interactive session")
+
+# Uncomment to run interactive session:
+# handler.run()

examples/python/cli/repo_map_example.py

@@ -0,0 +1,24 @@
+"""
+Repository Map Example for PraisonAI CLI.
+
+Intelligent codebase mapping with symbol extraction.
+Docs: https://docs.praison.ai/cli/repo-map
+"""
+
+from praisonai.cli.features import RepoMapHandler
+
+# Initialize with current directory
+handler = RepoMapHandler(verbose=True)
+repo_map = handler.initialize(root=".")
+
+# Get the repository map
+print("=== Repository Map ===")
+map_str = handler.get_map()
+print(map_str[:500] if len(map_str) > 500 else map_str)
+print("..." if len(map_str) > 500 else "")
+
+# Refresh after file changes
+handler.refresh()
+
+# Get symbols for a specific file (if exists)
+# symbols = handler.get_context("MyClass")

examples/python/cli/sandbox_execution_example.py

@@ -0,0 +1,42 @@
+"""
+Sandbox Execution Example for PraisonAI CLI.
+
+Secure isolated command execution (--sandbox flag).
+Docs: https://docs.praison.ai/cli/sandbox-execution
+"""
+
+from praisonai.cli.features import SandboxExecutorHandler
+from praisonai.cli.features.sandbox_executor import SandboxMode
+
+# Initialize with basic sandbox
+handler = SandboxExecutorHandler(verbose=True)
+sandbox = handler.initialize(mode="basic")
+
+print(f"Sandbox mode: {handler.get_mode()}")
+print(f"Sandbox enabled: {handler.is_enabled}")
+
+# Execute a safe command
+print("\n=== Safe Command ===")
+result = handler.execute("echo 'Hello from sandbox!'")
+print(f"Success: {result.success}")
+print(f"Output: {result.stdout.strip()}")
+print(f"Sandboxed: {result.was_sandboxed}")
+
+# Try a blocked command
+print("\n=== Blocked Command ===")
+result = handler.execute("rm -rf /")
+print(f"Success: {result.success}")
+print(f"Violations: {result.policy_violations}")
+
+# Validate command before execution
+print("\n=== Command Validation ===")
+violations = handler.validate_command("sudo apt install something")
+if violations:
+    print(f"Command blocked: {violations}")
+else:
+    print("Command allowed")
+
+# Available modes
+print("\n=== Available Modes ===")
+for mode in SandboxMode:
+    print(f"  {mode.value}")

examples/python/cli/slash_commands_example.py

@@ -0,0 +1,51 @@
+"""
+Slash Commands Example for PraisonAI CLI.
+
+Demonstrates interactive slash commands like /help, /cost, /model.
+Docs: https://docs.praison.ai/cli/slash-commands
+"""
+
+from praisonai.cli.features import SlashCommandHandler
+from praisonai.cli.features.slash_commands import SlashCommand, CommandKind
+
+# Create handler
+handler = SlashCommandHandler()
+
+# Execute commands
+print("=== /help ===")
+result = handler.execute("/help")
+print(result.get("message", ""))
+
+print("\n=== /cost ===")
+result = handler.execute("/cost")
+print(result.get("message", ""))
+
+print("\n=== /model ===")
+result = handler.execute("/model")
+print(result.get("message", ""))
+
+# Check if input is a command
+print("\n=== Command Detection ===")
+print(f"/help is command: {handler.is_command('/help')}")
+print(f"hello is command: {handler.is_command('hello')}")
+
+# Get completions for auto-complete
+print("\n=== Completions for '/he' ===")
+completions = handler.get_completions("/he")
+print(completions)
+
+# Register custom command
+def my_custom_handler(args, context):
+    return {"type": "custom", "message": f"Custom command with args: {args}"}
+
+custom_cmd = SlashCommand(
+    name="mycommand",
+    description="My custom command",
+    handler=my_custom_handler,
+    kind=CommandKind.ACTION,
+    aliases=["mc"]
+)
+
+handler.register(custom_cmd)
+result = handler.execute("/mycommand arg1 arg2")
+print(f"\n=== Custom Command ===\n{result.get('message', '')}")

examples/python/code/claude_code_example.py

@@ -0,0 +1,34 @@
+"""
+Claude Code CLI Integration Example
+
+Demonstrates how to use Claude Code CLI as an agent tool in PraisonAI.
+"""
+
+import asyncio
+from praisonai.integrations import ClaudeCodeIntegration
+
+async def main():
+    # Create Claude Code integration
+    claude = ClaudeCodeIntegration(
+        workspace=".",
+        output_format="json",
+        skip_permissions=True,
+        system_prompt="Be concise and follow best practices",
+        allowed_tools=["Read", "Write", "Bash"]
+    )
+    
+    # Check availability
+    print(f"Claude CLI available: {claude.is_available}")
+    print(f"Claude SDK available: {claude.sdk_available}")
+    
+    # Execute a coding task
+    result = await claude.execute("List all Python files in the current directory")
+    print(f"Result: {result}")
+    
+    # Stream output
+    print("\nStreaming output:")
+    async for event in claude.stream("Explain the project structure"):
+        print(event)
+
+if __name__ == "__main__":
+    asyncio.run(main())

examples/python/code/code_editing_example.py

@@ -0,0 +1,60 @@
+"""
+Code Editing Tools Example
+
+Demonstrates how to use PraisonAI's code editing tools with SEARCH/REPLACE diffs.
+"""
+
+from praisonai.code import (
+    set_workspace,
+    code_read_file,
+    code_write_file,
+    code_apply_diff,
+    code_list_files,
+    code_execute_command,
+    CODE_TOOLS
+)
+
+# Set workspace
+set_workspace(".")
+
+# List files
+print("Files in current directory:")
+files = code_list_files(".", recursive=False)
+for f in files:
+    print(f"  {f}")
+
+# Read a file
+print("\nReading example file:")
+try:
+    content = code_read_file("README.md", start_line=1, end_line=10)
+    print(content)
+except Exception as e:
+    print(f"  File not found: {e}")
+
+# Write a file
+print("\nWriting test file:")
+code_write_file("test_output.txt", "Hello from PraisonAI Code Tools!")
+print("  Created test_output.txt")
+
+# Apply a diff
+print("\nApplying diff:")
+diff = """
+<<<<<<< SEARCH
+Hello from PraisonAI Code Tools!
+=======
+Hello from PraisonAI Code Tools!
+This line was added by a diff.
+>>>>>>> REPLACE
+"""
+result = code_apply_diff("test_output.txt", diff)
+print(f"  {result}")
+
+# Execute command
+print("\nExecuting command:")
+output = code_execute_command("cat test_output.txt")
+print(f"  {output}")
+
+# Cleanup
+import os
+os.remove("test_output.txt")
+print("\nCleanup complete.")

examples/python/code/codex_cli_example.py

@@ -0,0 +1,36 @@
+"""
+Codex CLI Integration Example
+
+Demonstrates how to use OpenAI Codex CLI as an agent tool in PraisonAI.
+"""
+
+import asyncio
+from praisonai.integrations import CodexCLIIntegration
+
+async def main():
+    # Create Codex CLI integration
+    codex = CodexCLIIntegration(
+        workspace=".",
+        full_auto=True,
+        sandbox="danger-full-access",
+        json_output=True
+    )
+    
+    # Check availability
+    print(f"Codex CLI available: {codex.is_available}")
+    print(f"Full Auto: {codex.full_auto}")
+    print(f"Sandbox: {codex.sandbox}")
+    
+    # Execute a coding task
+    result = await codex.execute("List files in the current directory")
+    print(f"Result: {result}")
+    
+    # Stream JSON events
+    print("\nStreaming events:")
+    async for event in codex.stream("Explain the project"):
+        event_type = event.get("type", "")
+        if event_type == "item.completed":
+            print(f"  {event}")
+
+if __name__ == "__main__":
+    asyncio.run(main())

examples/python/code/cursor_cli_example.py

@@ -0,0 +1,37 @@
+"""
+Cursor CLI Integration Example
+
+Demonstrates how to use Cursor CLI as an agent tool in PraisonAI.
+"""
+
+import asyncio
+from praisonai.integrations import CursorCLIIntegration
+
+async def main():
+    # Create Cursor CLI integration
+    cursor = CursorCLIIntegration(
+        workspace=".",
+        output_format="json",
+        force=True,
+        model="gpt-5",
+        stream_partial=True
+    )
+    
+    # Check availability
+    print(f"Cursor CLI available: {cursor.is_available}")
+    print(f"Force mode: {cursor.force}")
+    print(f"Model: {cursor.model}")
+    
+    # Execute a coding task
+    result = await cursor.execute("List files in the current directory")
+    print(f"Result: {result}")
+    
+    # Stream output
+    print("\nStreaming output:")
+    async for event in cursor.stream("Explain the project"):
+        content = event.get("content", "")
+        if content:
+            print(content, end="", flush=True)
+
+if __name__ == "__main__":
+    asyncio.run(main())

examples/python/code/external_agents_example.py

@@ -0,0 +1,58 @@
+"""
+External Agents Integration Example
+
+Demonstrates how to use all external CLI integrations as agent tools.
+"""
+
+import asyncio
+from praisonai.integrations import (
+    ClaudeCodeIntegration,
+    GeminiCLIIntegration,
+    CodexCLIIntegration,
+    CursorCLIIntegration,
+    get_available_integrations
+)
+
+async def main():
+    # Check availability of all integrations
+    print("=== External CLI Integrations ===\n")
+    
+    availability = get_available_integrations()
+    print("Availability:")
+    for name, avail in availability.items():
+        status = "✅" if avail else "❌"
+        print(f"  {status} {name}")
+    
+    # Create integrations
+    print("\n--- Claude Code ---")
+    claude = ClaudeCodeIntegration(workspace=".", skip_permissions=True)
+    print(f"CLI: {claude.cli_command}")
+    print(f"Available: {claude.is_available}")
+    tool = claude.as_tool()
+    print(f"Tool: {tool.__name__}")
+    
+    print("\n--- Gemini CLI ---")
+    gemini = GeminiCLIIntegration(workspace=".", model="gemini-2.5-flash")
+    print(f"CLI: {gemini.cli_command}")
+    print(f"Available: {gemini.is_available}")
+    tool = gemini.as_tool()
+    print(f"Tool: {tool.__name__}")
+    
+    print("\n--- Codex CLI ---")
+    codex = CodexCLIIntegration(workspace=".", full_auto=True)
+    print(f"CLI: {codex.cli_command}")
+    print(f"Available: {codex.is_available}")
+    tool = codex.as_tool()
+    print(f"Tool: {tool.__name__}")
+    
+    print("\n--- Cursor CLI ---")
+    cursor = CursorCLIIntegration(workspace=".", force=True)
+    print(f"CLI: {cursor.cli_command}")
+    print(f"Available: {cursor.is_available}")
+    tool = cursor.as_tool()
+    print(f"Tool: {tool.__name__}")
+    
+    print("\n=== All integrations ready ===")
+
+if __name__ == "__main__":
+    asyncio.run(main())

examples/python/code/gemini_cli_example.py

@@ -0,0 +1,33 @@
+"""
+Gemini CLI Integration Example
+
+Demonstrates how to use Gemini CLI as an agent tool in PraisonAI.
+"""
+
+import asyncio
+from praisonai.integrations import GeminiCLIIntegration
+
+async def main():
+    # Create Gemini CLI integration
+    gemini = GeminiCLIIntegration(
+        workspace=".",
+        model="gemini-2.5-flash",
+        output_format="json",
+        include_directories=["../lib", "../docs"]
+    )
+    
+    # Check availability
+    print(f"Gemini CLI available: {gemini.is_available}")
+    print(f"Model: {gemini.model}")
+    
+    # Execute a coding task
+    result = await gemini.execute("Analyze the code structure")
+    print(f"Result: {result}")
+    
+    # Execute with stats
+    result, stats = await gemini.execute_with_stats("Explain main.py")
+    print(f"Result: {result}")
+    print(f"Stats: {stats}")
+
+if __name__ == "__main__":
+    asyncio.run(main())

examples/python/general/auto_agents_example.py

@@ -1,11 +1,48 @@
+"""
+AutoAgents Example
+
+AutoAgents automatically creates and manages AI agents based on high-level instructions.
+It features dynamic agent count based on task complexity.
+
+Documentation: https://docs.praison.ai/features/autoagents
+CLI Reference: https://docs.praison.ai/nocode/auto
+
+New Features:
+- Dynamic agent count (1-4 based on task complexity)
+- Workflow patterns: sequential, parallel, routing, orchestrator-workers, evaluator-optimizer
+- Pattern recommendation based on task keywords
+- Tool preservation from LLM suggestions
+"""
+
 from praisonaiagents import AutoAgents
 from praisonaiagents.tools import duckduckgo
 
+# Basic usage - AutoAgents analyzes complexity and creates optimal agents
 agents = AutoAgents(
     instructions="Search for information about AI Agents",
     tools=[duckduckgo],
-    process="sequential",
-    verbose=True
+    process="sequential",  # or "hierarchical"
+    verbose=True,
+    max_agents=3  # Maximum number of agents to create
 )
 
-agents.start()
+result = agents.start()
+print(result)
+
+# =============================================================================
+# CLI Usage Examples (for reference)
+# =============================================================================
+# 
+# # Auto-generate agents (dynamic count based on complexity)
+# praisonai --auto "Write a haiku about spring"
+# praisonai --auto "Research AI trends, analyze data, write report"
+#
+# # Auto-generate workflow with specific pattern
+# praisonai workflow auto "Research and write" --pattern sequential
+# praisonai workflow auto "Research from multiple sources" --pattern parallel
+# praisonai workflow auto "Comprehensive analysis" --pattern orchestrator-workers
+# praisonai workflow auto "Refine content quality" --pattern evaluator-optimizer
+#
+# # With framework selection
+# praisonai --framework crewai --auto "Create a movie script"
+# praisonai --framework autogen --auto "Create a movie script"

examples/python/mcp/multiple-mcp-servers.py

@@ -0,0 +1,108 @@
+"""
+Multiple MCP Servers Example
+
+This example demonstrates how to use multiple MCP servers with a single agent.
+You can combine different MCP servers (time, memory, filesystem, etc.) along
+with regular Python functions as tools.
+
+Requirements:
+- pip install praisonaiagents
+- uvx (for mcp-server-time)
+- npx (for @modelcontextprotocol servers)
+"""
+
+from praisonaiagents import Agent
+from praisonaiagents.mcp import MCP
+
+
+def my_custom_tool(query: str) -> str:
+    """
+    A custom Python function that can be used alongside MCP tools.
+    
+    Args:
+        query: The search query
+        
+    Returns:
+        A formatted result string
+    """
+    return f"Custom tool result for: {query}"
+
+
+def main():
+    # Example 1: Single MCP server (basic usage)
+    print("=" * 60)
+    print("Example 1: Single MCP Server")
+    print("=" * 60)
+    
+    agent_single = Agent(
+        name="TimeAgent",
+        instructions="You are a helpful assistant that can tell time.",
+        tools=MCP("uvx mcp-server-time")
+    )
+    
+    response = agent_single.start("What time is it in Tokyo?")
+    print(f"Response: {response}\n")
+    
+    # Example 2: Multiple MCP servers
+    print("=" * 60)
+    print("Example 2: Multiple MCP Servers")
+    print("=" * 60)
+    
+    agent_multi = Agent(
+        name="MultiMCPAgent",
+        instructions="""You are a helpful assistant with access to multiple tools:
+        - Time tools for getting current time and converting timezones
+        - Memory tools for storing and retrieving information
+        Use the appropriate tool based on the user's request.""",
+        tools=[
+            MCP("uvx mcp-server-time"),                     # Time tools
+            MCP("npx @modelcontextprotocol/server-memory"), # Memory tools
+        ]
+    )
+    
+    response = agent_multi.start("What time is it in New York and London?")
+    print(f"Response: {response}\n")
+    
+    # Example 3: Multiple MCP servers + custom functions
+    print("=" * 60)
+    print("Example 3: MCP Servers + Custom Functions")
+    print("=" * 60)
+    
+    agent_mixed = Agent(
+        name="MixedToolsAgent",
+        instructions="""You are a helpful assistant with access to:
+        - Time tools for timezone operations
+        - A custom search tool for general queries
+        Choose the right tool for each task.""",
+        tools=[
+            MCP("uvx mcp-server-time"),  # MCP server
+            my_custom_tool                # Regular Python function
+        ]
+    )
+    
+    response = agent_mixed.start("What time is it in Paris?")
+    print(f"Response: {response}\n")
+    
+    # Example 4: MCP with environment variables
+    print("=" * 60)
+    print("Example 4: MCP with Environment Variables")
+    print("=" * 60)
+    
+    # Note: Replace with your actual API key
+    # agent_with_env = Agent(
+    #     name="SearchAgent",
+    #     instructions="You can search the web for information.",
+    #     tools=[
+    #         MCP(
+    #             command="npx",
+    #             args=["-y", "@modelcontextprotocol/server-brave-search"],
+    #             env={"BRAVE_API_KEY": "your-api-key-here"}
+    #         ),
+    #         MCP("uvx mcp-server-time")
+    #     ]
+    # )
+    print("(Skipped - requires BRAVE_API_KEY)")
+
+
+if __name__ == "__main__":
+    main()

examples/python/scheduled_agents/news_checker_agent.py

@@ -0,0 +1,115 @@
+"""
+24/7 News Checker Agent Example
+
+This example demonstrates how to create a continuously running agent
+that checks for AI news every hour and prints the results.
+
+Usage:
+    python news_checker_agent.py
+    
+The agent will:
+1. Run immediately on start
+2. Check for latest AI news every 1 hour
+3. Print results to console
+4. Continue running until stopped (Ctrl+C)
+"""
+
+import os
+import sys
+from datetime import datetime
+
+# Add parent directory to path for imports
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), '../../..'))
+
+from praisonaiagents import Agent
+from praisonai.agent_scheduler import AgentScheduler
+from tools import search_tool
+
+
+def on_success(result):
+    """Callback function called when agent execution succeeds."""
+    print("\n" + "="*80)
+    print(f"✅ SUCCESS - {datetime.now().strftime('%Y-%m-%d %H:%M:%S')}")
+    print("="*80)
+    print(result)
+    print("="*80 + "\n")
+
+
+def on_failure(error):
+    """Callback function called when agent execution fails."""
+    print("\n" + "="*80)
+    print(f"❌ FAILURE - {datetime.now().strftime('%Y-%m-%d %H:%M:%S')}")
+    print("="*80)
+    print(f"Error: {error}")
+    print("="*80 + "\n")
+
+
+def main():
+    """Main function to set up and run the scheduled news checker agent."""
+    
+    print("\n🤖 Starting 24/7 News Checker Agent")
+    print("="*80)
+    
+    # Create the agent with web search capability
+    agent = Agent(
+        name="NewsChecker",
+        role="AI News Monitor",
+        goal="Monitor and summarize the latest AI and technology news",
+        backstory="""You are an expert AI news analyst who stays up-to-date 
+        with the latest developments in artificial intelligence, machine learning, 
+        and technology. You provide concise, accurate summaries of important news.""",
+        tools=[search_tool],
+        verbose=True
+    )
+    
+    # Define the task
+    task = """Search for the latest AI news from today. 
+    Focus on major developments, breakthroughs, or announcements.
+    Provide a summary of the top 3-5 most important news items.
+    Include the source and key details for each item."""
+    
+    # Create scheduler with callbacks
+    scheduler = AgentScheduler(
+        agent=agent,
+        task=task,
+        on_success=on_success,
+        on_failure=on_failure
+    )
+    
+    print(f"Agent: {agent.name}")
+    print(f"Task: {task}")
+    print(f"Schedule: Every 1 hour")
+    print("="*80)
+    print("\n⏰ Starting scheduler... (Press Ctrl+C to stop)\n")
+    
+    # Start the scheduler
+    # - "hourly" or "*/1h" = every 1 hour
+    # - run_immediately=True = run once before starting the schedule
+    scheduler.start(
+        schedule_expr="hourly",  # Run every hour
+        max_retries=3,           # Retry up to 3 times on failure
+        run_immediately=True     # Run immediately on start
+    )
+    
+    try:
+        # Keep the main thread alive
+        while scheduler.is_running:
+            import time
+            time.sleep(1)
+            
+    except KeyboardInterrupt:
+        print("\n\n🛑 Stopping scheduler...")
+        scheduler.stop()
+        
+        # Print final statistics
+        stats = scheduler.get_stats()
+        print("\n📊 Final Statistics:")
+        print(f"  Total Executions: {stats['total_executions']}")
+        print(f"  Successful: {stats['successful_executions']}")
+        print(f"  Failed: {stats['failed_executions']}")
+        print(f"  Success Rate: {stats['success_rate']:.1f}%")
+        print("\n✅ Agent stopped successfully\n")
+
+
+if __name__ == "__main__":
+    main()

examples/python/scheduled_agents/news_checker_live.py

@@ -0,0 +1,143 @@
+"""
+Live News Checker Agent - 24/7 Scheduled Agent
+
+This example demonstrates a real working agent that checks AI news
+every hour using the AgentScheduler with actual API calls.
+
+Requirements:
+    - OPENAI_API_KEY environment variable set
+    - praisonaiagents package installed
+    
+Usage:
+    python news_checker_live.py
+"""
+
+import os
+import sys
+from datetime import datetime
+
+# Add parent directory to path
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), '../../..'))
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), '../../../..'))
+
+from praisonaiagents import Agent
+from praisonai.scheduler import AgentScheduler
+
+# Import search_tool from root tools.py
+try:
+    from tools import search_tool
+except ImportError:
+    # Fallback: create inline search tool
+    from duckduckgo_search import DDGS
+    
+    def search_tool(query: str) -> list:
+        """Search the web using DuckDuckGo."""
+        try:
+            results = []
+            ddgs = DDGS()
+            for result in ddgs.text(keywords=query, max_results=5):
+                results.append({
+                    "title": result.get("title", ""),
+                    "url": result.get("href", ""),
+                    "snippet": result.get("body", ""),
+                })
+            return results
+        except Exception as e:
+            print(f"Search error: {e}")
+            return []
+
+
+def on_success(result):
+    """Callback when agent execution succeeds."""
+    print("\n" + "="*80)
+    print(f"✅ SUCCESS - {datetime.now().strftime('%Y-%m-%d %H:%M:%S')}")
+    print("="*80)
+    print(result)
+    print("="*80 + "\n")
+
+
+def on_failure(error):
+    """Callback when agent execution fails."""
+    print("\n" + "="*80)
+    print(f"❌ FAILURE - {datetime.now().strftime('%Y-%m-%d %H:%M:%S')}")
+    print("="*80)
+    print(f"Error: {error}")
+    print("="*80 + "\n")
+
+
+def main():
+    """Main function to run the scheduled news checker."""
+    
+    # Check for API key
+    if not os.getenv("OPENAI_API_KEY"):
+        print("❌ Error: OPENAI_API_KEY environment variable not set")
+        print("Please set it with: export OPENAI_API_KEY=your_key_here")
+        sys.exit(1)
+    
+    print("\n🤖 Starting 24/7 AI News Checker Agent")
+    print("="*80)
+    
+    # Create the agent with web search capability
+    agent = Agent(
+        name="AI News Monitor",
+        role="Technology News Analyst",
+        goal="Monitor and summarize the latest AI and technology news",
+        backstory="""You are an expert AI news analyst who stays up-to-date 
+        with the latest developments in artificial intelligence, machine learning, 
+        and technology. You provide concise, accurate summaries of important news.""",
+        tools=[search_tool],
+        verbose=True,
+        self_reflect=False  # Disable reflection for faster execution
+    )
+    
+    # Define the task
+    task = """Search for the latest AI news from today. 
+    Focus on major developments, breakthroughs, or announcements.
+    Provide a summary of the top 3 most important news items.
+    Include the source and key details for each item."""
+    
+    # Create scheduler with callbacks
+    scheduler = AgentScheduler(
+        agent=agent,
+        task=task,
+        on_success=on_success,
+        on_failure=on_failure
+    )
+    
+    print(f"Agent: {agent.name}")
+    print(f"Task: {task}")
+    print(f"Schedule: Every 1 hour")
+    print("="*80)
+    print("\n⏰ Starting scheduler... (Press Ctrl+C to stop)\n")
+    
+    # Start the scheduler
+    # For testing: use "*/2m" (every 2 minutes) instead of "hourly"
+    # For production: use "hourly"
+    scheduler.start(
+        schedule_expr="*/2m",  # Every 2 minutes for testing
+        max_retries=3,
+        run_immediately=True  # Run once immediately
+    )
+    
+    try:
+        # Keep the main thread alive
+        while scheduler.is_running:
+            import time
+            time.sleep(1)
+            
+    except KeyboardInterrupt:
+        print("\n\n🛑 Stopping scheduler...")
+        scheduler.stop()
+        
+        # Print final statistics
+        stats = scheduler.get_stats()
+        print("\n📊 Final Statistics:")
+        print(f"  Total Executions: {stats['total_executions']}")
+        print(f"  Successful: {stats['successful_executions']}")
+        print(f"  Failed: {stats['failed_executions']}")
+        print(f"  Success Rate: {stats['success_rate']:.1f}%")
+        print("\n✅ Agent stopped successfully\n")
+
+
+if __name__ == "__main__":
+    main()

examples/python/scheduled_agents/simple_test.py

@@ -0,0 +1,61 @@
+"""
+Simple test script for 24/7 Agent Scheduler
+
+This script demonstrates the basic usage of AgentScheduler
+for running agents periodically.
+
+Usage:
+    python simple_test.py
+"""
+
+import sys
+import os
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), '../../..'))
+
+from unittest.mock import Mock
+from praisonai.scheduler import AgentScheduler
+
+# Create a mock agent for testing
+mock_agent = Mock()
+mock_agent.name = "TestAgent"
+mock_agent.start = Mock(return_value="Test execution successful!")
+
+# Create scheduler
+scheduler = AgentScheduler(
+    agent=mock_agent,
+    task="Test task execution"
+)
+
+print("🤖 Testing AgentScheduler")
+print("=" * 60)
+
+# Test 1: Execute once
+print("\n✅ Test 1: Execute once")
+result = scheduler.execute_once()
+print(f"Result: {result}")
+
+# Test 2: Get stats
+print("\n✅ Test 2: Get statistics")
+stats = scheduler.get_stats()
+print(f"Stats: {stats}")
+
+# Test 3: Start scheduler (will run every 5 seconds)
+print("\n✅ Test 3: Start scheduler (*/5s)")
+scheduler.start("*/5s", run_immediately=True)
+
+import time
+print("Running for 12 seconds...")
+time.sleep(12)
+
+# Stop scheduler
+print("\n✅ Test 4: Stop scheduler")
+scheduler.stop()
+
+# Final stats
+print("\n📊 Final Statistics:")
+final_stats = scheduler.get_stats()
+for key, value in final_stats.items():
+    print(f"  {key}: {value}")
+
+print("\n✅ All tests passed!")
+print("=" * 60)

examples/python/tools/e2b/agents.yaml

@@ -1,8 +1,8 @@
 framework: praisonai
 topic: crawl and extract all data from each url from this page https://news.ycombinator.com/
-roles:
+agents:  # Canonical: use 'agents' instead of 'roles'
   web_scraper:
-    backstory: An expert in data extraction from websites, adept at navigating and
+    instructions:  # Canonical An expert in data extraction from websites, adept at navigating and
       retrieving detailed information.
     goal: Extract URLs from the source page
     role: Web Scraper
@@ -15,7 +15,7 @@ roles:
     tools:
     - 'code_interpret'
   url_data_extractor:
-    backstory: Specializes in crawling websites to gather comprehensive data, ensuring
+    instructions:  # Canonical Specializes in crawling websites to gather comprehensive data, ensuring
       nothing is missed from each link.
     goal: Crawl each URL for data extraction
     role: URL Data Extractor
@@ -29,7 +29,7 @@ roles:
     tools:
     - 'code_interpret'
   data_organizer:
-    backstory: Known for keen attention to detail, capable of structuring vast amounts
+    instructions:  # Canonical Known for keen attention to detail, capable of structuring vast amounts
       of information into usable formats efficiently.
     goal: Organize extracted data into a structured format
     role: Data Organizer

examples/scheduler/test_5sec_schedule.yaml

@@ -0,0 +1,14 @@
+framework: praisonai
+
+agents:
+  - name: "5 Second Timer"
+    role: "Timer Agent"
+    instructions: "Display the current time"
+    verbose: true
+
+task: "Show current time"
+
+schedule:
+  interval: "*/5s"
+  max_retries: 3
+  run_immediately: true

examples/scheduler/test_counter.yaml

@@ -0,0 +1,14 @@
+framework: praisonai
+
+agents:
+  - name: "Counter Agent"
+    role: "Counter"
+    instructions: "Count from 1 to 5, incrementing by 1 each time. Just output the numbers clearly."
+    verbose: true
+
+task: "Print numbers 1, 2, 3, 4, 5 (one number per line)"
+
+schedule:
+  interval: "*/5s"
+  max_retries: 3
+  run_immediately: true

examples/scheduler/test_counter_with_limits.yaml

@@ -0,0 +1,16 @@
+framework: praisonai
+
+agents:
+  - name: "Counter Agent with Limits"
+    role: "Counter"
+    instructions: "Count from 1 to 5, incrementing by 1 each time. Just output the numbers clearly."
+    verbose: true
+
+task: "Print numbers 1, 2, 3, 4, 5 (one number per line)"
+
+schedule:
+  interval: "*/5s"           # Run every 5 seconds
+  max_retries: 3             # Retry up to 3 times on failure
+  run_immediately: true      # Run once immediately on start
+  timeout: 30                # Max 30 seconds per execution
+  max_cost: 0.01             # Stop after spending $0.01 USD

examples/scheduler/test_default_budget.yaml

@@ -0,0 +1,15 @@
+framework: praisonai
+
+agents:
+  - name: "Budget Test Agent"
+    role: "Tester"
+    instructions: "Count from 1 to 3"
+    verbose: true
+
+task: "Print numbers 1, 2, 3"
+
+schedule:
+  interval: "*/3s"
+  max_retries: 2
+  run_immediately: true
+  # No max_cost specified - should default to $1.00

examples/skills/basic_skill_usage.py

@@ -0,0 +1,73 @@
+"""
+Basic Agent Skills Usage Example
+
+This example demonstrates how to use Agent Skills with PraisonAI Agents.
+Skills are modular capabilities that can be loaded and used by agents.
+"""
+
+from praisonaiagents import Agent, SkillManager
+
+# Example 1: Using SkillManager directly
+print("=" * 50)
+print("Example 1: Using SkillManager directly")
+print("=" * 50)
+
+# Create a skill manager
+manager = SkillManager()
+
+# Discover skills from a directory
+# This scans the directory for subdirectories containing SKILL.md files
+skill_count = manager.discover(["./pdf-processing"], include_defaults=False)
+print(f"Discovered {skill_count} skill(s)")
+
+# List available skills
+for skill in manager:
+    print(f"  - {skill.properties.name}: {skill.properties.description[:50]}...")
+
+# Generate prompt XML for system prompt injection
+prompt_xml = manager.to_prompt()
+print("\nGenerated prompt XML:")
+print(prompt_xml)
+
+# Example 2: Using skills with an Agent
+print("\n" + "=" * 50)
+print("Example 2: Using skills with an Agent")
+print("=" * 50)
+
+# Create an agent with skills
+# Skills are lazy-loaded only when accessed
+agent = Agent(
+    name="PDF Assistant",
+    instructions="You are a helpful assistant that can process PDF documents.",
+    skills=["./pdf-processing"],  # Direct skill paths
+)
+
+# Access the skill manager through the agent
+if agent.skill_manager:
+    print(f"Agent has {len(agent.skill_manager)} skill(s) loaded")
+    
+    # Get the skills prompt for system prompt injection
+    skills_prompt = agent.get_skills_prompt()
+    print("\nAgent skills prompt:")
+    print(skills_prompt)
+
+# Example 3: Discovering skills from multiple directories
+print("\n" + "=" * 50)
+print("Example 3: Using skills_dirs for discovery")
+print("=" * 50)
+
+# Create an agent that discovers skills from directories
+agent_with_discovery = Agent(
+    name="Multi-Skill Agent",
+    instructions="You are a versatile assistant with multiple skills.",
+    skills_dirs=["./"],  # Scan current directory for skill subdirectories
+)
+
+if agent_with_discovery.skill_manager:
+    print(f"Discovered {len(agent_with_discovery.skill_manager)} skill(s)")
+    for name in agent_with_discovery.skill_manager.skill_names:
+        print(f"  - {name}")
+
+print("\n" + "=" * 50)
+print("Examples complete!")
+print("=" * 50)

examples/skills/custom_skill_example.py

@@ -0,0 +1,163 @@
+"""
+Custom Skill Creation Example
+
+This example demonstrates how to programmatically create and validate
+custom Agent Skills using PraisonAI Agents.
+"""
+
+import tempfile
+from pathlib import Path
+
+from praisonaiagents.skills import (
+    SkillManager,
+    SkillLoader,
+    validate,
+    read_properties,
+)
+
+
+def create_custom_skill(
+    name: str,
+    description: str,
+    instructions: str,
+    base_dir: str = None
+) -> Path:
+    """Create a custom skill programmatically.
+    
+    Args:
+        name: Skill name (kebab-case)
+        description: What the skill does
+        instructions: Markdown instructions for the skill
+        base_dir: Directory to create skill in (default: temp dir)
+        
+    Returns:
+        Path to the created skill directory
+    """
+    if base_dir is None:
+        base_dir = tempfile.mkdtemp()
+    
+    skill_dir = Path(base_dir) / name
+    skill_dir.mkdir(parents=True, exist_ok=True)
+    
+    # Create SKILL.md
+    skill_md_content = f"""---
+name: {name}
+description: {description}
+license: Apache-2.0
+metadata:
+  author: custom
+  version: "1.0"
+---
+
+{instructions}
+"""
+    (skill_dir / "SKILL.md").write_text(skill_md_content)
+    
+    # Create optional directories
+    (skill_dir / "scripts").mkdir(exist_ok=True)
+    (skill_dir / "references").mkdir(exist_ok=True)
+    (skill_dir / "assets").mkdir(exist_ok=True)
+    
+    return skill_dir
+
+
+def main():
+    print("=" * 60)
+    print("Custom Skill Creation Example")
+    print("=" * 60)
+    
+    # Create a custom skill
+    print("\n1. Creating a custom skill...")
+    
+    skill_path = create_custom_skill(
+        name="code-review",
+        description="Review code for best practices, bugs, and improvements. Use when asked to review or analyze code quality.",
+        instructions="""
+# Code Review Skill
+
+## Overview
+This skill enables thorough code review with focus on:
+- Code quality and readability
+- Potential bugs and edge cases
+- Performance considerations
+- Security vulnerabilities
+- Best practices adherence
+
+## Instructions
+
+When reviewing code:
+
+1. **Read the code carefully** - Understand the purpose and flow
+2. **Check for bugs** - Look for null checks, edge cases, error handling
+3. **Evaluate readability** - Variable names, comments, structure
+4. **Consider performance** - Time/space complexity, unnecessary operations
+5. **Security review** - Input validation, injection risks, data exposure
+6. **Suggest improvements** - Provide actionable recommendations
+
+## Output Format
+
+Structure your review as:
+- Summary: Brief overview of the code
+- Issues: List of problems found (severity: high/medium/low)
+- Suggestions: Recommended improvements
+- Positive aspects: What's done well
+"""
+    )
+    
+    print(f"   Created skill at: {skill_path}")
+    
+    # Validate the skill
+    print("\n2. Validating the skill...")
+    errors = validate(skill_path)
+    
+    if errors:
+        print("   Validation errors:")
+        for error in errors:
+            print(f"   - {error}")
+    else:
+        print("   ✓ Skill is valid!")
+    
+    # Load and inspect the skill
+    print("\n3. Loading skill properties...")
+    props = read_properties(skill_path)
+    print(f"   Name: {props.name}")
+    print(f"   Description: {props.description[:60]}...")
+    print(f"   License: {props.license}")
+    
+    # Use with SkillLoader for progressive loading
+    print("\n4. Using SkillLoader for progressive disclosure...")
+    loader = SkillLoader()
+    
+    # Level 1: Load metadata only
+    skill = loader.load_metadata(str(skill_path))
+    print(f"   Level 1 - Metadata loaded: {skill.properties.name}")
+    print(f"   Instructions loaded: {skill.is_activated}")
+    
+    # Level 2: Activate (load instructions)
+    loader.activate(skill)
+    print(f"   Level 2 - Activated: {skill.is_activated}")
+    print(f"   Instructions preview: {skill.instructions[:100]}...")
+    
+    # Use with SkillManager
+    print("\n5. Using with SkillManager...")
+    manager = SkillManager()
+    manager.add_skill(str(skill_path))
+    
+    # Generate prompt XML
+    prompt = manager.to_prompt()
+    print("   Generated prompt XML:")
+    print(prompt)
+    
+    # Clean up
+    print("\n6. Cleanup...")
+    import shutil
+    shutil.rmtree(skill_path.parent)
+    print("   Temporary files removed.")
+    
+    print("\n" + "=" * 60)
+    print("Example complete!")
+    print("=" * 60)
+
+
+if __name__ == "__main__":
+    main()

examples/skills/pdf-processing/SKILL.md

@@ -0,0 +1,38 @@
+---
+name: pdf-processing
+description: Process and extract information from PDF documents. Use this skill when the user asks to read, analyze, or extract data from PDF files.
+license: Apache-2.0
+compatibility: Works with PraisonAI Agents
+metadata:
+  author: praisonai
+  version: "1.0"
+---
+
+# PDF Processing Skill
+
+## Overview
+
+This skill enables agents to process PDF documents, extract text content, and analyze the information within them.
+
+## When to Use
+
+Activate this skill when:
+- User asks to read or analyze a PDF file
+- User needs to extract text from a PDF
+- User wants to summarize a PDF document
+- User needs to search for information within a PDF
+
+## Instructions
+
+1. First, verify the PDF file exists at the specified path
+2. Use appropriate tools to read the PDF content
+3. Extract text while preserving structure where possible
+4. For large PDFs, process in chunks to manage context
+5. Summarize or analyze based on user's specific request
+
+## Best Practices
+
+- Always confirm the file path before processing
+- Handle encrypted PDFs gracefully with appropriate error messages
+- For scanned PDFs, note that OCR may be required
+- Preserve important formatting like tables and lists

examples/yaml/agents_workflow.yaml

@@ -11,7 +11,7 @@
 
 framework: praisonai
 process: workflow
-topic: "Research AI trends and create a comprehensive report"
+input: "Research AI trends and create a comprehensive report"  # The input passed to {{input}} in steps
 
 # Workflow configuration (new)
 workflow:
@@ -31,17 +31,17 @@ variables:
     - Natural Language Processing
 
 # Agent definitions using familiar 'roles' format
-roles:
+agents:  # Canonical: use 'agents' instead of 'roles'
   classifier:
     role: Request Classifier
-    backstory: |
+    instructions:  # Canonical |
       You are an expert at classifying requests into categories.
       Respond with ONLY one word: 'technical', 'creative', or 'research'.
     goal: Classify incoming requests accurately
     
   researcher:
     role: Research Analyst
-    backstory: |
+    instructions:  # Canonical |
       You are a skilled research analyst with expertise in technology trends.
       You provide concise, factual research findings with citations.
     goal: Research topics thoroughly and provide insights
@@ -52,28 +52,28 @@ roles:
       
   tech_expert:
     role: Technical Expert
-    backstory: |
+    instructions:  # Canonical |
       You are a technical expert with deep knowledge in AI and ML.
       You provide detailed technical explanations.
     goal: Handle technical questions with expertise
     
   creative_writer:
     role: Creative Writer
-    backstory: |
+    instructions:  # Canonical |
       You are a creative writer who can make complex topics engaging.
       You write in a clear, accessible style.
     goal: Create engaging content from research
     
   market_analyst:
     role: Market Analyst
-    backstory: |
+    instructions:  # Canonical |
       You analyze market trends and competitive landscapes.
       You provide actionable business insights.
     goal: Analyze market trends and opportunities
     
   aggregator:
     role: Synthesizer
-    backstory: |
+    instructions:  # Canonical |
       You combine findings from multiple sources into coherent reports.
       You identify patterns and key insights.
     goal: Synthesize all findings into comprehensive reports

examples/yaml/workflows/loop_workflow.yaml

@@ -39,4 +39,4 @@ steps:
       over: topics
       
   - agent: summarizer
-    action: "Create a comprehensive summary of all the analyzed topics"
+    action: "Create a comprehensive summary of all the analyzed topics based on these results: {{previous_output}}"

examples/yaml/workflows/parallel_workflow.yaml

@@ -3,6 +3,7 @@
 
 name: Parallel Research Workflow
 description: Multiple agents research different aspects concurrently, then an aggregator combines results
+input: "AI-powered customer service solutions"  # The input passed to {{input}} in steps
 
 workflow:
   verbose: true
@@ -43,4 +44,4 @@ steps:
         action: "Research customer needs for: {{input}}"
         
   - agent: aggregator
-    action: "Synthesize all the research findings into a comprehensive report"
+    action: "Synthesize all the research findings into a comprehensive report: {{previous_output}}"

examples/yaml/workflows/routing_workflow.yaml

@@ -3,6 +3,7 @@
 
 name: Agentic Routing Workflow
 description: A classifier agent routes requests to specialized handler agents
+input: "Write a poem about the ocean"  # The input passed to {{input}} in steps
 
 workflow:
   verbose: true

examples/yaml/workflows/simple_workflow.yaml

@@ -3,6 +3,7 @@
 
 name: Simple Research Workflow
 description: A simple workflow where a researcher gathers info and a writer summarizes it
+input: "The benefits of renewable energy"  # The input passed to {{input}} in steps
 
 workflow:
   verbose: true

src/praisonai-agents/.gitignore

@@ -87,3 +87,4 @@ graph.py
 chroma_db/
 .qodo
 *_COMPARISON_ANALYSIS.md
+*_REPORT.md

src/praisonai-agents/README.md

@@ -0,0 +1 @@
+../../README.md

src/praisonai-agents/benchmarks/BENCHMARK_RESULTS.md

@@ -0,0 +1,35 @@
+# PraisonAI Agents - Benchmark Results
+
+**Generated:** 2025-12-18 14:40:22
+**Iterations:** 100
+**Test:** Agent instantiation (without tools)
+
+## Results
+
+| Framework | Avg Time (μs) | Relative |
+|-----------|---------------|----------|
+| **PraisonAI** | **3.77** | **1.00x (fastest)** |
+| OpenAI Agents SDK | 5.26 | 1.39x |
+| Agno | 5.64 | 1.49x |
+| PraisonAI (LiteLLM) | 7.56 | 2.00x |
+| PydanticAI | 226.94 | 60.16x |
+| LangGraph | 4,558.71 | 1,209x |
+| CrewAI | 15,607.92 | 4,138x |
+
+## Package Versions
+
+| Package | Version |
+|---------|--------|
+| PraisonAI | 0.1.5 |
+| Agno | 2.3.14 |
+| PydanticAI | 1.35.0 |
+| OpenAI Agents SDK | 0.6.3 |
+| LangGraph | 1.0.5 |
+| CrewAI | 1.6.1 |
+
+## How to Reproduce
+
+```bash
+cd praisonai-agents
+python benchmarks/simple_benchmark.py
+```

src/praisonai-agents/benchmarks/EXECUTION_BENCHMARK_RESULTS.md

@@ -0,0 +1,31 @@
+# PraisonAI Agents - Real Execution Benchmark
+
+**Generated:** 2025-12-18 14:43:52
+**Model:** gpt-4o-mini
+**Iterations:** 3
+**Prompt:** "What is 2+2? Reply with just the number."
+
+## Results
+
+| Framework | Method | Avg Time | Relative |
+|-----------|--------|----------|----------|
+| **PraisonAI** | `agent.start()` | **0.45s** | **1.00x (fastest)** |
+| PraisonAI (LiteLLM) | `agent.start()` | 0.55s | 1.22x |
+| CrewAI | `crew.kickoff()` | 0.58s | 1.28x |
+| Agno | `agent.run()` | 0.92s | 2.05x |
+
+## Package Versions
+
+| Package | Version |
+|---------|--------|
+| PraisonAI | 0.1.5 |
+| Agno | 2.3.14 |
+| CrewAI | 1.6.1 |
+
+## How to Reproduce
+
+```bash
+export OPENAI_API_KEY=your_key
+cd praisonai-agents
+python benchmarks/execution_benchmark.py
+```

src/praisonai-agents/benchmarks/REAL_BENCHMARK_RESULTS.md

@@ -0,0 +1,72 @@
+# PraisonAI Agents - Real-World Execution Benchmark
+
+**Generated:** 2025-12-18 14:33:49
+**Model:** gpt-4o-mini
+**Iterations:** 3 per test
+
+## Results Summary
+
+| Test | PraisonAI (avg) | Agno (avg) | Difference |
+|------|-----------------|------------|------------|
+| Single Agent | 2.04s | 3.18s | PraisonAI 1.56x faster |
+| 2 Agents | 2.49s | 20.49s | PraisonAI 8.21x faster |
+| 3 Agents | 6.52s | 13.97s | PraisonAI 2.14x faster |
+
+## Detailed Results
+
+### praisonai_single
+- Average: 2.04s
+- Min: 1.95s
+- Max: 2.19s
+- Std Dev: 0.13s
+- Times: ['2.19s', '1.95s', '2.00s']
+
+### agno_single
+- Average: 3.18s
+- Min: 3.00s
+- Max: 3.51s
+- Std Dev: 0.28s
+- Times: ['3.51s', '3.00s', '3.04s']
+
+### praisonai_two_agents
+- Average: 2.49s
+- Min: 2.22s
+- Max: 2.65s
+- Std Dev: 0.24s
+- Times: ['2.65s', '2.22s', '2.62s']
+
+### agno_two_agents
+- Average: 20.49s
+- Min: 9.96s
+- Max: 35.05s
+- Std Dev: 13.02s
+- Times: ['16.46s', '9.96s', '35.05s']
+
+### praisonai_three_agents
+- Average: 6.52s
+- Min: 5.62s
+- Max: 7.39s
+- Std Dev: 0.88s
+- Times: ['5.62s', '6.54s', '7.39s']
+
+### agno_three_agents
+- Average: 13.97s
+- Min: 11.69s
+- Max: 15.32s
+- Std Dev: 1.99s
+- Times: ['14.91s', '11.69s', '15.32s']
+
+## Package Versions
+
+| Package | Version |
+|---------|--------|
+| PraisonAI | 0.1.5 |
+| Agno | 2.3.14 |
+
+## How to Reproduce
+
+```bash
+export OPENAI_API_KEY=your_key
+cd praisonai-agents
+python benchmarks/real_benchmark.py
+```

src/praisonai-agents/benchmarks/SIMPLE_BENCHMARK_RESULTS.md

@@ -0,0 +1,22 @@
+# PraisonAI Agents - Benchmark Results
+
+**Generated:** 2025-12-18 13:23:43
+
+## Agent Instantiation Time
+
+| Framework | Avg Time (μs) | Relative |
+|-----------|---------------|----------|
+| **PraisonAI** | **4.16** | **1.00x (fastest)** |
+| OpenAI Agents SDK | 5.55 | 1.33x |
+| Agno | 5.58 | 1.34x |
+| PraisonAI (LiteLLM) | 8.52 | 2.05x |
+| PydanticAI | 225.87 | 54.25x |
+| LangGraph | 4,349.54 | 1,045x |
+| CrewAI | 15,911.11 | 3,821x |
+
+## How to Reproduce
+
+```bash
+cd praisonai-agents
+python benchmarks/simple_benchmark.py
+```

src/praisonai-agents/benchmarks/TOOLS_BENCHMARK_RESULTS.md

@@ -0,0 +1,33 @@
+# PraisonAI Agents - Tools Benchmark Results
+
+**Generated:** 2025-12-18 14:40:35
+**Iterations:** 100
+**Test:** Agent instantiation WITH TOOLS
+
+## Results
+
+| Framework | Avg Time (μs) | Relative |
+|-----------|---------------|----------|
+| **PraisonAI** | **3.24** | **1.00x (fastest)** |
+| Agno | 5.12 | 1.58x |
+| PraisonAI (LiteLLM) | 8.59 | 2.65x |
+| OpenAI Agents SDK | 279.95 | 86.44x |
+| LangGraph | 2,310.82 | 713x |
+| CrewAI | 15,773.44 | 4,870x |
+
+## Package Versions
+
+| Package | Version |
+|---------|--------|
+| PraisonAI | 0.1.5 |
+| Agno | 2.3.14 |
+| OpenAI Agents SDK | 0.6.3 |
+| LangGraph | 1.0.5 |
+| CrewAI | 1.6.1 |
+
+## How to Reproduce
+
+```bash
+cd praisonai-agents
+python benchmarks/tools_benchmark.py
+```

src/praisonai-agents/benchmarks/deep_profiling.py

@@ -0,0 +1,368 @@
+#!/usr/bin/env python3
+"""
+PraisonAI Agents - Deep Profiling Benchmark
+
+Profiles each step of agent initialization to identify performance bottlenecks.
+
+Key Optimizations Applied:
+1. Lazy RulesManager - filesystem operations deferred until first access
+2. Lazy os.getcwd() - fast_context_path resolved only when needed
+3. Lazy console initialization - Rich Console created on demand
+
+Usage:
+    python benchmarks/deep_profiling.py
+"""
+
+import time
+import tracemalloc
+import cProfile
+import pstats
+import io
+from typing import Literal, Callable, Dict, List
+from dataclasses import dataclass
+
+# ============================================================================
+# Timing Utilities
+# ============================================================================
+
+@dataclass
+class StepTiming:
+    name: str
+    duration_us: float  # microseconds
+    memory_kb: float    # kilobytes
+
+
+class StepProfiler:
+    """Profile individual steps during initialization"""
+    
+    def __init__(self):
+        self.steps: List[StepTiming] = []
+        self._start_time = None
+        self._start_memory = None
+    
+    def start_step(self, name: str):
+        tracemalloc.start()
+        self._start_time = time.perf_counter()
+        self._step_name = name
+    
+    def end_step(self):
+        end_time = time.perf_counter()
+        current, peak = tracemalloc.get_traced_memory()
+        tracemalloc.stop()
+        
+        duration_us = (end_time - self._start_time) * 1_000_000
+        memory_kb = peak / 1024
+        
+        self.steps.append(StepTiming(
+            name=self._step_name,
+            duration_us=duration_us,
+            memory_kb=memory_kb
+        ))
+    
+    def print_report(self, title: str):
+        print(f"\n{'='*70}")
+        print(f"STEP-BY-STEP PROFILING: {title}")
+        print(f"{'='*70}")
+        
+        total_time = sum(s.duration_us for s in self.steps)
+        total_memory = sum(s.memory_kb for s in self.steps)
+        
+        print(f"\n{'Step':<40} {'Time (μs)':<15} {'Memory (KB)':<15} {'% Time':<10}")
+        print("-" * 80)
+        
+        for step in sorted(self.steps, key=lambda x: x.duration_us, reverse=True):
+            pct = (step.duration_us / total_time * 100) if total_time > 0 else 0
+            print(f"{step.name:<40} {step.duration_us:<15.2f} {step.memory_kb:<15.2f} {pct:<10.1f}%")
+        
+        print("-" * 80)
+        print(f"{'TOTAL':<40} {total_time:<15.2f} {total_memory:<15.2f} {'100.0':<10}%")
+
+
+def time_function(func: Callable, iterations: int = 100) -> Dict:
+    """Time a function with detailed statistics"""
+    times = []
+    
+    # Warmup
+    for _ in range(10):
+        func()
+    
+    # Measure
+    for _ in range(iterations):
+        start = time.perf_counter()
+        func()
+        end = time.perf_counter()
+        times.append((end - start) * 1_000_000)  # Convert to microseconds
+    
+    return {
+        'avg': sum(times) / len(times),
+        'min': min(times),
+        'max': max(times),
+        'median': sorted(times)[len(times)//2]
+    }
+
+
+# ============================================================================
+# Profile PraisonAI Agent Initialization - Step by Step
+# ============================================================================
+
+def profile_praisonai_detailed():
+    """Profile each step of PraisonAI Agent initialization"""
+    print("\n" + "="*70)
+    print("DETAILED PROFILING: PraisonAI Agent")
+    print("="*70)
+    
+    profiler = StepProfiler()
+    
+    # Step 1: Import
+    profiler.start_step("1. Import praisonaiagents")
+    from praisonaiagents import Agent as PraisonAgent
+    profiler.end_step()
+    
+    # Step 2: Define tool
+    profiler.start_step("2. Define tool function")
+    def get_weather(city: Literal["nyc", "sf"]):
+        """Use this to get weather information."""
+        if city == "nyc":
+            return "It might be cloudy in nyc"
+        elif city == "sf":
+            return "It's always sunny in sf"
+    tools = [get_weather]
+    profiler.end_step()
+    
+    # Step 3: Full instantiation
+    profiler.start_step("3. Full Agent instantiation")
+    agent = PraisonAgent(
+        name="Test Agent",
+        instructions="Be concise, reply with one sentence.",
+        llm="gpt-4o-mini",
+        tools=tools,
+        verbose=False
+    )
+    profiler.end_step()
+    
+    profiler.print_report("PraisonAI Agent")
+    
+    # Now profile the __init__ internals using cProfile
+    print("\n" + "="*70)
+    print("cProfile ANALYSIS: PraisonAI Agent.__init__")
+    print("="*70)
+    
+    pr = cProfile.Profile()
+    pr.enable()
+    
+    for _ in range(100):
+        agent = PraisonAgent(
+            name="Test Agent",
+            instructions="Be concise, reply with one sentence.",
+            llm="gpt-4o-mini",
+            tools=tools,
+            verbose=False
+        )
+    
+    pr.disable()
+    
+    # Print top 30 functions by cumulative time
+    s = io.StringIO()
+    ps = pstats.Stats(pr, stream=s).sort_stats('cumulative')
+    ps.print_stats(30)
+    print(s.getvalue())
+    
+    return agent
+
+
+# ============================================================================
+# Profile Agno Agent Initialization - Step by Step
+# ============================================================================
+
+def profile_agno_detailed():
+    """Profile each step of Agno Agent initialization"""
+    print("\n" + "="*70)
+    print("DETAILED PROFILING: Agno Agent")
+    print("="*70)
+    
+    profiler = StepProfiler()
+    
+    # Step 1: Import
+    profiler.start_step("1. Import agno")
+    from agno.agent import Agent as AgnoAgent
+    from agno.models.openai import OpenAIChat
+    profiler.end_step()
+    
+    # Step 2: Define tool
+    profiler.start_step("2. Define tool function")
+    def get_weather(city: Literal["nyc", "sf"]):
+        """Use this to get weather information."""
+        if city == "nyc":
+            return "It might be cloudy in nyc"
+        elif city == "sf":
+            return "It's always sunny in sf"
+    tools = [get_weather]
+    profiler.end_step()
+    
+    # Step 3: Full instantiation
+    profiler.start_step("3. Full Agent instantiation")
+    agent = AgnoAgent(model=OpenAIChat(id="gpt-4o-mini"), tools=tools)
+    profiler.end_step()
+    
+    profiler.print_report("Agno Agent")
+    
+    # Now profile the __init__ internals using cProfile
+    print("\n" + "="*70)
+    print("cProfile ANALYSIS: Agno Agent.__init__")
+    print("="*70)
+    
+    pr = cProfile.Profile()
+    pr.enable()
+    
+    for _ in range(100):
+        agent = AgnoAgent(model=OpenAIChat(id="gpt-4o-mini"), tools=tools)
+    
+    pr.disable()
+    
+    # Print top 30 functions by cumulative time
+    s = io.StringIO()
+    ps = pstats.Stats(pr, stream=s).sort_stats('cumulative')
+    ps.print_stats(30)
+    print(s.getvalue())
+    
+    return agent
+
+
+# ============================================================================
+# Profile Individual Components of PraisonAI
+# ============================================================================
+
+def profile_praisonai_components():
+    """Profile individual components of PraisonAI Agent initialization"""
+    print("\n" + "="*70)
+    print("COMPONENT-LEVEL PROFILING: PraisonAI Agent")
+    print("="*70)
+    
+    from praisonaiagents import Agent as PraisonAgent
+    
+    def get_weather(city: Literal["nyc", "sf"]):
+        if city == "nyc":
+            return "It might be cloudy in nyc"
+        elif city == "sf":
+            return "It's always sunny in sf"
+    tools = [get_weather]
+    
+    # Test different configurations
+    configs = [
+        ("Minimal (no tools, no llm)", {"name": "Test", "verbose": False}),
+        ("With name only", {"name": "Test Agent", "verbose": False}),
+        ("With instructions", {"name": "Test", "instructions": "Be concise", "verbose": False}),
+        ("With llm (gpt-4o-mini)", {"name": "Test", "llm": "gpt-4o-mini", "verbose": False}),
+        ("With llm (openai/gpt-4o-mini)", {"name": "Test", "llm": "openai/gpt-4o-mini", "verbose": False}),
+        ("With tools", {"name": "Test", "tools": tools, "verbose": False}),
+        ("Full config (OpenAI SDK)", {"name": "Test", "instructions": "Be concise", "llm": "gpt-4o-mini", "tools": tools, "verbose": False}),
+        ("Full config (LiteLLM)", {"name": "Test", "instructions": "Be concise", "llm": "openai/gpt-4o-mini", "tools": tools, "verbose": False}),
+    ]
+    
+    print(f"\n{'Configuration':<40} {'Avg Time (μs)':<15} {'Min (μs)':<15} {'Max (μs)':<15}")
+    print("-" * 85)
+    
+    for name, config in configs:
+        def create_agent():
+            return PraisonAgent(**config)
+        
+        stats = time_function(create_agent, iterations=100)
+        print(f"{name:<40} {stats['avg']:<15.2f} {stats['min']:<15.2f} {stats['max']:<15.2f}")
+
+
+# ============================================================================
+# Compare Specific Operations
+# ============================================================================
+
+def compare_specific_operations():
+    """Compare specific operations between PraisonAI and Agno"""
+    print("\n" + "="*70)
+    print("OPERATION COMPARISON: PraisonAI vs Agno")
+    print("="*70)
+    
+    from praisonaiagents import Agent as PraisonAgent
+    from agno.agent import Agent as AgnoAgent
+    from agno.models.openai import OpenAIChat
+    
+    def get_weather(city: Literal["nyc", "sf"]):
+        if city == "nyc":
+            return "It might be cloudy in nyc"
+        elif city == "sf":
+            return "It's always sunny in sf"
+    tools = [get_weather]
+    
+    operations = []
+    
+    # Operation 1: Bare minimum instantiation
+    def praisonai_minimal():
+        return PraisonAgent(name="Test", verbose=False)
+    
+    def agno_minimal():
+        return AgnoAgent(model=OpenAIChat(id="gpt-4o-mini"))
+    
+    praison_stats = time_function(praisonai_minimal, 100)
+    agno_stats = time_function(agno_minimal, 100)
+    operations.append(("Minimal instantiation", praison_stats['avg'], agno_stats['avg']))
+    
+    # Operation 2: With tools
+    def praisonai_with_tools():
+        return PraisonAgent(name="Test", tools=tools, verbose=False)
+    
+    def agno_with_tools():
+        return AgnoAgent(model=OpenAIChat(id="gpt-4o-mini"), tools=tools)
+    
+    praison_stats = time_function(praisonai_with_tools, 100)
+    agno_stats = time_function(agno_with_tools, 100)
+    operations.append(("With tools", praison_stats['avg'], agno_stats['avg']))
+    
+    # Operation 3: Full config
+    def praisonai_full():
+        return PraisonAgent(
+            name="Test Agent",
+            instructions="Be concise",
+            llm="gpt-4o-mini",
+            tools=tools,
+            verbose=False
+        )
+    
+    def agno_full():
+        return AgnoAgent(
+            model=OpenAIChat(id="gpt-4o-mini"),
+            tools=tools,
+            instructions=["Be concise"]
+        )
+    
+    praison_stats = time_function(praisonai_full, 100)
+    agno_stats = time_function(agno_full, 100)
+    operations.append(("Full config", praison_stats['avg'], agno_stats['avg']))
+    
+    print(f"\n{'Operation':<30} {'PraisonAI (μs)':<18} {'Agno (μs)':<18} {'Ratio':<10}")
+    print("-" * 76)
+    
+    for op_name, praison_time, agno_time in operations:
+        ratio = praison_time / agno_time if agno_time > 0 else 0
+        print(f"{op_name:<30} {praison_time:<18.2f} {agno_time:<18.2f} {ratio:<10.1f}x")
+
+
+# ============================================================================
+# Main
+# ============================================================================
+
+if __name__ == "__main__":
+    import os
+    os.environ.setdefault("OPENAI_API_KEY", "sk-test")
+    
+    print("="*70)
+    print("PraisonAI Agents - Deep Profiling Benchmark")
+    print("="*70)
+    print("\nThis analysis identifies performance bottlenecks in agent initialization.\n")
+    
+    # Run all profiling
+    profile_praisonai_detailed()
+    profile_agno_detailed()
+    profile_praisonai_components()
+    compare_specific_operations()
+    
+    print("\n" + "="*70)
+    print("PROFILING COMPLETE")
+    print("="*70)

src/praisonai-agents/benchmarks/execution_benchmark.py

@@ -0,0 +1,296 @@
+#!/usr/bin/env python3
+"""
+PraisonAI Agents - Real Execution Benchmark
+
+Benchmarks ACTUAL agent execution with LLM API calls.
+Tests single agent running a simple task across frameworks.
+
+Usage:
+    export OPENAI_API_KEY=your_key
+    python benchmarks/execution_benchmark.py
+    python benchmarks/execution_benchmark.py --no-save
+    python benchmarks/execution_benchmark.py --iterations 5
+    python benchmarks/execution_benchmark.py --model gpt-4o --iterations 10
+"""
+
+import time
+import os
+import argparse
+from importlib.metadata import version as get_version
+
+# Defaults (can be overridden via CLI)
+DEFAULT_MODEL = "gpt-4o-mini"
+DEFAULT_ITERATIONS = 3
+DEFAULT_PROMPT = "What is 2+2? Reply with just the number."
+
+
+def get_package_versions():
+    """Get version numbers for benchmarked packages."""
+    packages = {
+        'PraisonAI': 'praisonaiagents',
+        'Agno': 'agno',
+        'CrewAI': 'crewai',
+    }
+    versions = {}
+    for name, pkg in packages.items():
+        try:
+            versions[name] = get_version(pkg)
+        except:
+            versions[name] = 'not installed'
+    return versions
+
+
+def benchmark_praisonai(model, iterations, prompt):
+    """Benchmark PraisonAI agent execution using agent.start()"""
+    from praisonaiagents import Agent
+    
+    print("\n--- PraisonAI ---")
+    print("Method: agent.start()")
+    
+    times = []
+    for i in range(iterations):
+        agent = Agent(
+            name="Calculator",
+            instructions="You are a helpful assistant. Be very brief.",
+            llm=model,
+            verbose=False
+        )
+        
+        start = time.perf_counter()
+        response = agent.start(prompt)
+        elapsed = time.perf_counter() - start
+        
+        times.append(elapsed)
+        print(f"  Run {i+1}: {elapsed:.2f}s - Response: {str(response)[:50]}")
+    
+    avg = sum(times) / len(times)
+    print(f"  Average: {avg:.2f}s")
+    return avg
+
+
+def benchmark_praisonai_litellm(model, iterations, prompt):
+    """Benchmark PraisonAI with LiteLLM backend using agent.start()"""
+    from praisonaiagents import Agent
+    
+    print("\n--- PraisonAI (LiteLLM) ---")
+    print("Method: agent.start()")
+    
+    times = []
+    for i in range(iterations):
+        agent = Agent(
+            name="Calculator",
+            instructions="You are a helpful assistant. Be very brief.",
+            llm=f"openai/{model}",
+            verbose=False
+        )
+        
+        start = time.perf_counter()
+        response = agent.start(prompt)
+        elapsed = time.perf_counter() - start
+        
+        times.append(elapsed)
+        print(f"  Run {i+1}: {elapsed:.2f}s - Response: {str(response)[:50]}")
+    
+    avg = sum(times) / len(times)
+    print(f"  Average: {avg:.2f}s")
+    return avg
+
+
+def benchmark_agno(model, iterations, prompt):
+    """Benchmark Agno agent execution using agent.run()"""
+    from agno.agent import Agent
+    from agno.models.openai import OpenAIChat
+    
+    print("\n--- Agno ---")
+    print("Method: agent.run()")
+    
+    times = []
+    for i in range(iterations):
+        agent = Agent(
+            model=OpenAIChat(id=model),
+            instructions=["You are a helpful assistant. Be very brief."],
+        )
+        
+        start = time.perf_counter()
+        response = agent.run(prompt)
+        elapsed = time.perf_counter() - start
+        
+        times.append(elapsed)
+        # Extract content from response
+        content = str(response.content)[:50] if hasattr(response, 'content') else str(response)[:50]
+        print(f"  Run {i+1}: {elapsed:.2f}s - Response: {content}")
+    
+    avg = sum(times) / len(times)
+    print(f"  Average: {avg:.2f}s")
+    return avg
+
+
+def benchmark_crewai(model, iterations, prompt):
+    """Benchmark CrewAI agent execution using crew.kickoff()"""
+    from crewai import Agent, Task, Crew
+    
+    print("\n--- CrewAI ---")
+    print("Method: crew.kickoff()")
+    
+    times = []
+    for i in range(iterations):
+        agent = Agent(
+            role="Calculator",
+            goal="Answer math questions",
+            backstory="You are a helpful assistant.",
+            llm=model,
+            verbose=False
+        )
+        
+        task = Task(
+            description=prompt,
+            expected_output="A number",
+            agent=agent
+        )
+        
+        crew = Crew(
+            agents=[agent],
+            tasks=[task],
+            verbose=False
+        )
+        
+        start = time.perf_counter()
+        response = crew.kickoff()
+        elapsed = time.perf_counter() - start
+        
+        times.append(elapsed)
+        print(f"  Run {i+1}: {elapsed:.2f}s - Response: {str(response)[:50]}")
+    
+    avg = sum(times) / len(times)
+    print(f"  Average: {avg:.2f}s")
+    return avg
+
+
+def save_results(results: dict, model: str, iterations: int, prompt: str):
+    """Save results to markdown file."""
+    import os
+    from datetime import datetime
+    
+    filepath = os.path.join(os.path.dirname(__file__), 'EXECUTION_BENCHMARK_RESULTS.md')
+    versions = get_package_versions()
+    
+    # Sort by time
+    sorted_results = sorted(results.items(), key=lambda x: x[1])
+    fastest = sorted_results[0][1]
+    
+    with open(filepath, 'w') as f:
+        f.write('# PraisonAI Agents - Real Execution Benchmark\n\n')
+        f.write(f'**Generated:** {datetime.now().strftime("%Y-%m-%d %H:%M:%S")}\n')
+        f.write(f'**Model:** {model}\n')
+        f.write(f'**Iterations:** {iterations}\n')
+        f.write(f'**Prompt:** "{prompt}"\n\n')
+        f.write('## Results\n\n')
+        f.write('| Framework | Method | Avg Time | Relative |\n')
+        f.write('|-----------|--------|----------|----------|\n')
+        
+        methods = {
+            'PraisonAI': 'agent.start()',
+            'PraisonAI (LiteLLM)': 'agent.start()',
+            'Agno': 'agent.run()',
+            'CrewAI': 'crew.kickoff()',
+        }
+        
+        for name, avg in sorted_results:
+            ratio = avg / fastest
+            method = methods.get(name, 'unknown')
+            if ratio == 1.0:
+                f.write(f'| **{name}** | `{method}` | **{avg:.2f}s** | **1.00x (fastest)** |\n')
+            else:
+                f.write(f'| {name} | `{method}` | {avg:.2f}s | {ratio:.2f}x |\n')
+        
+        f.write('\n## Package Versions\n\n')
+        f.write('| Package | Version |\n')
+        f.write('|---------|--------|\n')
+        for pkg, ver in versions.items():
+            f.write(f'| {pkg} | {ver} |\n')
+        
+        f.write('\n## How to Reproduce\n\n')
+        f.write('```bash\n')
+        f.write('export OPENAI_API_KEY=your_key\n')
+        f.write('cd praisonai-agents\n')
+        f.write('python benchmarks/execution_benchmark.py\n')
+        f.write('```\n')
+    
+    print(f'\nResults saved to: {filepath}')
+
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description='PraisonAI Agents - Real Execution Benchmark')
+    parser.add_argument('--no-save', action='store_true', help='Do not save results to file')
+    parser.add_argument('--model', '-m', type=str, default=DEFAULT_MODEL, help=f'Model to use (default: {DEFAULT_MODEL})')
+    parser.add_argument('--iterations', '-i', type=int, default=DEFAULT_ITERATIONS, help=f'Number of iterations (default: {DEFAULT_ITERATIONS})')
+    parser.add_argument('--prompt', '-p', type=str, default=DEFAULT_PROMPT, help=f'Prompt to use (default: "{DEFAULT_PROMPT}")')
+    args = parser.parse_args()
+    
+    # Check API key
+    if not os.environ.get('OPENAI_API_KEY'):
+        print("ERROR: Set OPENAI_API_KEY environment variable")
+        exit(1)
+    
+    model = args.model
+    iterations = args.iterations
+    prompt = args.prompt
+    
+    print('=' * 60)
+    print('PraisonAI Agents - Real Execution Benchmark')
+    print('=' * 60)
+    print(f'\nModel: {model}')
+    print(f'Iterations: {iterations}')
+    print(f'Prompt: "{prompt}"')
+    
+    versions = get_package_versions()
+    print('\nPackage versions:')
+    for pkg, ver in versions.items():
+        print(f'  {pkg}: {ver}')
+    
+    results = {}
+    
+    # Run benchmarks
+    try:
+        results['PraisonAI'] = benchmark_praisonai(model, iterations, prompt)
+    except Exception as e:
+        print(f"PraisonAI error: {e}")
+    
+    try:
+        results['PraisonAI (LiteLLM)'] = benchmark_praisonai_litellm(model, iterations, prompt)
+    except Exception as e:
+        print(f"PraisonAI (LiteLLM) error: {e}")
+    
+    try:
+        results['Agno'] = benchmark_agno(model, iterations, prompt)
+    except Exception as e:
+        print(f"Agno error: {e}")
+    
+    try:
+        results['CrewAI'] = benchmark_crewai(model, iterations, prompt)
+    except Exception as e:
+        print(f"CrewAI error: {e}")
+    
+    # Summary
+    print('\n' + '=' * 60)
+    print('SUMMARY')
+    print('=' * 60)
+    
+    if results:
+        sorted_results = sorted(results.items(), key=lambda x: x[1])
+        fastest = sorted_results[0][1]
+        
+        print(f"\n{'Framework':<25} {'Avg Time':<12} {'Relative':<10}")
+        print('-' * 47)
+        
+        for name, avg in sorted_results:
+            ratio = avg / fastest
+            print(f'{name:<25} {avg:<12.2f}s {ratio:.2f}x')
+    
+    print('\n' + '=' * 60)
+    
+    # Save results (unless --no-save)
+    if results and not args.no_save:
+        save_results(results, model, iterations, prompt)
+    elif args.no_save:
+        print('\nResults not saved (--no-save flag used)')

src/praisonai-agents/benchmarks/performance_benchmark.py

@@ -0,0 +1,523 @@
+"""
+PraisonAI Agents - Comprehensive Performance Benchmark
+
+Measures agent instantiation time and memory footprint across
+multiple AI agent frameworks using industry-standard methodology.
+
+Usage:
+    python benchmarks/performance_benchmark.py
+"""
+
+import gc
+import time
+import tracemalloc
+import statistics
+from typing import Literal, List, Callable
+from dataclasses import dataclass, field
+
+
+@dataclass
+class PerformanceResult:
+    """Performance measurement results"""
+    run_times: List[float] = field(default_factory=list)
+    memory_usages: List[float] = field(default_factory=list)
+    
+    @property
+    def avg_run_time(self) -> float:
+        return statistics.mean(self.run_times) if self.run_times else 0
+    
+    @property
+    def avg_memory_usage(self) -> float:
+        return statistics.mean(self.memory_usages) if self.memory_usages else 0
+    
+    @property
+    def min_run_time(self) -> float:
+        return min(self.run_times) if self.run_times else 0
+    
+    @property
+    def max_run_time(self) -> float:
+        return max(self.run_times) if self.run_times else 0
+    
+    @property
+    def std_dev_run_time(self) -> float:
+        return statistics.stdev(self.run_times) if len(self.run_times) > 1 else 0
+    
+    @property
+    def median_run_time(self) -> float:
+        return statistics.median(self.run_times) if self.run_times else 0
+
+
+def compute_tracemalloc_baseline(samples: int = 3) -> float:
+    """Compute baseline memory usage for tracemalloc"""
+    def empty_func():
+        return
+    
+    results = []
+    for _ in range(samples):
+        gc.collect()
+        tracemalloc.start()
+        empty_func()
+        _, peak = tracemalloc.get_traced_memory()
+        tracemalloc.stop()
+        results.append(peak / 1024 / 1024)  # Convert to MiB
+    
+    return sum(results) / len(results) if results else 0
+
+
+def measure_time(func: Callable, iterations: int = 1000, warmup: int = 10) -> List[float]:
+    """Measure execution time for a function"""
+    # Warmup runs
+    for _ in range(warmup):
+        func()
+    
+    # Measured runs
+    times = []
+    for _ in range(iterations):
+        start = time.perf_counter()
+        func()
+        end = time.perf_counter()
+        times.append(end - start)
+    
+    return times
+
+
+def measure_memory(func: Callable, iterations: int = 1000, warmup: int = 10) -> List[float]:
+    """Measure memory usage for a function"""
+    baseline = compute_tracemalloc_baseline()
+    
+    # Warmup runs
+    for _ in range(warmup):
+        func()
+    
+    # Measured runs
+    memory_usages = []
+    for _ in range(iterations):
+        gc.collect()
+        tracemalloc.start()
+        func()
+        _, peak = tracemalloc.get_traced_memory()
+        tracemalloc.stop()
+        
+        peak_mib = peak / 1024 / 1024
+        adjusted = max(0, peak_mib - baseline)
+        memory_usages.append(adjusted)
+    
+    return memory_usages
+
+
+def run_benchmark(func: Callable, name: str, iterations: int = 1000, warmup: int = 10) -> PerformanceResult:
+    """Run full benchmark for a function"""
+    print(f"\n{'='*60}")
+    print(f"Benchmarking: {name}")
+    print(f"{'='*60}")
+    print(f"Warmup runs: {warmup}")
+    print(f"Measured iterations: {iterations}")
+    
+    print("\nMeasuring runtime...")
+    run_times = measure_time(func, iterations, warmup)
+    
+    print("Measuring memory...")
+    memory_usages = measure_memory(func, iterations, warmup)
+    
+    result = PerformanceResult(run_times=run_times, memory_usages=memory_usages)
+    
+    print(f"\nResults for {name}:")
+    print(f"  Average Time: {result.avg_run_time:.6f} seconds ({result.avg_run_time * 1_000_000:.2f} μs)")
+    print(f"  Min Time:     {result.min_run_time:.6f} seconds")
+    print(f"  Max Time:     {result.max_run_time:.6f} seconds")
+    print(f"  Std Dev:      {result.std_dev_run_time:.6f} seconds")
+    print(f"  Median Time:  {result.median_run_time:.6f} seconds")
+    print(f"  Avg Memory:   {result.avg_memory_usage:.6f} MiB ({result.avg_memory_usage * 1024:.2f} KiB)")
+    
+    return result
+
+
+# ============================================================================
+# Framework Imports
+# ============================================================================
+from praisonaiagents import Agent as PraisonAgent
+from agno.agent import Agent as AgnoAgent
+from agno.models.openai import OpenAIChat
+
+# Additional frameworks for comparison
+try:
+    from pydantic_ai import Agent as PydanticAgent
+    PYDANTIC_AVAILABLE = True
+except ImportError:
+    PYDANTIC_AVAILABLE = False
+
+try:
+    from crewai.agent import Agent as CrewAgent
+    from crewai.tools import tool as crewai_tool
+    CREWAI_AVAILABLE = True
+except ImportError:
+    CREWAI_AVAILABLE = False
+
+try:
+    from langchain_core.tools import tool as langchain_tool
+    from langchain_openai import ChatOpenAI
+    from langgraph.prebuilt import create_react_agent
+    LANGGRAPH_AVAILABLE = True
+except ImportError:
+    LANGGRAPH_AVAILABLE = False
+
+try:
+    from agents import Agent as OpenAIAgent, function_tool
+    OPENAI_AGENTS_AVAILABLE = True
+except ImportError:
+    OPENAI_AGENTS_AVAILABLE = False
+
+
+def get_weather(city: Literal["nyc", "sf"]):
+    """Use this to get weather information."""
+    if city == "nyc":
+        return "It might be cloudy in nyc"
+    elif city == "sf":
+        return "It's always sunny in sf"
+
+
+tools = [get_weather]
+
+
+def instantiate_praisonai_agent():
+    """Instantiate a PraisonAI agent with one tool."""
+    return PraisonAgent(
+        name="Test Agent",
+        instructions="Be concise, reply with one sentence.",
+        llm="gpt-4o-mini",
+        tools=tools,
+        verbose=False
+    )
+
+
+def instantiate_praisonai_litellm_agent():
+    """Instantiate a PraisonAI agent with LiteLLM backend."""
+    return PraisonAgent(
+        name="Test Agent",
+        instructions="Be concise, reply with one sentence.",
+        llm="openai/gpt-4o-mini",
+        tools=tools,
+        verbose=False
+    )
+
+
+def instantiate_agno_agent():
+    """Instantiate an Agno agent with one tool."""
+    return AgnoAgent(model=OpenAIChat(id="gpt-4o-mini"), tools=tools)
+
+
+def instantiate_pydantic_agent():
+    """Instantiate a PydanticAI agent with one tool."""
+    agent = PydanticAgent("openai:gpt-4o-mini", system_prompt="Be concise, reply with one sentence.")
+    @agent.tool_plain
+    def get_weather_pydantic(city: Literal["nyc", "sf"]):
+        """Use this to get weather information."""
+        if city == "nyc":
+            return "It might be cloudy in nyc"
+        elif city == "sf":
+            return "It's always sunny in sf"
+    return agent
+
+
+def instantiate_openai_agents():
+    """Instantiate an OpenAI Agents SDK agent with one tool."""
+    return OpenAIAgent(
+        name="Test agent",
+        instructions="Be concise, reply with one sentence.",
+        model="gpt-4o-mini",
+        tools=[function_tool(get_weather)],
+    )
+
+
+def instantiate_langgraph_agent():
+    """Instantiate a LangGraph agent with one tool."""
+    @langchain_tool
+    def get_weather_lg(city: Literal["nyc", "sf"]):
+        """Use this to get weather information."""
+        if city == "nyc":
+            return "It might be cloudy in nyc"
+        elif city == "sf":
+            return "It's always sunny in sf"
+    return create_react_agent(model=ChatOpenAI(model="gpt-4o-mini"), tools=[get_weather_lg])
+
+
+def instantiate_crewai_agent():
+    """Instantiate a CrewAI agent with one tool."""
+    @crewai_tool("Weather Tool")
+    def get_weather_crew(city: Literal["nyc", "sf"]):
+        """Use this to get weather information."""
+        if city == "nyc":
+            return "It might be cloudy in nyc"
+        elif city == "sf":
+            return "It's always sunny in sf"
+    return CrewAgent(
+        llm="gpt-4o-mini",
+        role="Test Agent",
+        goal="Be concise, reply with one sentence.",
+        tools=[get_weather_crew],
+        backstory="Test",
+    )
+
+
+def get_package_versions():
+    """Get version numbers for all benchmarked packages."""
+    from importlib.metadata import version as get_version
+    
+    packages = {
+        'PraisonAI': 'praisonaiagents',
+        'Agno': 'agno',
+        'PydanticAI': 'pydantic-ai',
+        'OpenAI Agents SDK': 'openai-agents',
+        'LangGraph': 'langgraph',
+        'CrewAI': 'crewai'
+    }
+    
+    versions = {}
+    for display_name, pkg_name in packages.items():
+        try:
+            versions[display_name] = get_version(pkg_name)
+        except Exception:
+            versions[display_name] = 'not installed'
+    
+    return versions
+
+
+def save_benchmark_results(results: dict, baseline):
+    """Save benchmark results to BENCHMARK_RESULTS.md and update README.md"""
+    import os
+    import re
+    from datetime import datetime
+    
+    benchmarks_dir = os.path.dirname(__file__)
+    filepath = os.path.join(benchmarks_dir, 'BENCHMARK_RESULTS.md')
+    
+    # Get package versions
+    versions = get_package_versions()
+    
+    display_names = {
+        'agno': 'Agno',
+        'praisonai': 'PraisonAI',
+        'praisonai_litellm': 'PraisonAI (LiteLLM)',
+        'pydantic': 'PydanticAI',
+        'openai_agents': 'OpenAI Agents SDK',
+        'langgraph': 'LangGraph',
+        'crewai': 'CrewAI'
+    }
+    
+    # Build table rows
+    table_rows = []
+    sorted_results = sorted(results.items(), key=lambda x: x[1].avg_run_time)
+    
+    for name, result in sorted_results:
+        display_name = display_names.get(name, name)
+        avg_us = result.avg_run_time * 1e6
+        ratio = result.avg_run_time / baseline.avg_run_time if baseline.avg_run_time > 0 else 0
+        
+        if name == 'praisonai':
+            table_rows.append(f'| **{display_name}** | **{avg_us:.2f}** | **1.00x (fastest)** |')
+        else:
+            avg_str = f'{avg_us:,.2f}' if avg_us >= 1000 else f'{avg_us:.2f}'
+            ratio_str = f'{ratio:,.0f}x' if ratio >= 100 else f'{ratio:.2f}x'
+            table_rows.append(f'| {display_name} | {avg_str} | {ratio_str} |')
+    
+    # Save to BENCHMARK_RESULTS.md
+    with open(filepath, 'w') as f:
+        f.write('# PraisonAI Agents - Benchmark Results\n\n')
+        f.write(f'**Generated:** {datetime.now().strftime("%Y-%m-%d %H:%M:%S")}\n')
+        f.write('**Methodology:** 10 warmup runs + 1000 measured iterations\n\n')
+        f.write('## Agent Instantiation Time\n\n')
+        f.write('| Framework | Avg Time (μs) | Relative |\n')
+        f.write('|-----------|---------------|----------|\n')
+        f.write('\n'.join(table_rows) + '\n')
+        f.write('\n## Package Versions\n\n')
+        f.write('| Package | Version |\n')
+        f.write('|---------|--------|\n')
+        for pkg, ver in versions.items():
+            f.write(f'| {pkg} | {ver} |\n')
+        f.write('\n## How to Reproduce\n\n')
+        f.write('```bash\n')
+        f.write('cd praisonai-agents\n')
+        f.write('python benchmarks/performance_benchmark.py\n')
+        f.write('```\n')
+    
+    print(f'\nResults saved to: {filepath}')
+    
+    # Update README.md
+    readme_path = os.path.join(benchmarks_dir, '..', '..', '..', 'README.md')
+    if os.path.exists(readme_path):
+        with open(readme_path, 'r') as f:
+            content = f.read()
+        
+        new_table = '''| Framework | Avg Time (μs) | Relative |
+|-----------|---------------|----------|
+''' + '\n'.join(table_rows)
+        
+        pattern = r'(\| Framework \| Avg Time \(μs\) \| Relative \|\n\|[-|]+\|\n)(\|[^\n]+\|\n)+'
+        
+        if re.search(pattern, content):
+            content = re.sub(pattern, new_table + '\n', content)
+            
+            with open(readme_path, 'w') as f:
+                f.write(content)
+            
+            print(f'README.md updated: {readme_path}')
+
+
+if __name__ == "__main__":
+    print("="*70)
+    print("PraisonAI Agents - Comprehensive Performance Benchmark")
+    print("="*70)
+    print("\nThis benchmark measures:")
+    print("  1. Agent instantiation time (how fast an agent object is created)")
+    print("  2. Memory footprint (how much memory an agent uses)")
+    print("\nMethodology:")
+    print("  - 10 warmup runs (not measured)")
+    print("  - 1000 measured iterations")
+    print("  - tracemalloc for memory measurement")
+    print("  - Baseline subtraction for accuracy")
+    print("\nFrameworks available:")
+    print("  - PraisonAI: ✅")
+    print(f"  - Agno: ✅")
+    print(f"  - PydanticAI: {'✅' if PYDANTIC_AVAILABLE else '❌'}")
+    print(f"  - OpenAI Agents SDK: {'✅' if OPENAI_AGENTS_AVAILABLE else '❌'}")
+    print(f"  - LangGraph: {'✅' if LANGGRAPH_AVAILABLE else '❌'}")
+    print(f"  - CrewAI: {'✅' if CREWAI_AVAILABLE else '❌'}")
+    
+    results = {}
+    
+    # Test PraisonAI first
+    try:
+        print("\n" + "="*70)
+        print("Testing PraisonAI Agents")
+        print("="*70)
+        results['praisonai'] = run_benchmark(
+            instantiate_praisonai_agent,
+            "PraisonAI",
+            iterations=1000,
+            warmup=10
+        )
+    except Exception as e:
+        print(f"Error testing PraisonAI: {e}")
+    
+    try:
+        print("\n" + "="*70)
+        print("Testing PraisonAI Agents (LiteLLM)")
+        print("="*70)
+        results['praisonai_litellm'] = run_benchmark(
+            instantiate_praisonai_litellm_agent,
+            "PraisonAI (LiteLLM)",
+            iterations=1000,
+            warmup=10
+        )
+    except Exception as e:
+        print(f"Error testing PraisonAI (LiteLLM): {e}")
+    
+    # Test Agno
+    try:
+        print("\n" + "="*70)
+        print("Testing Agno")
+        print("="*70)
+        results['agno'] = run_benchmark(
+            instantiate_agno_agent,
+            "Agno",
+            iterations=1000,
+            warmup=10
+        )
+    except Exception as e:
+        print(f"Error testing Agno: {e}")
+    
+    # Test PydanticAI
+    if PYDANTIC_AVAILABLE:
+        try:
+            print("\n" + "="*70)
+            print("Testing PydanticAI")
+            print("="*70)
+            results['pydantic'] = run_benchmark(
+                instantiate_pydantic_agent,
+                "PydanticAI Agent Instantiation",
+                iterations=1000,
+                warmup=10
+            )
+        except Exception as e:
+            print(f"Error testing PydanticAI: {e}")
+    
+    # Test OpenAI Agents SDK
+    if OPENAI_AGENTS_AVAILABLE:
+        try:
+            print("\n" + "="*70)
+            print("Testing OpenAI Agents SDK")
+            print("="*70)
+            results['openai_agents'] = run_benchmark(
+                instantiate_openai_agents,
+                "OpenAI Agents SDK Instantiation",
+                iterations=1000,
+                warmup=10
+            )
+        except Exception as e:
+            print(f"Error testing OpenAI Agents SDK: {e}")
+    
+    # Test LangGraph
+    if LANGGRAPH_AVAILABLE:
+        try:
+            print("\n" + "="*70)
+            print("Testing LangGraph")
+            print("="*70)
+            results['langgraph'] = run_benchmark(
+                instantiate_langgraph_agent,
+                "LangGraph Agent Instantiation",
+                iterations=1000,
+                warmup=10
+            )
+        except Exception as e:
+            print(f"Error testing LangGraph: {e}")
+    
+    # Test CrewAI
+    if CREWAI_AVAILABLE:
+        try:
+            print("\n" + "="*70)
+            print("Testing CrewAI")
+            print("="*70)
+            results['crewai'] = run_benchmark(
+                instantiate_crewai_agent,
+                "CrewAI Agent Instantiation",
+                iterations=1000,
+                warmup=10
+            )
+        except Exception as e:
+            print(f"Error testing CrewAI: {e}")
+    
+    # Final Comparison Table
+    print("\n" + "="*70)
+    print("RESULTS SUMMARY")
+    print("="*70)
+    
+    if 'praisonai' in results:
+        baseline = results['praisonai']
+        
+        print(f"\n{'Framework':<25} {'Avg Time (μs)':<15} {'Avg Memory (KiB)':<18} {'Relative':<15}")
+        print("-" * 73)
+        
+        # Sort by time
+        sorted_results = sorted(results.items(), key=lambda x: x[1].avg_run_time)
+        
+        for name, result in sorted_results:
+            time_ratio = result.avg_run_time / baseline.avg_run_time if baseline.avg_run_time > 0 else 0
+            
+            time_str = f"{time_ratio:.2f}x"
+            
+            display_name = {
+                'agno': 'Agno',
+                'praisonai': 'PraisonAI',
+                'praisonai_litellm': 'PraisonAI (LiteLLM)',
+                'pydantic': 'PydanticAI',
+                'openai_agents': 'OpenAI Agents SDK',
+                'langgraph': 'LangGraph',
+                'crewai': 'CrewAI'
+            }.get(name, name)
+            
+            print(f"{display_name:<25} {result.avg_run_time*1e6:<15.2f} {result.avg_memory_usage*1024:<18.2f} {time_str:<15}")
+        
+        print("\n" + "="*70)
+        
+        # Save results to files
+        save_benchmark_results(results, baseline)

src/praisonai-agents/benchmarks/real_benchmark.py

@@ -0,0 +1,481 @@
+#!/usr/bin/env python3
+"""
+PraisonAI Agents - Real-World Execution Benchmark
+
+Benchmarks actual agent execution with LLM calls across frameworks.
+Tests single agent, 2-agent, and 3-agent workflows.
+
+Usage:
+    python benchmarks/real_benchmark.py
+"""
+
+import time
+import statistics
+import os
+from typing import Literal, List
+from dataclasses import dataclass, field
+from importlib.metadata import version as get_version
+
+# Ensure API key is set
+if not os.environ.get('OPENAI_API_KEY'):
+    print("ERROR: OPENAI_API_KEY environment variable not set")
+    exit(1)
+
+
+@dataclass
+class ExecutionResult:
+    """Execution benchmark results"""
+    times: List[float] = field(default_factory=list)
+    
+    @property
+    def avg_time(self) -> float:
+        return statistics.mean(self.times) if self.times else 0
+    
+    @property
+    def min_time(self) -> float:
+        return min(self.times) if self.times else 0
+    
+    @property
+    def max_time(self) -> float:
+        return max(self.times) if self.times else 0
+    
+    @property
+    def std_dev(self) -> float:
+        return statistics.stdev(self.times) if len(self.times) > 1 else 0
+
+
+def get_package_versions():
+    """Get version numbers for benchmarked packages."""
+    packages = {
+        'PraisonAI': 'praisonaiagents',
+        'Agno': 'agno',
+    }
+    versions = {}
+    for display_name, pkg_name in packages.items():
+        try:
+            versions[display_name] = get_version(pkg_name)
+        except Exception:
+            versions[display_name] = 'not installed'
+    return versions
+
+
+# ============================================================================
+# Tool Definitions
+# ============================================================================
+
+def get_weather(city: Literal["nyc", "sf"]) -> str:
+    """Get weather information for a city."""
+    if city == "nyc":
+        return "New York: Cloudy, 45°F, 60% humidity"
+    return "San Francisco: Sunny, 72°F, 40% humidity"
+
+
+def get_time(timezone: Literal["est", "pst"]) -> str:
+    """Get current time in a timezone."""
+    if timezone == "est":
+        return "Eastern Time: 2:30 PM"
+    return "Pacific Time: 11:30 AM"
+
+
+def calculate(expression: str) -> str:
+    """Calculate a math expression."""
+    try:
+        result = eval(expression)
+        return f"Result: {result}"
+    except Exception:
+        return "Error: Invalid expression"
+
+
+# ============================================================================
+# PraisonAI Benchmarks
+# ============================================================================
+
+def benchmark_praisonai_single(iterations: int = 3) -> ExecutionResult:
+    """Benchmark single PraisonAI agent execution."""
+    from praisonaiagents import Agent
+    
+    result = ExecutionResult()
+    
+    for i in range(iterations):
+        print(f"  PraisonAI single agent iteration {i+1}/{iterations}...")
+        
+        agent = Agent(
+            name="Assistant",
+            instructions="You are a helpful assistant. Be very brief.",
+            llm="gpt-4o-mini",
+            tools=[get_weather],
+            verbose=False
+        )
+        
+        start = time.perf_counter()
+        response = agent.start("What's the weather in NYC? Reply in one sentence.")
+        elapsed = time.perf_counter() - start
+        
+        result.times.append(elapsed)
+        print(f"    Time: {elapsed:.2f}s")
+    
+    return result
+
+
+def benchmark_praisonai_two_agents(iterations: int = 3) -> ExecutionResult:
+    """Benchmark 2 PraisonAI agents working together."""
+    from praisonaiagents import Agent, PraisonAIAgents
+    
+    result = ExecutionResult()
+    
+    for i in range(iterations):
+        print(f"  PraisonAI 2-agent iteration {i+1}/{iterations}...")
+        
+        researcher = Agent(
+            name="Researcher",
+            role="Research Assistant",
+            goal="Gather weather information",
+            instructions="Get weather data and pass to reporter. Be brief.",
+            llm="gpt-4o-mini",
+            tools=[get_weather],
+            verbose=False
+        )
+        
+        reporter = Agent(
+            name="Reporter",
+            role="News Reporter",
+            goal="Create weather report",
+            instructions="Create a one-sentence weather summary.",
+            llm="gpt-4o-mini",
+            verbose=False
+        )
+        
+        agents = PraisonAIAgents(
+            agents=[researcher, reporter],
+            process="sequential",
+            verbose=False
+        )
+        
+        start = time.perf_counter()
+        response = agents.start("Get NYC weather and create a brief report.")
+        elapsed = time.perf_counter() - start
+        
+        result.times.append(elapsed)
+        print(f"    Time: {elapsed:.2f}s")
+    
+    return result
+
+
+def benchmark_praisonai_three_agents(iterations: int = 3) -> ExecutionResult:
+    """Benchmark 3 PraisonAI agents working together."""
+    from praisonaiagents import Agent, PraisonAIAgents
+    
+    result = ExecutionResult()
+    
+    for i in range(iterations):
+        print(f"  PraisonAI 3-agent iteration {i+1}/{iterations}...")
+        
+        weather_agent = Agent(
+            name="WeatherAgent",
+            role="Weather Specialist",
+            goal="Get weather data",
+            instructions="Fetch weather information. Be brief.",
+            llm="gpt-4o-mini",
+            tools=[get_weather],
+            verbose=False
+        )
+        
+        time_agent = Agent(
+            name="TimeAgent",
+            role="Time Specialist",
+            goal="Get time information",
+            instructions="Fetch timezone information. Be brief.",
+            llm="gpt-4o-mini",
+            tools=[get_time],
+            verbose=False
+        )
+        
+        summarizer = Agent(
+            name="Summarizer",
+            role="Summary Writer",
+            goal="Combine information",
+            instructions="Create a one-sentence summary combining weather and time.",
+            llm="gpt-4o-mini",
+            verbose=False
+        )
+        
+        agents = PraisonAIAgents(
+            agents=[weather_agent, time_agent, summarizer],
+            process="sequential",
+            verbose=False
+        )
+        
+        start = time.perf_counter()
+        response = agents.start("Get NYC weather and EST time, then summarize.")
+        elapsed = time.perf_counter() - start
+        
+        result.times.append(elapsed)
+        print(f"    Time: {elapsed:.2f}s")
+    
+    return result
+
+
+# ============================================================================
+# Agno Benchmarks
+# ============================================================================
+
+def benchmark_agno_single(iterations: int = 3) -> ExecutionResult:
+    """Benchmark single Agno agent execution."""
+    from agno.agent import Agent
+    from agno.models.openai import OpenAIChat
+    
+    result = ExecutionResult()
+    
+    for i in range(iterations):
+        print(f"  Agno single agent iteration {i+1}/{iterations}...")
+        
+        agent = Agent(
+            model=OpenAIChat(id="gpt-4o-mini"),
+            instructions=["You are a helpful assistant. Be very brief."],
+            tools=[get_weather],
+        )
+        
+        start = time.perf_counter()
+        response = agent.run("What's the weather in NYC? Reply in one sentence.")
+        elapsed = time.perf_counter() - start
+        
+        result.times.append(elapsed)
+        print(f"    Time: {elapsed:.2f}s")
+    
+    return result
+
+
+def benchmark_agno_two_agents(iterations: int = 3) -> ExecutionResult:
+    """Benchmark 2 Agno agents working together (using Team)."""
+    from agno.agent import Agent
+    from agno.models.openai import OpenAIChat
+    from agno.team import Team
+    
+    result = ExecutionResult()
+    
+    for i in range(iterations):
+        print(f"  Agno 2-agent iteration {i+1}/{iterations}...")
+        
+        researcher = Agent(
+            name="Researcher",
+            model=OpenAIChat(id="gpt-4o-mini"),
+            instructions=["Get weather data. Be brief."],
+            tools=[get_weather],
+        )
+        
+        reporter = Agent(
+            name="Reporter",
+            model=OpenAIChat(id="gpt-4o-mini"),
+            instructions=["Create a one-sentence weather summary."],
+        )
+        
+        team = Team(
+            members=[researcher, reporter],
+            model=OpenAIChat(id="gpt-4o-mini"),
+        )
+        
+        start = time.perf_counter()
+        response = team.run("Get NYC weather and create a brief report.")
+        elapsed = time.perf_counter() - start
+        
+        result.times.append(elapsed)
+        print(f"    Time: {elapsed:.2f}s")
+    
+    return result
+
+
+def benchmark_agno_three_agents(iterations: int = 3) -> ExecutionResult:
+    """Benchmark 3 Agno agents working together."""
+    from agno.agent import Agent
+    from agno.models.openai import OpenAIChat
+    from agno.team import Team
+    
+    result = ExecutionResult()
+    
+    for i in range(iterations):
+        print(f"  Agno 3-agent iteration {i+1}/{iterations}...")
+        
+        weather_agent = Agent(
+            name="WeatherAgent",
+            model=OpenAIChat(id="gpt-4o-mini"),
+            instructions=["Fetch weather information. Be brief."],
+            tools=[get_weather],
+        )
+        
+        time_agent = Agent(
+            name="TimeAgent",
+            model=OpenAIChat(id="gpt-4o-mini"),
+            instructions=["Fetch timezone information. Be brief."],
+            tools=[get_time],
+        )
+        
+        summarizer = Agent(
+            name="Summarizer",
+            model=OpenAIChat(id="gpt-4o-mini"),
+            instructions=["Create a one-sentence summary combining weather and time."],
+        )
+        
+        team = Team(
+            members=[weather_agent, time_agent, summarizer],
+            model=OpenAIChat(id="gpt-4o-mini"),
+        )
+        
+        start = time.perf_counter()
+        response = team.run("Get NYC weather and EST time, then summarize.")
+        elapsed = time.perf_counter() - start
+        
+        result.times.append(elapsed)
+        print(f"    Time: {elapsed:.2f}s")
+    
+    return result
+
+
+# ============================================================================
+# Save Results
+# ============================================================================
+
+def save_results(results: dict, versions: dict):
+    """Save benchmark results to file."""
+    import os
+    from datetime import datetime
+    
+    filepath = os.path.join(os.path.dirname(__file__), 'REAL_BENCHMARK_RESULTS.md')
+    
+    with open(filepath, 'w') as f:
+        f.write('# PraisonAI Agents - Real-World Execution Benchmark\n\n')
+        f.write(f'**Generated:** {datetime.now().strftime("%Y-%m-%d %H:%M:%S")}\n')
+        f.write('**Model:** gpt-4o-mini\n')
+        f.write('**Iterations:** 3 per test\n\n')
+        
+        f.write('## Results Summary\n\n')
+        f.write('| Test | PraisonAI (avg) | Agno (avg) | Difference |\n')
+        f.write('|------|-----------------|------------|------------|\n')
+        
+        tests = ['single', 'two_agents', 'three_agents']
+        test_names = ['Single Agent', '2 Agents', '3 Agents']
+        
+        for test, name in zip(tests, test_names):
+            praison_key = f'praisonai_{test}'
+            agno_key = f'agno_{test}'
+            
+            if praison_key in results and agno_key in results:
+                praison_avg = results[praison_key].avg_time
+                agno_avg = results[agno_key].avg_time
+                
+                if praison_avg < agno_avg:
+                    diff = f'PraisonAI {agno_avg/praison_avg:.2f}x faster'
+                else:
+                    diff = f'Agno {praison_avg/agno_avg:.2f}x faster'
+                
+                f.write(f'| {name} | {praison_avg:.2f}s | {agno_avg:.2f}s | {diff} |\n')
+        
+        f.write('\n## Detailed Results\n\n')
+        
+        for key, result in results.items():
+            f.write(f'### {key}\n')
+            f.write(f'- Average: {result.avg_time:.2f}s\n')
+            f.write(f'- Min: {result.min_time:.2f}s\n')
+            f.write(f'- Max: {result.max_time:.2f}s\n')
+            f.write(f'- Std Dev: {result.std_dev:.2f}s\n')
+            f.write(f'- Times: {[f"{t:.2f}s" for t in result.times]}\n\n')
+        
+        f.write('## Package Versions\n\n')
+        f.write('| Package | Version |\n')
+        f.write('|---------|--------|\n')
+        for pkg, ver in versions.items():
+            f.write(f'| {pkg} | {ver} |\n')
+        
+        f.write('\n## How to Reproduce\n\n')
+        f.write('```bash\n')
+        f.write('export OPENAI_API_KEY=your_key\n')
+        f.write('cd praisonai-agents\n')
+        f.write('python benchmarks/real_benchmark.py\n')
+        f.write('```\n')
+    
+    print(f'\nResults saved to: {filepath}')
+
+
+# ============================================================================
+# Main
+# ============================================================================
+
+if __name__ == "__main__":
+    print("="*70)
+    print("PraisonAI Agents - Real-World Execution Benchmark")
+    print("="*70)
+    print("\nThis benchmark measures actual agent execution time including:")
+    print("  - LLM API calls")
+    print("  - Tool execution")
+    print("  - Multi-agent coordination")
+    print("\nModel: gpt-4o-mini")
+    print("Iterations: 3 per test")
+    
+    versions = get_package_versions()
+    print(f"\nPackage versions:")
+    for pkg, ver in versions.items():
+        print(f"  {pkg}: {ver}")
+    
+    results = {}
+    
+    # Single Agent Tests
+    print("\n" + "="*70)
+    print("SINGLE AGENT BENCHMARK")
+    print("="*70)
+    
+    print("\nTesting PraisonAI single agent...")
+    results['praisonai_single'] = benchmark_praisonai_single()
+    
+    print("\nTesting Agno single agent...")
+    results['agno_single'] = benchmark_agno_single()
+    
+    # 2-Agent Tests
+    print("\n" + "="*70)
+    print("2-AGENT BENCHMARK")
+    print("="*70)
+    
+    print("\nTesting PraisonAI 2 agents...")
+    results['praisonai_two_agents'] = benchmark_praisonai_two_agents()
+    
+    print("\nTesting Agno 2 agents (Team)...")
+    results['agno_two_agents'] = benchmark_agno_two_agents()
+    
+    # 3-Agent Tests
+    print("\n" + "="*70)
+    print("3-AGENT BENCHMARK")
+    print("="*70)
+    
+    print("\nTesting PraisonAI 3 agents...")
+    results['praisonai_three_agents'] = benchmark_praisonai_three_agents()
+    
+    print("\nTesting Agno 3 agents (Team)...")
+    results['agno_three_agents'] = benchmark_agno_three_agents()
+    
+    # Summary
+    print("\n" + "="*70)
+    print("RESULTS SUMMARY")
+    print("="*70)
+    
+    print(f"\n{'Test':<20} {'PraisonAI (avg)':<18} {'Agno (avg)':<15} {'Difference':<20}")
+    print("-" * 73)
+    
+    tests = [
+        ('Single Agent', 'praisonai_single', 'agno_single'),
+        ('2 Agents', 'praisonai_two_agents', 'agno_two_agents'),
+        ('3 Agents', 'praisonai_three_agents', 'agno_three_agents'),
+    ]
+    
+    for name, praison_key, agno_key in tests:
+        praison_avg = results[praison_key].avg_time
+        agno_avg = results[agno_key].avg_time
+        
+        if praison_avg < agno_avg:
+            diff = f'PraisonAI {agno_avg/praison_avg:.2f}x faster'
+        else:
+            diff = f'Agno {praison_avg/agno_avg:.2f}x faster'
+        
+        print(f"{name:<20} {praison_avg:<18.2f}s {agno_avg:<15.2f}s {diff:<20}")
+    
+    print("\n" + "="*70)
+    
+    # Save results
+    save_results(results, versions)

src/praisonai-agents/benchmarks/response_benchmark.py

@@ -0,0 +1,235 @@
+"""
+PraisonAI Agents - Response Time Benchmark
+
+Measures actual LLM response time with gpt-4o-mini across frameworks.
+
+Usage:
+    python benchmarks/response_benchmark.py
+"""
+
+import time
+import statistics
+import os
+from typing import List, Literal
+from dataclasses import dataclass, field
+
+# Ensure we use gpt-4o-mini
+os.environ['OPENAI_MODEL_NAME'] = 'gpt-4o-mini'
+
+
+@dataclass
+class ResponseResult:
+    """Response time measurement results"""
+    response_times: List[float] = field(default_factory=list)
+    
+    @property
+    def avg_time(self) -> float:
+        return statistics.mean(self.response_times) if self.response_times else 0
+    
+    @property
+    def min_time(self) -> float:
+        return min(self.response_times) if self.response_times else 0
+    
+    @property
+    def max_time(self) -> float:
+        return max(self.response_times) if self.response_times else 0
+    
+    @property
+    def std_dev(self) -> float:
+        return statistics.stdev(self.response_times) if len(self.response_times) > 1 else 0
+
+
+def get_weather(city: Literal["nyc", "sf"]):
+    """Use this to get weather information."""
+    if city == "nyc":
+        return "It might be cloudy in nyc"
+    elif city == "sf":
+        return "It's always sunny in sf"
+
+
+def benchmark_praisonai_response(iterations: int = 5) -> ResponseResult:
+    """Benchmark PraisonAI agent response time"""
+    from praisonaiagents import Agent
+    
+    result = ResponseResult()
+    
+    for i in range(iterations):
+        print(f"  PraisonAI iteration {i+1}/{iterations}...")
+        
+        # Create fresh agent each time
+        agent = Agent(
+            name="Test Agent",
+            instructions="Be concise, reply with one sentence.",
+            llm="gpt-4o-mini",
+            verbose=False
+        )
+        
+        start = time.perf_counter()
+        response = agent.start("What is the capital of France?")
+        end = time.perf_counter()
+        
+        result.response_times.append(end - start)
+        print(f"    Response time: {end - start:.3f}s")
+    
+    return result
+
+
+def benchmark_agno_response(iterations: int = 5) -> ResponseResult:
+    """Benchmark Agno agent response time"""
+    from agno.agent import Agent
+    from agno.models.openai import OpenAIChat
+    
+    result = ResponseResult()
+    
+    for i in range(iterations):
+        print(f"  Agno iteration {i+1}/{iterations}...")
+        
+        # Create fresh agent each time
+        agent = Agent(
+            model=OpenAIChat(id="gpt-4o-mini"),
+            system_message="Be concise, reply with one sentence.",
+        )
+        
+        start = time.perf_counter()
+        response = agent.run("What is the capital of France?")
+        end = time.perf_counter()
+        
+        result.response_times.append(end - start)
+        print(f"    Response time: {end - start:.3f}s")
+    
+    return result
+
+
+def benchmark_praisonai_with_tool(iterations: int = 5) -> ResponseResult:
+    """Benchmark PraisonAI agent response time with tool calling"""
+    from praisonaiagents import Agent
+    
+    result = ResponseResult()
+    
+    for i in range(iterations):
+        print(f"  PraisonAI (with tool) iteration {i+1}/{iterations}...")
+        
+        agent = Agent(
+            name="Weather Agent",
+            instructions="You are a weather assistant. Use the get_weather tool to answer questions.",
+            llm="gpt-4o-mini",
+            tools=[get_weather],
+            verbose=False
+        )
+        
+        start = time.perf_counter()
+        response = agent.start("What's the weather in NYC?")
+        end = time.perf_counter()
+        
+        result.response_times.append(end - start)
+        print(f"    Response time: {end - start:.3f}s")
+    
+    return result
+
+
+def benchmark_agno_with_tool(iterations: int = 5) -> ResponseResult:
+    """Benchmark Agno agent response time with tool calling"""
+    from agno.agent import Agent
+    from agno.models.openai import OpenAIChat
+    
+    result = ResponseResult()
+    
+    for i in range(iterations):
+        print(f"  Agno (with tool) iteration {i+1}/{iterations}...")
+        
+        agent = Agent(
+            model=OpenAIChat(id="gpt-4o-mini"),
+            system_message="You are a weather assistant. Use the get_weather tool to answer questions.",
+            tools=[get_weather],
+        )
+        
+        start = time.perf_counter()
+        response = agent.run("What's the weather in NYC?")
+        end = time.perf_counter()
+        
+        result.response_times.append(end - start)
+        print(f"    Response time: {end - start:.3f}s")
+    
+    return result
+
+
+if __name__ == "__main__":
+    print("="*70)
+    print("PraisonAI Agents - Response Time Benchmark")
+    print("="*70)
+    print("\nModel: gpt-4o-mini")
+    print("Measures end-to-end response time including:")
+    print("  - Agent instantiation")
+    print("  - API call to OpenAI")
+    print("  - Response processing")
+    
+    iterations = 5
+    results = {}
+    
+    # Test 1: Simple response (no tools)
+    print("\n" + "="*70)
+    print("TEST 1: Simple Response (No Tools)")
+    print("="*70)
+    
+    print("\nBenchmarking PraisonAI...")
+    try:
+        results['praisonai_simple'] = benchmark_praisonai_response(iterations)
+    except Exception as e:
+        print(f"Error: {e}")
+        import traceback
+        traceback.print_exc()
+    
+    print("\nBenchmarking Agno...")
+    try:
+        results['agno_simple'] = benchmark_agno_response(iterations)
+    except Exception as e:
+        print(f"Error: {e}")
+        import traceback
+        traceback.print_exc()
+    
+    # Test 2: Response with tool calling
+    print("\n" + "="*70)
+    print("TEST 2: Response with Tool Calling")
+    print("="*70)
+    
+    print("\nBenchmarking PraisonAI with tool...")
+    try:
+        results['praisonai_tool'] = benchmark_praisonai_with_tool(iterations)
+    except Exception as e:
+        print(f"Error: {e}")
+        import traceback
+        traceback.print_exc()
+    
+    print("\nBenchmarking Agno with tool...")
+    try:
+        results['agno_tool'] = benchmark_agno_with_tool(iterations)
+    except Exception as e:
+        print(f"Error: {e}")
+        import traceback
+        traceback.print_exc()
+    
+    # Summary
+    print("\n" + "="*70)
+    print("SUMMARY RESULTS")
+    print("="*70)
+    
+    print(f"\n{'Test':<30} {'PraisonAI (avg)':<20} {'Agno (avg)':<20} {'Ratio':<15}")
+    print("-" * 85)
+    
+    if 'praisonai_simple' in results and 'agno_simple' in results:
+        p_time = results['praisonai_simple'].avg_time
+        a_time = results['agno_simple'].avg_time
+        ratio = p_time / a_time if a_time > 0 else float('inf')
+        print(f"{'Simple Response':<30} {p_time:<20.3f}s {a_time:<20.3f}s {ratio:.2f}x")
+    
+    if 'praisonai_tool' in results and 'agno_tool' in results:
+        p_time = results['praisonai_tool'].avg_time
+        a_time = results['agno_tool'].avg_time
+        ratio = p_time / a_time if a_time > 0 else float('inf')
+        print(f"{'With Tool Calling':<30} {p_time:<20.3f}s {a_time:<20.3f}s {ratio:.2f}x")
+    
+    print("\n" + "="*70)
+    print("NOTE")
+    print("="*70)
+    print("\nResponse time is dominated by OpenAI API latency.")
+    print("Agent instantiation overhead is negligible in real-world usage.")

src/praisonai-agents/benchmarks/results.console

@@ -0,0 +1,194 @@
+❯ python benchmarks/performance_benchmark.py
+/Users/praison/praisonai-package/src/praisonai-agents/praisonaiagents/_warning_patch.py:47: UserWarning: Mixing V1 models and V2 models (or constructs, like `TypeAdapter`) is not supported. Please upgrade `Settings` to V2.
+  _original_warn(message, category, stacklevel, source)
+[13:04:15] INFO     [13:04:15]          config.py:50
+                    config.py:50 INFO               
+                    Using config path:              
+                    /Users/praison/.con             
+                    fig/crewai/settings             
+                    .json                           
+======================================================================
+PraisonAI Agents - Comprehensive Performance Benchmark
+======================================================================
+
+This benchmark measures:
+  1. Agent instantiation time (how fast an agent object is created)
+  2. Memory footprint (how much memory an agent uses)
+
+Methodology:
+  - 10 warmup runs (not measured)
+  - 1000 measured iterations
+  - tracemalloc for memory measurement
+  - Baseline subtraction for accuracy
+
+Frameworks available:
+  - PraisonAI: ✅
+  - Agno: ✅
+  - PydanticAI: ✅
+  - OpenAI Agents SDK: ✅
+  - LangGraph: ✅
+  - CrewAI: ✅
+
+======================================================================
+Testing PraisonAI Agents
+======================================================================
+
+============================================================
+Benchmarking: PraisonAI
+============================================================
+Warmup runs: 10
+Measured iterations: 1000
+
+Measuring runtime...
+Measuring memory...
+
+Results for PraisonAI:
+  Average Time: 0.000003 seconds (3.20 μs)
+  Min Time:     0.000003 seconds
+  Max Time:     0.000025 seconds
+  Std Dev:      0.000001 seconds
+  Median Time:  0.000003 seconds
+  Avg Memory:   0.003678 MiB (3.77 KiB)
+
+======================================================================
+Testing PraisonAI Agents (LiteLLM)
+======================================================================
+
+============================================================
+Benchmarking: PraisonAI (LiteLLM)
+============================================================
+Warmup runs: 10
+Measured iterations: 1000
+
+Measuring runtime...
+Measuring memory...
+
+Results for PraisonAI (LiteLLM):
+  Average Time: 0.000007 seconds (7.31 μs)
+  Min Time:     0.000007 seconds
+  Max Time:     0.000060 seconds
+  Std Dev:      0.000002 seconds
+  Median Time:  0.000007 seconds
+  Avg Memory:   0.006419 MiB (6.57 KiB)
+
+======================================================================
+Testing Agno
+======================================================================
+
+============================================================
+Benchmarking: Agno
+============================================================
+Warmup runs: 10
+Measured iterations: 1000
+
+Measuring runtime...
+Measuring memory...
+
+Results for Agno:
+  Average Time: 0.000005 seconds (5.30 μs)
+  Min Time:     0.000005 seconds
+  Max Time:     0.000039 seconds
+  Std Dev:      0.000002 seconds
+  Median Time:  0.000005 seconds
+  Avg Memory:   0.006599 MiB (6.76 KiB)
+
+======================================================================
+Testing PydanticAI
+======================================================================
+
+============================================================
+Benchmarking: PydanticAI Agent Instantiation
+============================================================
+Warmup runs: 10
+Measured iterations: 1000
+
+Measuring runtime...
+Measuring memory...
+
+Results for PydanticAI Agent Instantiation:
+  Average Time: 0.000247 seconds (247.34 μs)
+  Min Time:     0.000218 seconds
+  Max Time:     0.000552 seconds
+  Std Dev:      0.000031 seconds
+  Median Time:  0.000237 seconds
+  Avg Memory:   0.028486 MiB (29.17 KiB)
+
+======================================================================
+Testing OpenAI Agents SDK
+======================================================================
+
+============================================================
+Benchmarking: OpenAI Agents SDK Instantiation
+============================================================
+Warmup runs: 10
+Measured iterations: 1000
+
+Measuring runtime...
+Measuring memory...
+
+Results for OpenAI Agents SDK Instantiation:
+  Average Time: 0.000282 seconds (282.25 μs)
+  Min Time:     0.000247 seconds
+  Max Time:     0.001110 seconds
+  Std Dev:      0.000055 seconds
+  Median Time:  0.000263 seconds
+  Avg Memory:   0.029032 MiB (29.73 KiB)
+
+======================================================================
+Testing LangGraph
+======================================================================
+
+============================================================
+Benchmarking: LangGraph Agent Instantiation
+============================================================
+Warmup runs: 10
+Measured iterations: 1000
+
+Measuring runtime...
+Measuring memory...
+
+Results for LangGraph Agent Instantiation:
+  Average Time: 0.003200 seconds (3200.15 μs)
+  Min Time:     0.002808 seconds
+  Max Time:     0.116363 seconds
+  Std Dev:      0.003628 seconds
+  Median Time:  0.002972 seconds
+  Avg Memory:   0.207509 MiB (212.49 KiB)
+
+======================================================================
+Testing CrewAI
+======================================================================
+
+============================================================
+Benchmarking: CrewAI Agent Instantiation
+============================================================
+Warmup runs: 10
+Measured iterations: 1000
+
+Measuring runtime...
+Measuring memory...
+
+Results for CrewAI Agent Instantiation:
+  Average Time: 0.016772 seconds (16771.69 μs)
+  Min Time:     0.014404 seconds
+  Max Time:     0.091841 seconds
+  Std Dev:      0.003445 seconds
+  Median Time:  0.015816 seconds
+  Avg Memory:   0.065050 MiB (66.61 KiB)
+
+======================================================================
+RESULTS SUMMARY
+======================================================================
+
+Framework                 Avg Time (μs)   Avg Memory (KiB)   Relative       
+-------------------------------------------------------------------------
+PraisonAI                 3.20            3.77               1.00x          
+Agno                      5.30            6.76               1.66x          
+PraisonAI (LiteLLM)       7.31            6.57               2.28x          
+PydanticAI                247.34          29.17              77.25x         
+OpenAI Agents SDK         282.25          29.73              88.15x         
+LangGraph                 3200.15         212.49             999.46x        
+CrewAI                    16771.69        66.61              5238.06x       
+
+======================================================================
+❯ 

src/praisonai-agents/benchmarks/simple_benchmark.py

@@ -0,0 +1,243 @@
+#!/usr/bin/env python3
+"""
+PraisonAI Agents - Performance Benchmark
+
+Compares agent instantiation times across popular AI agent frameworks.
+
+Usage:
+    python benchmarks/simple_benchmark.py
+"""
+
+import time
+from typing import Literal
+
+
+ITERATIONS = 100
+
+
+def sample_tool(city: Literal['nyc', 'sf']):
+    """Sample tool for benchmark testing."""
+    if city == 'nyc':
+        return 'cloudy'
+    return 'sunny'
+
+
+def measure_instantiation(create_fn, iterations=ITERATIONS):
+    """Measure average instantiation time in microseconds."""
+    times = []
+    for _ in range(iterations):
+        start = time.perf_counter()
+        create_fn()
+        times.append((time.perf_counter() - start) * 1_000_000)
+    return sum(times) / len(times)
+
+
+def run_benchmark():
+    """Run the benchmark across all available frameworks (without tools)."""
+    results = {}
+    
+    print('=' * 60)
+    print('PraisonAI Agents - Performance Benchmark')
+    print('=' * 60)
+    print(f'\nIterations: {ITERATIONS}')
+    print('Metric: Agent instantiation time (microseconds)\n')
+    
+    # PraisonAI (without tools)
+    print("Testing PraisonAI...")
+    from praisonaiagents import Agent as PraisonAgent
+    
+    results['PraisonAI'] = measure_instantiation(
+        lambda: PraisonAgent(name='Test', llm='gpt-4o-mini', verbose=False)
+    )
+    
+    results['PraisonAI (LiteLLM)'] = measure_instantiation(
+        lambda: PraisonAgent(name='Test', llm='openai/gpt-4o-mini', verbose=False)
+    )
+    
+    # Other frameworks for comparison
+    print("Testing other frameworks...")
+    
+    try:
+        from agno.agent import Agent as AgnoAgent
+        from agno.models.openai import OpenAIChat
+        results['Agno'] = measure_instantiation(
+            lambda: AgnoAgent(model=OpenAIChat(id='gpt-4o-mini'))
+        )
+    except ImportError:
+        pass
+    
+    try:
+        from pydantic_ai import Agent as PydanticAgent
+        results['PydanticAI'] = measure_instantiation(
+            lambda: PydanticAgent('openai:gpt-4o-mini')
+        )
+    except ImportError:
+        pass
+    
+    try:
+        from agents import Agent as OpenAIAgent
+        results['OpenAI Agents SDK'] = measure_instantiation(
+            lambda: OpenAIAgent(name='Test', model='gpt-4o-mini')
+        )
+    except ImportError:
+        pass
+    
+    try:
+        from langchain_core.tools import tool as langchain_tool
+        from langchain_openai import ChatOpenAI
+        from langgraph.prebuilt import create_react_agent
+        
+        @langchain_tool
+        def get_weather_lg(city: Literal['nyc', 'sf']):
+            """Get weather info."""
+            return 'sunny' if city == 'sf' else 'cloudy'
+        
+        results['LangGraph'] = measure_instantiation(
+            lambda: create_react_agent(model=ChatOpenAI(model='gpt-4o-mini'), tools=[get_weather_lg])
+        )
+    except ImportError:
+        pass
+    
+    try:
+        from crewai.agent import Agent as CrewAgent
+        from crewai.tools import tool as crewai_tool
+        
+        @crewai_tool("Weather Tool")
+        def get_weather_crew(city: Literal['nyc', 'sf']):
+            """Get weather info."""
+            return 'sunny' if city == 'sf' else 'cloudy'
+        
+        results['CrewAI'] = measure_instantiation(
+            lambda: CrewAgent(
+                role='Weather Agent',
+                goal='Provide weather info',
+                backstory='A weather expert',
+                tools=[get_weather_crew],
+                verbose=False
+            )
+        )
+    except ImportError:
+        pass
+    
+    # Print results
+    print('\n' + '=' * 60)
+    print('RESULTS')
+    print('=' * 60)
+    
+    baseline = results.get('PraisonAI', 1)
+    print(f"\n{'Framework':<25} {'Avg Time (μs)':<15} {'Relative':<10}")
+    print('-' * 50)
+    
+    for name, avg in sorted(results.items(), key=lambda x: x[1]):
+        ratio = avg / baseline
+        print(f'{name:<25} {avg:<15.2f} {ratio:.2f}x')
+    
+    print('\n' + '=' * 60)
+    return results
+
+
+def get_package_versions():
+    """Get version numbers for benchmarked packages."""
+    from importlib.metadata import version as get_version
+    
+    packages = {
+        'PraisonAI': 'praisonaiagents',
+        'Agno': 'agno',
+        'PydanticAI': 'pydantic-ai',
+        'OpenAI Agents SDK': 'openai-agents',
+        'LangGraph': 'langgraph',
+        'CrewAI': 'crewai'
+    }
+    
+    versions = {}
+    for display_name, pkg_name in packages.items():
+        try:
+            versions[display_name] = get_version(pkg_name)
+        except Exception:
+            versions[display_name] = 'not installed'
+    
+    return versions
+
+
+def save_results(results: dict, filename: str = 'BENCHMARK_RESULTS.md'):
+    """Save benchmark results to a markdown file and update README."""
+    import os
+    import re
+    from datetime import datetime
+    
+    filepath = os.path.join(os.path.dirname(__file__), filename)
+    versions = get_package_versions()
+    
+    # Find the fastest result (baseline)
+    sorted_results = sorted(results.items(), key=lambda x: x[1])
+    fastest_time = sorted_results[0][1] if sorted_results else 1
+    
+    # Build the table rows
+    table_rows = []
+    for name, avg in sorted_results:
+        ratio = avg / fastest_time
+        if ratio == 1.0:
+            table_rows.append(f'| **{name}** | **{avg:.2f}** | **1.00x (fastest)** |')
+        else:
+            avg_str = f'{avg:,.2f}' if avg >= 1000 else f'{avg:.2f}'
+            ratio_str = f'{ratio:,.0f}x' if ratio >= 100 else f'{ratio:.2f}x'
+            table_rows.append(f'| {name} | {avg_str} | {ratio_str} |')
+    
+    # Save to BENCHMARK_RESULTS.md
+    with open(filepath, 'w') as f:
+        f.write('# PraisonAI Agents - Benchmark Results\n\n')
+        f.write(f'**Generated:** {datetime.now().strftime("%Y-%m-%d %H:%M:%S")}\n')
+        f.write(f'**Iterations:** {ITERATIONS}\n')
+        f.write('**Test:** Agent instantiation (without tools)\n\n')
+        f.write('## Results\n\n')
+        f.write('| Framework | Avg Time (μs) | Relative |\n')
+        f.write('|-----------|---------------|----------|\n')
+        f.write('\n'.join(table_rows) + '\n')
+        f.write('\n## Package Versions\n\n')
+        f.write('| Package | Version |\n')
+        f.write('|---------|--------|\n')
+        for pkg, ver in versions.items():
+            f.write(f'| {pkg} | {ver} |\n')
+        f.write('\n## How to Reproduce\n\n')
+        f.write('```bash\n')
+        f.write('cd praisonai-agents\n')
+        f.write('python benchmarks/simple_benchmark.py\n')
+        f.write('```\n')
+    
+    print(f'\nResults saved to: {filepath}')
+    
+    # Also update the main README.md
+    readme_path = os.path.join(os.path.dirname(__file__), '..', '..', '..', 'README.md')
+    if os.path.exists(readme_path):
+        update_readme(readme_path, table_rows)
+    
+    return filepath
+
+
+def update_readme(readme_path: str, table_rows: list):
+    """Update the performance section in README.md with latest results."""
+    import re
+    
+    with open(readme_path, 'r') as f:
+        content = f.read()
+    
+    # Build the new table
+    new_table = '''| Framework | Avg Time (μs) | Relative |
+|-----------|---------------|----------|
+''' + '\n'.join(table_rows)
+    
+    # Pattern to match the performance table
+    pattern = r'(\| Framework \| Avg Time \(μs\) \| Relative \|\n\|[-|]+\|\n)(\|[^\n]+\|\n)+'
+    
+    if re.search(pattern, content):
+        content = re.sub(pattern, new_table + '\n', content)
+        
+        with open(readme_path, 'w') as f:
+            f.write(content)
+        
+        print(f'README.md updated: {readme_path}')
+
+
+if __name__ == '__main__':
+    results = run_benchmark()
+    save_results(results)

src/praisonai-agents/benchmarks/tools_benchmark.py

@@ -0,0 +1,213 @@
+#!/usr/bin/env python3
+"""
+PraisonAI Agents - Tools Benchmark
+
+Compares agent instantiation times WITH TOOLS across frameworks.
+
+Usage:
+    python benchmarks/tools_benchmark.py
+"""
+
+import time
+from typing import Literal
+from importlib.metadata import version as get_version
+
+
+ITERATIONS = 100
+
+
+def sample_tool(city: Literal['nyc', 'sf']):
+    """Sample tool for benchmark testing."""
+    if city == 'nyc':
+        return 'cloudy'
+    return 'sunny'
+
+
+def measure_instantiation(create_fn, iterations=ITERATIONS):
+    """Measure average instantiation time in microseconds."""
+    times = []
+    
+    # Warmup
+    for _ in range(10):
+        create_fn()
+    
+    # Measure
+    for _ in range(iterations):
+        start = time.perf_counter()
+        create_fn()
+        elapsed = time.perf_counter() - start
+        times.append(elapsed * 1_000_000)  # Convert to microseconds
+    
+    return sum(times) / len(times)
+
+
+def run_benchmark():
+    """Run the benchmark across all available frameworks WITH TOOLS."""
+    tools = [sample_tool]
+    results = {}
+    
+    print('=' * 60)
+    print('PraisonAI Agents - Tools Benchmark')
+    print('=' * 60)
+    print(f'\nIterations: {ITERATIONS}')
+    print('Metric: Agent instantiation time WITH TOOLS (microseconds)\n')
+    
+    # PraisonAI with tools
+    print("Testing PraisonAI...")
+    from praisonaiagents import Agent as PraisonAgent
+    
+    results['PraisonAI'] = measure_instantiation(
+        lambda: PraisonAgent(name='Test', llm='gpt-4o-mini', tools=tools, verbose=False)
+    )
+    
+    results['PraisonAI (LiteLLM)'] = measure_instantiation(
+        lambda: PraisonAgent(name='Test', llm='openai/gpt-4o-mini', tools=tools, verbose=False)
+    )
+    
+    # Other frameworks with tools
+    print("Testing other frameworks...")
+    
+    try:
+        from agno.agent import Agent as AgnoAgent
+        from agno.models.openai import OpenAIChat
+        results['Agno'] = measure_instantiation(
+            lambda: AgnoAgent(model=OpenAIChat(id='gpt-4o-mini'), tools=tools)
+        )
+    except ImportError:
+        pass
+    
+    try:
+        from agents import Agent as OpenAIAgent, function_tool
+        results['OpenAI Agents SDK'] = measure_instantiation(
+            lambda: OpenAIAgent(name='Test', model='gpt-4o-mini', tools=[function_tool(sample_tool)])
+        )
+    except ImportError:
+        pass
+    
+    try:
+        from langchain_core.tools import tool as langchain_tool
+        from langchain_openai import ChatOpenAI
+        from langgraph.prebuilt import create_react_agent
+        
+        @langchain_tool
+        def get_weather_lg(city: Literal['nyc', 'sf']):
+            """Get weather info."""
+            return 'sunny' if city == 'sf' else 'cloudy'
+        
+        results['LangGraph'] = measure_instantiation(
+            lambda: create_react_agent(model=ChatOpenAI(model='gpt-4o-mini'), tools=[get_weather_lg])
+        )
+    except ImportError:
+        pass
+    
+    try:
+        from crewai.agent import Agent as CrewAgent
+        from crewai.tools import tool as crewai_tool
+        
+        @crewai_tool("Weather Tool")
+        def get_weather_crew(city: Literal['nyc', 'sf']):
+            """Get weather info."""
+            return 'sunny' if city == 'sf' else 'cloudy'
+        
+        results['CrewAI'] = measure_instantiation(
+            lambda: CrewAgent(
+                role='Weather Agent',
+                goal='Provide weather info',
+                backstory='A weather expert',
+                tools=[get_weather_crew],
+                verbose=False
+            )
+        )
+    except ImportError:
+        pass
+    
+    # Print results
+    print('\n' + '=' * 60)
+    print('RESULTS (WITH TOOLS)')
+    print('=' * 60)
+    
+    sorted_results = sorted(results.items(), key=lambda x: x[1])
+    baseline = sorted_results[0][1] if sorted_results else 1
+    
+    print(f"\n{'Framework':<25} {'Avg Time (μs)':<15} {'Relative':<10}")
+    print('-' * 50)
+    
+    for name, avg in sorted_results:
+        ratio = avg / baseline
+        print(f'{name:<25} {avg:<15.2f} {ratio:.2f}x')
+    
+    print('\n' + '=' * 60)
+    return results
+
+
+def get_package_versions():
+    """Get version numbers for benchmarked packages."""
+    packages = {
+        'PraisonAI': 'praisonaiagents',
+        'Agno': 'agno',
+        'OpenAI Agents SDK': 'openai-agents',
+        'LangGraph': 'langgraph',
+        'CrewAI': 'crewai'
+    }
+    
+    versions = {}
+    for display_name, pkg_name in packages.items():
+        try:
+            versions[display_name] = get_version(pkg_name)
+        except Exception:
+            versions[display_name] = 'not installed'
+    
+    return versions
+
+
+def save_results(results: dict, filename: str = 'TOOLS_BENCHMARK_RESULTS.md'):
+    """Save benchmark results to a markdown file."""
+    import os
+    from datetime import datetime
+    
+    filepath = os.path.join(os.path.dirname(__file__), filename)
+    versions = get_package_versions()
+    
+    # Find the fastest result (baseline)
+    sorted_results = sorted(results.items(), key=lambda x: x[1])
+    fastest_time = sorted_results[0][1] if sorted_results else 1
+    
+    # Build the table rows
+    table_rows = []
+    for name, avg in sorted_results:
+        ratio = avg / fastest_time
+        if ratio == 1.0:
+            table_rows.append(f'| **{name}** | **{avg:.2f}** | **1.00x (fastest)** |')
+        else:
+            avg_str = f'{avg:,.2f}' if avg >= 1000 else f'{avg:.2f}'
+            ratio_str = f'{ratio:,.0f}x' if ratio >= 100 else f'{ratio:.2f}x'
+            table_rows.append(f'| {name} | {avg_str} | {ratio_str} |')
+    
+    # Save to file
+    with open(filepath, 'w') as f:
+        f.write('# PraisonAI Agents - Tools Benchmark Results\n\n')
+        f.write(f'**Generated:** {datetime.now().strftime("%Y-%m-%d %H:%M:%S")}\n')
+        f.write(f'**Iterations:** {ITERATIONS}\n')
+        f.write('**Test:** Agent instantiation WITH TOOLS\n\n')
+        f.write('## Results\n\n')
+        f.write('| Framework | Avg Time (μs) | Relative |\n')
+        f.write('|-----------|---------------|----------|\n')
+        f.write('\n'.join(table_rows) + '\n')
+        f.write('\n## Package Versions\n\n')
+        f.write('| Package | Version |\n')
+        f.write('|---------|--------|\n')
+        for pkg, ver in versions.items():
+            f.write(f'| {pkg} | {ver} |\n')
+        f.write('\n## How to Reproduce\n\n')
+        f.write('```bash\n')
+        f.write('cd praisonai-agents\n')
+        f.write('python benchmarks/tools_benchmark.py\n')
+        f.write('```\n')
+    
+    print(f'\nResults saved to: {filepath}')
+    return filepath
+
+
+if __name__ == '__main__':
+    results = run_benchmark()
+    save_results(results)

src/praisonai-agents/callback_test_log.txt

@@ -0,0 +1,9 @@
+Starting test with verbose=False
+CALLBACK EXECUTED!
+Message: 
+You need to do the following task: Say hello world.
+Expected Output: A greeting.
+Please provide only the final result of your work. Do not add any conversation or extra explanation.
+Response: Hello, World!
+Other kwargs: {'markdown': True, 'generation_time': 2.685833215713501, 'agent_name': 'TestAgent', 'agent_role': 'Assistant', 'agent_tools': None, 'task_name': 'simple_task', 'task_description': 'Say hello world', 'task_id': 0}
+--------------------------------------------------

src/praisonai-agents/examples/web_search_example.py

@@ -0,0 +1,90 @@
+"""
+Unified Web Search Example
+
+This example demonstrates the search_web tool that automatically tries
+multiple search providers with fallback support.
+
+Search Provider Priority:
+1. Tavily (requires TAVILY_API_KEY + tavily-python)
+2. Exa (requires EXA_API_KEY + exa_py)
+3. You.com (requires YDC_API_KEY + youdotcom)
+4. DuckDuckGo (requires duckduckgo_search package, no API key)
+5. SearxNG (requires requests + running SearxNG instance)
+
+Installation:
+    pip install praisonaiagents duckduckgo_search
+
+Usage:
+    python web_search_example.py
+"""
+
+from praisonaiagents import Agent
+from praisonaiagents.tools import search_web, get_available_providers
+
+
+def main():
+    print("=" * 60)
+    print("Unified Web Search - Provider Status")
+    print("=" * 60)
+    
+    # Check which providers are available
+    providers = get_available_providers()
+    for p in providers:
+        status = "✓ Available" if p["available"] else f"✗ {p['reason']}"
+        print(f"  {p['name']:12} {status}")
+    
+    print("\n" + "=" * 60)
+    print("Example 1: Basic Search")
+    print("=" * 60)
+    
+    # Simple search - automatically uses best available provider
+    results = search_web("Python programming best practices", max_results=3)
+    
+    if results and "error" not in results[0]:
+        print(f"\nFound {len(results)} results using {results[0].get('provider', 'unknown')}:\n")
+        for r in results:
+            print(f"  Title: {r.get('title', 'N/A')[:60]}...")
+            print(f"  URL: {r.get('url', 'N/A')}")
+            print()
+    else:
+        print(f"\nSearch failed: {results}")
+    
+    print("=" * 60)
+    print("Example 2: Specify Providers")
+    print("=" * 60)
+    
+    # Only try specific providers
+    results = search_web(
+        "machine learning tutorials",
+        max_results=3,
+        providers=["duckduckgo", "tavily"]  # Try these in order
+    )
+    
+    if results and "error" not in results[0]:
+        print(f"\nFound {len(results)} results using {results[0].get('provider', 'unknown')}:\n")
+        for r in results:
+            print(f"  Title: {r.get('title', 'N/A')[:60]}...")
+            print(f"  URL: {r.get('url', 'N/A')}")
+            print()
+    else:
+        print(f"\nSearch failed: {results}")
+    
+    print("=" * 60)
+    print("Example 3: With PraisonAI Agent")
+    print("=" * 60)
+    
+    # Use search_web as an agent tool
+    agent = Agent(
+        name="SearchAgent",
+        role="Web Researcher",
+        goal="Find and summarize information from the web",
+        instructions="Search the web and provide concise summaries of findings.",
+        tools=[search_web]
+    )
+    
+    result = agent.start("What are the top 3 AI trends in 2025?")
+    print(f"\nAgent Response:\n{result}")
+
+
+if __name__ == "__main__":
+    main()

src/praisonai-agents/praisonaiagents/__init__.py

@@ -3,7 +3,7 @@ Praison AI Agents - A package for hierarchical AI agent task execution
 """
 
 # Apply warning patch BEFORE any imports to intercept warnings at the source
-from . import _warning_patch
+from . import _warning_patch  # noqa: F401
 
 # Import centralized logging configuration FIRST
 from . import _logging
@@ -78,6 +78,19 @@ def __getattr__(name):
     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
     raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
 
 # Planning mode support
@@ -289,6 +302,23 @@ __all__ = [
     'ExpandResult'
 ]
 
+# Add workflow exports
+__all__.extend([
+    'Workflow',
+    'WorkflowStep', 
+    'WorkflowContext',
+    'StepResult',
+    'Route',
+    'Parallel',
+    'Loop',
+    'Repeat',
+    'route',
+    'parallel',
+    'loop',
+    'repeat',
+    'Pipeline'
+])
+
 # Add MCP to __all__ if available
 if _mcp_available:
     __all__.append('MCP')
@@ -305,3 +335,29 @@ __all__.extend([
     'LineRange'
 ])
 
+# Add Agent Skills exports (lazy loaded)
+__all__.extend([
+    '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
+

src/praisonai-agents/praisonaiagents/_logging.py

@@ -55,10 +55,10 @@ def _get_all_noisy_loggers() -> List[str]:
 
 def _configure_loggers():
     """Configure all loggers based on LOGLEVEL environment variable."""
-    loglevel = os.environ.get('LOGLEVEL', 'INFO').upper()
+    loglevel = os.environ.get('LOGLEVEL', 'WARNING').upper()
     
-    # When DEBUG is set, allow some HTTP logging for API endpoints
-    if loglevel == 'DEBUG':
+    # When DEBUG or INFO is set, allow some HTTP logging for API endpoints
+    if loglevel in ('DEBUG', 'INFO'):
         allowed_debug_loggers = {"httpx", "httpx._client", "openai._client"}
         
         for logger_name in _get_all_noisy_loggers():
@@ -109,10 +109,10 @@ def configure_root_logger():
     """Configure the root logger with RichHandler."""
     from rich.logging import RichHandler
     
-    loglevel = os.environ.get('LOGLEVEL', 'INFO').upper()
+    loglevel = os.environ.get('LOGLEVEL', 'WARNING').upper()
     
     logging.basicConfig(
-        level=getattr(logging, loglevel, logging.INFO),
+        level=getattr(logging, loglevel, logging.WARNING),
         format="%(asctime)s %(filename)s:%(lineno)d %(levelname)s %(message)s",
         datefmt="[%X]",
         handlers=[RichHandler(rich_tracebacks=True)],

src/praisonai-agents/praisonaiagents/_warning_patch.py

@@ -8,9 +8,10 @@ import functools
 import sys
 
 # Apply aggressive warning filters first
+warnings.filterwarnings("ignore", message=".*There is no current event loop.*", category=DeprecationWarning)
 warnings.filterwarnings("ignore", message=".*Pydantic serializer warnings.*")
 warnings.filterwarnings("ignore", message=".*PydanticSerializationUnexpectedValue.*")
-warnings.filterwarnings("ignore", message=".*Expected 9 fields but got.*")
+warnings.filterwarnings("ignore", message=".*Expected \\d+ fields but got.*")
 warnings.filterwarnings("ignore", message=".*Expected `StreamingChoices` but got.*")
 warnings.filterwarnings("ignore", message=".*serialized value may not be as expected.*")
 warnings.filterwarnings("ignore", message=".*Use 'content=<...>' to upload raw bytes/text content.*")
@@ -27,9 +28,12 @@ SUPPRESSED_PATTERNS = [
     "The `dict` method is deprecated; use `model_dump` instead",
     "Pydantic serializer warnings",
     "PydanticSerializationUnexpectedValue",
-    "Expected 9 fields but got 5 for type `Message`",
+    "Expected",  # Catches all "Expected N fields but got M" patterns
     "Expected `StreamingChoices` but got `Choices`",
-    "serialized value may not be as expected"
+    "serialized value may not be as expected",
+    "Mixing V1 models and V2 models",
+    "Please upgrade `Settings` to V2",
+    "There is no current event loop"
 ]
 
 @functools.wraps(_original_warn)
@@ -41,7 +45,7 @@ def _patched_warn(message, category=None, stacklevel=1, source=None):
         if pattern in msg_str:
             return
     
-    if category == UserWarning and "pydantic" in msg_str.lower():
+    if category is UserWarning and "pydantic" in msg_str.lower():
         return
     
     _original_warn(message, category, stacklevel, source)
@@ -55,7 +59,7 @@ def _patched_warn_explicit(message, category, filename, lineno, module=None, reg
         if pattern in msg_str:
             return
     
-    if category == UserWarning and "pydantic" in msg_str.lower():
+    if category is UserWarning and "pydantic" in msg_str.lower():
         return
     
     if module and "pydantic" in str(module):

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

@@ -250,7 +250,9 @@ class Agent:
         fast_context_parallelism: int = 8,
         fast_context_timeout: float = 30.0,
         history_in_context: Optional[int] = None,
-        auto_save: Optional[str] = None
+        auto_save: Optional[str] = None,
+        skills: Optional[List[str]] = None,
+        skills_dirs: Optional[List[str]] = None
     ):
         """Initialize an Agent instance.
 
@@ -551,8 +553,9 @@ Your Goal: {self.goal}
         self.auto_save = auto_save  # Session name for auto-saving
         
         # Initialize rules manager for persistent context (like Cursor/Windsurf)
+        # NOTE: Lazy initialization - rules are loaded only when accessed (performance optimization)
         self._rules_manager = None
-        self._init_rules_manager()
+        self._rules_manager_initialized = False
         
         # Handle web_search fallback: inject DuckDuckGo tool for unsupported models
         if web_search and not self._model_supports_web_search():
@@ -599,13 +602,19 @@ Your Goal: {self.goal}
 
         # Fast Context configuration (lazy loaded)
         self.fast_context_enabled = fast_context
-        self.fast_context_path = fast_context_path or os.getcwd()
+        self._fast_context_path = fast_context_path  # Store raw, resolve lazily
         self.fast_context_model = fast_context_model
         self.fast_context_max_turns = fast_context_max_turns
         self.fast_context_parallelism = fast_context_parallelism
         self.fast_context_timeout = fast_context_timeout
         self._fast_context_instance = None  # Lazy loaded
 
+        # Agent Skills configuration (lazy loaded for zero performance impact)
+        self._skills = skills
+        self._skills_dirs = skills_dirs
+        self._skill_manager = None  # Lazy loaded
+        self._skills_initialized = False
+
     @property
     def console(self):
         """Lazily initialize Rich Console only when needed."""
@@ -614,6 +623,46 @@ Your Goal: {self.goal}
             self._console = Console()
         return self._console
     
+    @property
+    def fast_context_path(self):
+        """Lazily resolve fast_context_path - avoids os.getcwd() on every init."""
+        if self._fast_context_path is None:
+            self._fast_context_path = os.getcwd()
+        return self._fast_context_path
+    
+    @fast_context_path.setter
+    def fast_context_path(self, value):
+        self._fast_context_path = value
+    
+    @property
+    def skill_manager(self):
+        """Lazily initialize SkillManager only when skills are accessed."""
+        if self._skill_manager is None and (self._skills or self._skills_dirs):
+            from ..skills import SkillManager
+            self._skill_manager = SkillManager()
+            
+            # Add explicit skill paths
+            if self._skills:
+                for skill_path in self._skills:
+                    self._skill_manager.add_skill(skill_path)
+            
+            # Discover skills from directories
+            if self._skills_dirs:
+                self._skill_manager.discover(self._skills_dirs, include_defaults=False)
+            
+            self._skills_initialized = True
+        return self._skill_manager
+    
+    def get_skills_prompt(self) -> str:
+        """Get the XML prompt for available skills.
+        
+        Returns:
+            XML string with <available_skills> block, or empty string if no skills
+        """
+        if self.skill_manager is None:
+            return ""
+        return self.skill_manager.to_prompt()
+    
     @property
     def _openai_client(self):
         """Lazily initialize OpenAI client only when needed."""
@@ -785,6 +834,21 @@ Your Goal: {self.goal}
             logging.warning(f"FastContext search failed: {e}")
             return None
     
+    @property
+    def rules_manager(self):
+        """
+        Lazy-initialized RulesManager for persistent rules/instructions.
+        
+        This property initializes the RulesManager only when first accessed,
+        avoiding expensive filesystem operations during agent instantiation.
+        
+        Returns:
+            RulesManager instance or None if not available
+        """
+        if not self._rules_manager_initialized:
+            self._init_rules_manager()
+        return self._rules_manager
+    
     def _init_rules_manager(self):
         """
         Initialize RulesManager for persistent rules/instructions.
@@ -793,7 +857,10 @@ Your Goal: {self.goal}
         - ~/.praison/rules/ (global)
         - .praison/rules/ (workspace)
         - Subdirectory rules
+        
+        NOTE: This is called lazily via the rules_manager property for performance.
         """
+        self._rules_manager_initialized = True
         try:
             from ..memory.rules_manager import RulesManager
             import os
@@ -828,10 +895,10 @@ Your Goal: {self.goal}
         Returns:
             Formatted rules context string
         """
-        if not self._rules_manager:
+        if not self.rules_manager:
             return ""
         
-        return self._rules_manager.build_rules_context(
+        return self.rules_manager.build_rules_context(
             file_path=file_path,
             include_manual=include_manual
         )
@@ -1239,8 +1306,8 @@ Your Goal: {self.goal}
 Your Role: {self.role}\n
 Your Goal: {self.goal}"""
         
-        # Add rules context if rules manager is enabled
-        if self._rules_manager:
+        # Add rules context if rules manager is enabled (lazy initialization)
+        if self._rules_manager_initialized and self._rules_manager:
             rules_context = self.get_rules_context()
             if rules_context:
                 system_prompt += f"\n\n## Rules (Guidelines you must follow)\n{rules_context}"
@@ -1508,13 +1575,13 @@ Your Goal: {self.goal}"""
         logging.debug(f"{self.name} executing tool {function_name} with arguments: {arguments}")
 
         # Check if approval is required for this tool
-        from ..approval import is_approval_required, console_approval_callback, get_risk_level, mark_approved, ApprovalDecision
+        from ..approval import is_approval_required, console_approval_callback, get_risk_level, mark_approved, ApprovalDecision, get_approval_callback
         if is_approval_required(function_name):
             risk_level = get_risk_level(function_name)
-            logging.info(f"Tool {function_name} requires approval (risk level: {risk_level})")
+            logging.debug(f"Tool {function_name} requires approval (risk level: {risk_level})")
             
             # Use global approval callback or default console callback
-            callback = approval_callback or console_approval_callback
+            callback = get_approval_callback() or console_approval_callback
             
             try:
                 decision = callback(function_name, arguments, risk_level)
@@ -1539,23 +1606,54 @@ Your Goal: {self.goal}"""
         # Special handling for MCP tools
         # Check if tools is an MCP instance with the requested function name
         from ..mcp.mcp import MCP
+        
+        # Helper function to execute MCP tool
+        def _execute_mcp_tool(mcp_instance, func_name, args):
+            """Execute a tool from an MCP instance."""
+            # Handle SSE MCP client
+            if hasattr(mcp_instance, 'is_sse') and mcp_instance.is_sse:
+                if hasattr(mcp_instance, 'sse_client'):
+                    for tool in mcp_instance.sse_client.tools:
+                        if tool.name == func_name:
+                            logging.debug(f"Found matching SSE MCP tool: {func_name}")
+                            return True, tool(**args)
+            # Handle HTTP Stream MCP client
+            if hasattr(mcp_instance, 'is_http_stream') and mcp_instance.is_http_stream:
+                if hasattr(mcp_instance, 'http_stream_client'):
+                    for tool in mcp_instance.http_stream_client.tools:
+                        if tool.name == func_name:
+                            logging.debug(f"Found matching HTTP Stream MCP tool: {func_name}")
+                            return True, tool(**args)
+            # Handle WebSocket MCP client
+            if hasattr(mcp_instance, 'is_websocket') and mcp_instance.is_websocket:
+                if hasattr(mcp_instance, 'websocket_client'):
+                    for tool in mcp_instance.websocket_client.tools:
+                        if tool.name == func_name:
+                            logging.debug(f"Found matching WebSocket MCP tool: {func_name}")
+                            return True, tool(**args)
+            # Handle stdio MCP client
+            if hasattr(mcp_instance, 'runner'):
+                for mcp_tool in mcp_instance.runner.tools:
+                    if hasattr(mcp_tool, 'name') and mcp_tool.name == func_name:
+                        logging.debug(f"Found matching MCP tool: {func_name}")
+                        return True, mcp_instance.runner.call_tool(func_name, args)
+            return False, None
+        
+        # Check if tools is a single MCP instance
         if isinstance(self.tools, MCP):
             logging.debug(f"Looking for MCP tool {function_name}")
-            
-            # Handle SSE MCP client
-            if hasattr(self.tools, 'is_sse') and self.tools.is_sse:
-                if hasattr(self.tools, 'sse_client'):
-                    for tool in self.tools.sse_client.tools:
-                        if tool.name == function_name:
-                            logging.debug(f"Found matching SSE MCP tool: {function_name}")
-                            return tool(**arguments)
-            # Handle stdio MCP client
-            elif hasattr(self.tools, 'runner'):
-                # Check if any of the MCP tools match the function name
-                for mcp_tool in self.tools.runner.tools:
-                    if hasattr(mcp_tool, 'name') and mcp_tool.name == function_name:
-                        logging.debug(f"Found matching MCP tool: {function_name}")
-                        return self.tools.runner.call_tool(function_name, arguments)
+            found, result = _execute_mcp_tool(self.tools, function_name, arguments)
+            if found:
+                return result
+        
+        # Check if tools is a list that may contain MCP instances
+        if isinstance(self.tools, (list, tuple)):
+            for tool in self.tools:
+                if isinstance(tool, MCP):
+                    logging.debug(f"Looking for MCP tool {function_name} in MCP instance")
+                    found, result = _execute_mcp_tool(tool, function_name, arguments)
+                    if found:
+                        return result
 
         # Try to find the function in the agent's tools list first
         func = None
@@ -1870,7 +1968,8 @@ Your Goal: {self.goal}"""
                 if tool_param is not None:
                     from ..mcp.mcp import MCP
                     if isinstance(tool_param, MCP) and hasattr(tool_param, 'to_openai_tool'):
-                        logging.debug("Converting MCP tool to OpenAI format")
+                        # Single MCP instance
+                        logging.debug("Converting single MCP tool to OpenAI format")
                         openai_tool = tool_param.to_openai_tool()
                         if openai_tool:
                             # Handle both single tool and list of tools
@@ -1879,6 +1978,22 @@ Your Goal: {self.goal}"""
                             else:
                                 tool_param = [openai_tool]
                             logging.debug(f"Converted MCP tool: {tool_param}")
+                    elif isinstance(tool_param, (list, tuple)):
+                        # List that may contain MCP instances - convert each MCP to OpenAI format
+                        converted_tools = []
+                        for t in tool_param:
+                            if isinstance(t, MCP) and hasattr(t, 'to_openai_tool'):
+                                logging.debug("Converting MCP instance in list to OpenAI format")
+                                openai_tools = t.to_openai_tool()
+                                if isinstance(openai_tools, list):
+                                    converted_tools.extend(openai_tools)
+                                elif openai_tools:
+                                    converted_tools.append(openai_tools)
+                            else:
+                                # Keep non-MCP tools as-is
+                                converted_tools.append(t)
+                        tool_param = converted_tools
+                        logging.debug(f"Converted {len(converted_tools)} tools from list")
                 
                 # Store chat history length for potential rollback
                 chat_history_length = len(self.chat_history)

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

@@ -1370,6 +1370,61 @@ Context:
             agent_names = ", ".join([agent.name for agent in self.agents])
             print(f"📊 Available agents for this endpoint ({len(self.agents)}): {agent_names}")
             
+            # Create per-agent endpoints for individual agent access
+            # This allows n8n and other tools to call specific agents
+            agents_dict = {agent.name.lower().replace(' ', '_'): agent for agent in self.agents}
+            
+            # Add GET endpoint to list available agents
+            @_agents_shared_apps[port].get(f"{path}/list")
+            async def list_agents():
+                return {
+                    "agents": [
+                        {"name": agent.name, "id": agent.name.lower().replace(' ', '_')}
+                        for agent in self.agents
+                    ]
+                }
+            
+            # Add per-agent POST endpoints
+            for agent_id, agent_instance in agents_dict.items():
+                agent_path = f"{path}/{agent_id}"
+                
+                # Create a closure to capture the agent instance
+                def create_agent_handler(agent):
+                    async def handle_single_agent(request: Request):
+                        try:
+                            request_data = await request.json()
+                            query = request_data.get("query", "")
+                            if not query:
+                                raise HTTPException(status_code=400, detail="Missing 'query' field")
+                        except:
+                            raise HTTPException(status_code=400, detail="Invalid JSON body")
+                        
+                        try:
+                            if asyncio.iscoroutinefunction(agent.chat):
+                                response = await agent.achat(query)
+                            else:
+                                loop = asyncio.get_running_loop()
+                                response = await loop.run_in_executor(None, lambda q=query: agent.chat(q))
+                            
+                            return {
+                                "agent": agent.name,
+                                "query": query,
+                                "response": response
+                            }
+                        except Exception as e:
+                            logging.error(f"Error with agent {agent.name}: {str(e)}", exc_info=True)
+                            return JSONResponse(
+                                status_code=500,
+                                content={"error": f"Agent error: {str(e)}"}
+                            )
+                    return handle_single_agent
+                
+                # Register the endpoint
+                _agents_shared_apps[port].post(agent_path)(create_agent_handler(agent_instance))
+                _agents_registered_endpoints[port][agent_path] = f"{endpoint_id}_{agent_id}"
+            
+            print(f"🔗 Per-agent endpoints: {', '.join([f'{path}/{aid}' for aid in agents_dict.keys()])}")
+            
             # Start the server if it's not already running for this port
             if not _agents_server_started.get(port, False):
                 # Mark the server as started first to prevent duplicate starts

src/praisonai-agents/praisonaiagents/approval.py

@@ -43,6 +43,15 @@ def set_approval_callback(callback_fn: Callable):
     global approval_callback
     approval_callback = callback_fn
 
+def get_approval_callback() -> Optional[Callable]:
+    """Get the current approval callback function.
+    
+    Returns the custom callback if set, otherwise None.
+    This should be used instead of directly accessing approval_callback
+    to ensure the latest callback is always used.
+    """
+    return approval_callback
+
 def mark_approved(tool_name: str):
     """Mark a tool as approved in the current context."""
     approved = _approved_context.get(set())
@@ -90,7 +99,7 @@ def require_approval(risk_level: RiskLevel = "high"):
             except Exception as e:
                 # Fallback to sync approval if async fails
                 logging.warning(f"Async approval failed, using sync fallback: {e}")
-                callback = approval_callback or console_approval_callback
+                callback = get_approval_callback() or console_approval_callback
                 decision = callback(tool_name, kwargs, risk_level)
             
             if not decision.approved:
@@ -198,7 +207,7 @@ async def request_approval(function_name: str, arguments: Dict[str, Any]) -> App
     risk_level = TOOL_RISK_LEVELS.get(function_name, "medium")
     
     # Use custom callback if set, otherwise use console callback
-    callback = approval_callback or console_approval_callback
+    callback = get_approval_callback() or console_approval_callback
     
     try:
         # Handle async callbacks

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

@@ -0,0 +1,68 @@
+"""
+PraisonAI Agents Evaluation Framework.
+
+Provides comprehensive evaluation capabilities for AI agents with zero performance
+impact when not in use through lazy loading.
+
+Evaluator Types:
+    - AccuracyEvaluator: Compare output against expected output using LLM-as-judge
+    - PerformanceEvaluator: Measure runtime and memory usage
+    - ReliabilityEvaluator: Verify expected tool calls are made
+    - CriteriaEvaluator: Evaluate against custom criteria
+
+Example:
+    >>> from praisonaiagents.eval import AccuracyEvaluator
+    >>> evaluator = AccuracyEvaluator(
+    ...     agent=my_agent,
+    ...     input_text="What is 2+2?",
+    ...     expected_output="4"
+    ... )
+    >>> result = evaluator.run(print_summary=True)
+"""
+
+__all__ = [
+    "BaseEvaluator",
+    "AccuracyEvaluator",
+    "PerformanceEvaluator",
+    "ReliabilityEvaluator",
+    "CriteriaEvaluator",
+    "AccuracyResult",
+    "PerformanceResult",
+    "ReliabilityResult",
+    "CriteriaResult",
+    "EvaluationScore",
+    "PerformanceMetrics",
+    "ToolCallResult",
+    "CriteriaScore",
+]
+
+_LAZY_IMPORTS = {
+    "BaseEvaluator": ("base", "BaseEvaluator"),
+    "AccuracyEvaluator": ("accuracy", "AccuracyEvaluator"),
+    "PerformanceEvaluator": ("performance", "PerformanceEvaluator"),
+    "ReliabilityEvaluator": ("reliability", "ReliabilityEvaluator"),
+    "CriteriaEvaluator": ("criteria", "CriteriaEvaluator"),
+    "AccuracyResult": ("results", "AccuracyResult"),
+    "PerformanceResult": ("results", "PerformanceResult"),
+    "ReliabilityResult": ("results", "ReliabilityResult"),
+    "CriteriaResult": ("results", "CriteriaResult"),
+    "EvaluationScore": ("results", "EvaluationScore"),
+    "PerformanceMetrics": ("results", "PerformanceMetrics"),
+    "ToolCallResult": ("results", "ToolCallResult"),
+    "CriteriaScore": ("results", "CriteriaScore"),
+}
+
+
+def __getattr__(name: str):
+    """Lazy import mechanism for zero-cost imports when not used."""
+    if name in _LAZY_IMPORTS:
+        module_name, attr_name = _LAZY_IMPORTS[name]
+        import importlib
+        module = importlib.import_module(f".{module_name}", __name__)
+        return getattr(module, attr_name)
+    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
+
+
+def __dir__():
+    """Return list of available attributes for tab completion."""
+    return __all__

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

@@ -0,0 +1,239 @@
+"""
+Accuracy evaluator for PraisonAI Agents.
+
+Evaluates agent output accuracy by comparing against expected output using LLM-as-judge.
+"""
+
+import os
+import logging
+from typing import Any, Callable, List, Optional, Union, TYPE_CHECKING
+
+from .base import BaseEvaluator
+from .results import AccuracyResult, EvaluationScore
+
+if TYPE_CHECKING:
+    from ..agent.agent import Agent
+    from ..agents.agents import PraisonAIAgents
+
+logger = logging.getLogger(__name__)
+
+
+class AccuracyEvaluator(BaseEvaluator):
+    """
+    Evaluates the accuracy of agent outputs against expected outputs.
+    
+    Uses an LLM as a judge to score outputs on a scale of 1-10.
+    """
+    
+    def __init__(
+        self,
+        agent: Optional[Union["Agent", "PraisonAIAgents"]] = None,
+        func: Optional[Callable[..., str]] = None,
+        input_text: str = "",
+        expected_output: str = "",
+        num_iterations: int = 1,
+        model: Optional[str] = None,
+        name: Optional[str] = None,
+        save_results_path: Optional[str] = None,
+        verbose: bool = False,
+    ):
+        """
+        Initialize the accuracy evaluator.
+        
+        Args:
+            agent: Agent or PraisonAIAgents instance to evaluate
+            func: Alternative callable that returns output string
+            input_text: Input to provide to the agent/function
+            expected_output: Expected output to compare against
+            num_iterations: Number of evaluation iterations
+            model: LLM model to use for judging (defaults to gpt-4o-mini)
+            name: Name for this evaluation
+            save_results_path: Path to save results
+            verbose: Enable verbose output
+        """
+        super().__init__(name=name, save_results_path=save_results_path, verbose=verbose)
+        
+        self.agent = agent
+        self.func = func
+        self.input_text = input_text
+        self.expected_output = expected_output
+        self.num_iterations = num_iterations
+        self.model = model or os.getenv("OPENAI_MODEL_NAME", "gpt-4o-mini")
+        
+        if agent is None and func is None:
+            raise ValueError("Either 'agent' or 'func' must be provided")
+    
+    def _get_output(self) -> str:
+        """Get output from agent or function."""
+        if self.func:
+            return str(self.func(self.input_text))
+        
+        if hasattr(self.agent, 'chat'):
+            return str(self.agent.chat(self.input_text))
+        elif hasattr(self.agent, 'start'):
+            result = self.agent.start(self.input_text)
+            if hasattr(result, 'raw'):
+                return str(result.raw)
+            return str(result)
+        else:
+            raise ValueError("Agent must have 'chat' or 'start' method")
+    
+    def _judge_output(self, output: str) -> EvaluationScore:
+        """
+        Use LLM to judge the output against expected output.
+        
+        Args:
+            output: The actual output to evaluate
+            
+        Returns:
+            EvaluationScore with score and reasoning
+        """
+        try:
+            from openai import OpenAI
+        except ImportError:
+            raise ImportError("OpenAI package required for accuracy evaluation. Install with: pip install openai")
+        
+        client = OpenAI()
+        
+        prompt = f"""You are an expert evaluator. Compare the actual output against the expected output and provide a score from 1-10.
+
+Input: {self.input_text}
+
+Expected Output: {self.expected_output}
+
+Actual Output: {output}
+
+Scoring Guidelines:
+- 10: Perfect match in meaning and completeness
+- 8-9: Very close, minor differences that don't affect correctness
+- 6-7: Mostly correct but missing some details or has minor errors
+- 4-5: Partially correct but significant issues
+- 2-3: Mostly incorrect but shows some understanding
+- 1: Completely wrong or irrelevant
+
+Respond in this exact format:
+SCORE: [number 1-10]
+REASONING: [brief explanation]"""
+
+        response = client.chat.completions.create(
+            model=self.model,
+            messages=[{"role": "user", "content": prompt}],
+            temperature=0.1,
+            max_tokens=200
+        )
+        
+        response_text = response.choices[0].message.content or ""
+        
+        score = 5.0
+        reasoning = "Unable to parse response"
+        
+        lines = response_text.strip().split('\n')
+        for line in lines:
+            if line.startswith('SCORE:'):
+                try:
+                    score = float(line.replace('SCORE:', '').strip())
+                    score = max(1.0, min(10.0, score))
+                except ValueError:
+                    pass
+            elif line.startswith('REASONING:'):
+                reasoning = line.replace('REASONING:', '').strip()
+        
+        return EvaluationScore(
+            score=score,
+            reasoning=reasoning,
+            input_text=self.input_text,
+            output_text=output,
+            expected_output=self.expected_output
+        )
+    
+    def run(self, print_summary: bool = False) -> AccuracyResult:
+        """
+        Execute the accuracy evaluation.
+        
+        Args:
+            print_summary: Whether to print summary after evaluation
+            
+        Returns:
+            AccuracyResult with all evaluation scores
+        """
+        self.before_run()
+        
+        result = AccuracyResult(
+            eval_id=self.eval_id,
+            name=self.name
+        )
+        
+        for i in range(self.num_iterations):
+            if self.verbose:
+                logger.info(f"Running iteration {i + 1}/{self.num_iterations}")
+            
+            try:
+                output = self._get_output()
+                score = self._judge_output(output)
+                result.evaluations.append(score)
+                
+                if self.verbose:
+                    logger.info(f"  Score: {score.score}/10 - {score.reasoning}")
+            except Exception as e:
+                logger.error(f"Error in iteration {i + 1}: {e}")
+                result.evaluations.append(EvaluationScore(
+                    score=0.0,
+                    reasoning=f"Error: {str(e)}",
+                    input_text=self.input_text,
+                    output_text="",
+                    expected_output=self.expected_output
+                ))
+        
+        self.after_run(result)
+        
+        if print_summary:
+            result.print_summary()
+        
+        return result
+    
+    async def run_async(self, print_summary: bool = False) -> AccuracyResult:
+        """
+        Execute the accuracy evaluation asynchronously.
+        
+        Args:
+            print_summary: Whether to print summary after evaluation
+            
+        Returns:
+            AccuracyResult with all evaluation scores
+        """
+        await self.async_before_run()
+        result = self.run(print_summary=False)
+        await self.async_after_run(result)
+        
+        if print_summary:
+            result.print_summary()
+        
+        return result
+    
+    def evaluate_output(self, output: str, print_summary: bool = False) -> AccuracyResult:
+        """
+        Evaluate a pre-generated output without running the agent.
+        
+        Args:
+            output: The output to evaluate
+            print_summary: Whether to print summary after evaluation
+            
+        Returns:
+            AccuracyResult with evaluation score
+        """
+        self.before_run()
+        
+        result = AccuracyResult(
+            eval_id=self.eval_id,
+            name=self.name
+        )
+        
+        score = self._judge_output(output)
+        result.evaluations.append(score)
+        
+        self.after_run(result)
+        
+        if print_summary:
+            result.print_summary()
+        
+        return result

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

@@ -0,0 +1,143 @@
+"""
+Base evaluator class for PraisonAI Agents evaluation framework.
+
+This module provides the abstract base class that all evaluators inherit from.
+"""
+
+from abc import ABC, abstractmethod
+from typing import Any, Dict, Optional, Union, TYPE_CHECKING
+import uuid
+import logging
+
+if TYPE_CHECKING:
+    from .results import AccuracyResult, PerformanceResult, ReliabilityResult, CriteriaResult
+
+logger = logging.getLogger(__name__)
+
+
+class BaseEvaluator(ABC):
+    """
+    Abstract base class for all evaluators in PraisonAI Agents.
+    
+    Provides common functionality for evaluation lifecycle management,
+    result storage, and telemetry integration.
+    """
+    
+    def __init__(
+        self,
+        name: Optional[str] = None,
+        save_results_path: Optional[str] = None,
+        verbose: bool = False,
+    ):
+        """
+        Initialize the base evaluator.
+        
+        Args:
+            name: Optional name for this evaluation run
+            save_results_path: Optional file path to save results (supports {name}, {eval_id} placeholders)
+            verbose: Enable verbose output during evaluation
+        """
+        self.eval_id = str(uuid.uuid4())[:8]
+        self.name = name or f"eval_{self.eval_id}"
+        self.save_results_path = save_results_path
+        self.verbose = verbose
+        self._result = None
+    
+    def before_run(self) -> None:
+        """
+        Hook called before evaluation runs.
+        Override in subclasses for custom pre-evaluation logic.
+        """
+        if self.verbose:
+            logger.info(f"Starting evaluation: {self.name} (ID: {self.eval_id})")
+    
+    def after_run(self, result: Any) -> None:
+        """
+        Hook called after evaluation completes.
+        Override in subclasses for custom post-evaluation logic.
+        
+        Args:
+            result: The evaluation result
+        """
+        if self.verbose:
+            logger.info(f"Completed evaluation: {self.name}")
+        
+        if self.save_results_path:
+            self._save_result(result)
+    
+    async def async_before_run(self) -> None:
+        """
+        Async hook called before evaluation runs.
+        Override in subclasses for custom async pre-evaluation logic.
+        """
+        self.before_run()
+    
+    async def async_after_run(self, result: Any) -> None:
+        """
+        Async hook called after evaluation completes.
+        Override in subclasses for custom async post-evaluation logic.
+        
+        Args:
+            result: The evaluation result
+        """
+        self.after_run(result)
+    
+    @abstractmethod
+    def run(self, **kwargs) -> Any:
+        """
+        Execute the evaluation synchronously.
+        
+        Returns:
+            Evaluation result (type depends on evaluator)
+        """
+        pass
+    
+    async def run_async(self, **kwargs) -> Any:
+        """
+        Execute the evaluation asynchronously.
+        Default implementation calls sync run().
+        Override for true async evaluation.
+        
+        Returns:
+            Evaluation result (type depends on evaluator)
+        """
+        return self.run(**kwargs)
+    
+    def _save_result(self, result: Any) -> None:
+        """
+        Save evaluation result to file.
+        
+        Args:
+            result: The evaluation result to save
+        """
+        if not self.save_results_path:
+            return
+        
+        try:
+            import json
+            from pathlib import Path
+            
+            file_path = self.save_results_path.format(
+                name=self.name,
+                eval_id=self.eval_id
+            )
+            
+            path = Path(file_path)
+            path.parent.mkdir(parents=True, exist_ok=True)
+            
+            if hasattr(result, 'to_dict'):
+                data = result.to_dict()
+            elif hasattr(result, '__dict__'):
+                data = result.__dict__
+            else:
+                data = {"result": str(result)}
+            
+            path.write_text(json.dumps(data, indent=2, default=str))
+            
+            if self.verbose:
+                logger.info(f"Saved results to: {file_path}")
+        except Exception as e:
+            logger.warning(f"Failed to save results: {e}")
+    
+    def __repr__(self) -> str:
+        return f"{self.__class__.__name__}(name={self.name!r}, eval_id={self.eval_id!r})"

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

@@ -0,0 +1,289 @@
+"""
+Criteria evaluator for PraisonAI Agents.
+
+Evaluates agent outputs against custom criteria using LLM-as-judge.
+"""
+
+import os
+import logging
+from typing import Callable, List, Literal, Optional, Union, TYPE_CHECKING
+
+from .base import BaseEvaluator
+from .results import CriteriaResult, CriteriaScore
+
+if TYPE_CHECKING:
+    from ..agent.agent import Agent
+    from ..agents.agents import PraisonAIAgents
+
+logger = logging.getLogger(__name__)
+
+
+class CriteriaEvaluator(BaseEvaluator):
+    """
+    Evaluates agent outputs against custom criteria.
+    
+    Uses an LLM as a judge to score outputs based on user-defined criteria.
+    Supports both numeric (1-10) and binary (pass/fail) scoring.
+    """
+    
+    def __init__(
+        self,
+        criteria: str,
+        agent: Optional[Union["Agent", "PraisonAIAgents"]] = None,
+        func: Optional[Callable[..., str]] = None,
+        input_text: str = "",
+        scoring_type: Literal["numeric", "binary"] = "numeric",
+        threshold: float = 7.0,
+        num_iterations: int = 1,
+        model: Optional[str] = None,
+        on_fail: Optional[Callable[[CriteriaScore], None]] = None,
+        name: Optional[str] = None,
+        save_results_path: Optional[str] = None,
+        verbose: bool = False,
+    ):
+        """
+        Initialize the criteria evaluator.
+        
+        Args:
+            criteria: The criteria to evaluate against (e.g., "Response is helpful and accurate")
+            agent: Agent or PraisonAIAgents instance to evaluate
+            func: Alternative callable that returns output string
+            input_text: Input to provide to the agent/function
+            scoring_type: "numeric" (1-10 scale) or "binary" (pass/fail)
+            threshold: Score threshold for passing (numeric mode only)
+            num_iterations: Number of evaluation iterations
+            model: LLM model to use for judging (defaults to gpt-4o-mini)
+            on_fail: Callback function called when evaluation fails
+            name: Name for this evaluation
+            save_results_path: Path to save results
+            verbose: Enable verbose output
+        """
+        super().__init__(name=name, save_results_path=save_results_path, verbose=verbose)
+        
+        self.criteria = criteria
+        self.agent = agent
+        self.func = func
+        self.input_text = input_text
+        self.scoring_type = scoring_type
+        self.threshold = threshold
+        self.num_iterations = num_iterations
+        self.model = model or os.getenv("OPENAI_MODEL_NAME", "gpt-4o-mini")
+        self.on_fail = on_fail
+    
+    def _get_output(self) -> str:
+        """Get output from agent or function."""
+        if self.func:
+            return str(self.func(self.input_text))
+        
+        if self.agent is None:
+            raise ValueError("Either 'agent' or 'func' must be provided")
+        
+        if hasattr(self.agent, 'chat'):
+            return str(self.agent.chat(self.input_text))
+        elif hasattr(self.agent, 'start'):
+            result = self.agent.start(self.input_text)
+            if hasattr(result, 'raw'):
+                return str(result.raw)
+            return str(result)
+        else:
+            raise ValueError("Agent must have 'chat' or 'start' method")
+    
+    def _judge_output(self, output: str) -> CriteriaScore:
+        """
+        Use LLM to judge the output against criteria.
+        
+        Args:
+            output: The output to evaluate
+            
+        Returns:
+            CriteriaScore with score, passed status, and reasoning
+        """
+        try:
+            from openai import OpenAI
+        except ImportError:
+            raise ImportError("OpenAI package required for criteria evaluation. Install with: pip install openai")
+        
+        client = OpenAI()
+        
+        if self.scoring_type == "binary":
+            prompt = f"""You are an expert evaluator. Evaluate the following output against the given criteria.
+
+Criteria: {self.criteria}
+
+Output to evaluate:
+{output}
+
+Determine if the output PASSES or FAILS the criteria.
+
+Respond in this exact format:
+RESULT: [PASS or FAIL]
+REASONING: [brief explanation]"""
+        else:
+            prompt = f"""You are an expert evaluator. Evaluate the following output against the given criteria.
+
+Criteria: {self.criteria}
+
+Output to evaluate:
+{output}
+
+Score the output from 1-10 based on how well it meets the criteria.
+- 10: Perfectly meets all criteria
+- 8-9: Meets criteria very well with minor issues
+- 6-7: Meets most criteria but has some gaps
+- 4-5: Partially meets criteria
+- 2-3: Barely meets criteria
+- 1: Does not meet criteria at all
+
+Respond in this exact format:
+SCORE: [number 1-10]
+REASONING: [brief explanation]"""
+
+        response = client.chat.completions.create(
+            model=self.model,
+            messages=[{"role": "user", "content": prompt}],
+            temperature=0.1,
+            max_tokens=200
+        )
+        
+        response_text = response.choices[0].message.content or ""
+        
+        score = 5.0
+        passed = False
+        reasoning = "Unable to parse response"
+        
+        lines = response_text.strip().split('\n')
+        for line in lines:
+            if self.scoring_type == "binary":
+                if line.startswith('RESULT:'):
+                    result_str = line.replace('RESULT:', '').strip().upper()
+                    passed = result_str == 'PASS'
+                    score = 10.0 if passed else 0.0
+            else:
+                if line.startswith('SCORE:'):
+                    try:
+                        score = float(line.replace('SCORE:', '').strip())
+                        score = max(1.0, min(10.0, score))
+                        passed = score >= self.threshold
+                    except ValueError:
+                        pass
+            
+            if line.startswith('REASONING:'):
+                reasoning = line.replace('REASONING:', '').strip()
+        
+        return CriteriaScore(
+            score=score,
+            passed=passed,
+            reasoning=reasoning,
+            output_text=output,
+            criteria=self.criteria
+        )
+    
+    def run(self, print_summary: bool = False) -> CriteriaResult:
+        """
+        Execute the criteria evaluation.
+        
+        Args:
+            print_summary: Whether to print summary after evaluation
+            
+        Returns:
+            CriteriaResult with all evaluation scores
+        """
+        self.before_run()
+        
+        result = CriteriaResult(
+            criteria=self.criteria,
+            scoring_type=self.scoring_type,
+            threshold=self.threshold,
+            eval_id=self.eval_id,
+            name=self.name
+        )
+        
+        for i in range(self.num_iterations):
+            if self.verbose:
+                logger.info(f"Running iteration {i + 1}/{self.num_iterations}")
+            
+            try:
+                output = self._get_output()
+                score = self._judge_output(output)
+                result.evaluations.append(score)
+                
+                if self.verbose:
+                    status = "PASS" if score.passed else "FAIL"
+                    logger.info(f"  {status} (Score: {score.score}) - {score.reasoning}")
+                
+                if not score.passed and self.on_fail:
+                    self.on_fail(score)
+                    
+            except Exception as e:
+                logger.error(f"Error in iteration {i + 1}: {e}")
+                error_score = CriteriaScore(
+                    score=0.0,
+                    passed=False,
+                    reasoning=f"Error: {str(e)}",
+                    output_text="",
+                    criteria=self.criteria
+                )
+                result.evaluations.append(error_score)
+                
+                if self.on_fail:
+                    self.on_fail(error_score)
+        
+        self.after_run(result)
+        
+        if print_summary:
+            result.print_summary()
+        
+        return result
+    
+    async def run_async(self, print_summary: bool = False) -> CriteriaResult:
+        """
+        Execute the criteria evaluation asynchronously.
+        
+        Args:
+            print_summary: Whether to print summary after evaluation
+            
+        Returns:
+            CriteriaResult with all evaluation scores
+        """
+        await self.async_before_run()
+        result = self.run(print_summary=False)
+        await self.async_after_run(result)
+        
+        if print_summary:
+            result.print_summary()
+        
+        return result
+    
+    def evaluate_output(self, output: str, print_summary: bool = False) -> CriteriaResult:
+        """
+        Evaluate a pre-generated output without running the agent.
+        
+        Args:
+            output: The output to evaluate
+            print_summary: Whether to print summary after evaluation
+            
+        Returns:
+            CriteriaResult with evaluation score
+        """
+        self.before_run()
+        
+        result = CriteriaResult(
+            criteria=self.criteria,
+            scoring_type=self.scoring_type,
+            threshold=self.threshold,
+            eval_id=self.eval_id,
+            name=self.name
+        )
+        
+        score = self._judge_output(output)
+        result.evaluations.append(score)
+        
+        if not score.passed and self.on_fail:
+            self.on_fail(score)
+        
+        self.after_run(result)
+        
+        if print_summary:
+            result.print_summary()
+        
+        return result

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

@@ -0,0 +1,184 @@
+"""
+Performance evaluator for PraisonAI Agents.
+
+Measures runtime and memory usage of agent executions or functions.
+"""
+
+import time
+import logging
+from typing import Callable, Optional, TYPE_CHECKING
+
+from .base import BaseEvaluator
+from .results import PerformanceResult, PerformanceMetrics
+
+if TYPE_CHECKING:
+    from ..agent.agent import Agent
+    from ..agents.agents import PraisonAIAgents
+
+logger = logging.getLogger(__name__)
+
+
+class PerformanceEvaluator(BaseEvaluator):
+    """
+    Evaluates the performance of agent executions or functions.
+    
+    Measures runtime and memory usage across multiple iterations.
+    """
+    
+    def __init__(
+        self,
+        func: Optional[Callable[[], None]] = None,
+        agent: Optional["Agent"] = None,
+        input_text: str = "",
+        num_iterations: int = 10,
+        warmup_runs: int = 2,
+        track_memory: bool = True,
+        name: Optional[str] = None,
+        save_results_path: Optional[str] = None,
+        verbose: bool = False,
+    ):
+        """
+        Initialize the performance evaluator.
+        
+        Args:
+            func: Callable to benchmark (takes no arguments)
+            agent: Agent to benchmark (will call chat with input_text)
+            input_text: Input text for agent benchmarking
+            num_iterations: Number of benchmark iterations
+            warmup_runs: Number of warmup runs before measurement
+            track_memory: Whether to track memory usage
+            name: Name for this evaluation
+            save_results_path: Path to save results
+            verbose: Enable verbose output
+        """
+        super().__init__(name=name, save_results_path=save_results_path, verbose=verbose)
+        
+        self.func = func
+        self.agent = agent
+        self.input_text = input_text
+        self.num_iterations = num_iterations
+        self.warmup_runs = warmup_runs
+        self.track_memory = track_memory
+        
+        if func is None and agent is None:
+            raise ValueError("Either 'func' or 'agent' must be provided")
+    
+    def _get_memory_usage(self) -> float:
+        """Get current memory usage in MB."""
+        if not self.track_memory:
+            return 0.0
+        
+        try:
+            import tracemalloc
+            current, _ = tracemalloc.get_traced_memory()
+            return current / (1024 * 1024)
+        except Exception:
+            return 0.0
+    
+    def _execute(self) -> None:
+        """Execute the function or agent."""
+        if self.func:
+            self.func()
+        elif self.agent:
+            if hasattr(self.agent, 'chat'):
+                self.agent.chat(self.input_text)
+            elif hasattr(self.agent, 'start'):
+                self.agent.start(self.input_text)
+            else:
+                raise ValueError("Agent must have 'chat' or 'start' method")
+    
+    def run(self, print_summary: bool = False) -> PerformanceResult:
+        """
+        Execute the performance evaluation.
+        
+        Args:
+            print_summary: Whether to print summary after evaluation
+            
+        Returns:
+            PerformanceResult with all performance metrics
+        """
+        self.before_run()
+        
+        result = PerformanceResult(
+            warmup_runs=self.warmup_runs,
+            eval_id=self.eval_id,
+            name=self.name
+        )
+        
+        if self.track_memory:
+            try:
+                import tracemalloc
+                tracemalloc.start()
+            except Exception as e:
+                logger.warning(f"Could not start memory tracking: {e}")
+                self.track_memory = False
+        
+        if self.verbose:
+            logger.info(f"Running {self.warmup_runs} warmup iterations...")
+        
+        for i in range(self.warmup_runs):
+            try:
+                self._execute()
+            except Exception as e:
+                logger.warning(f"Warmup iteration {i + 1} failed: {e}")
+        
+        if self.verbose:
+            logger.info(f"Running {self.num_iterations} benchmark iterations...")
+        
+        for i in range(self.num_iterations):
+            memory_before = self._get_memory_usage()
+            
+            start_time = time.perf_counter()
+            try:
+                self._execute()
+                success = True
+            except Exception as e:
+                logger.error(f"Iteration {i + 1} failed: {e}")
+                success = False
+            end_time = time.perf_counter()
+            
+            memory_after = self._get_memory_usage()
+            
+            if success:
+                metrics = PerformanceMetrics(
+                    run_time=end_time - start_time,
+                    memory_usage=max(0, memory_after - memory_before),
+                    iteration=i + 1
+                )
+                result.metrics.append(metrics)
+                
+                if self.verbose:
+                    logger.info(f"  Iteration {i + 1}: {metrics.run_time:.4f}s, {metrics.memory_usage:.2f}MB")
+        
+        if self.track_memory:
+            try:
+                import tracemalloc
+                tracemalloc.stop()
+            except Exception:
+                pass
+        
+        self.after_run(result)
+        
+        if print_summary:
+            result.print_summary()
+        
+        return result
+    
+    async def run_async(self, print_summary: bool = False) -> PerformanceResult:
+        """
+        Execute the performance evaluation asynchronously.
+        
+        Args:
+            print_summary: Whether to print summary after evaluation
+            
+        Returns:
+            PerformanceResult with all performance metrics
+        """
+        await self.async_before_run()
+        result = self.run(print_summary=False)
+        await self.async_after_run(result)
+        
+        if print_summary:
+            result.print_summary()
+        
+        return result

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

@@ -0,0 +1,225 @@
+"""
+Reliability evaluator for PraisonAI Agents.
+
+Evaluates agent reliability by verifying expected tool calls are made.
+"""
+
+import logging
+from typing import Dict, List, Optional, Set, Any, TYPE_CHECKING
+
+from .base import BaseEvaluator
+from .results import ReliabilityResult, ToolCallResult
+
+if TYPE_CHECKING:
+    from ..agent.agent import Agent
+    from ..agents.agents import PraisonAIAgents
+
+logger = logging.getLogger(__name__)
+
+
+class ReliabilityEvaluator(BaseEvaluator):
+    """
+    Evaluates the reliability of agent tool usage.
+    
+    Verifies that expected tools are called during agent execution.
+    """
+    
+    def __init__(
+        self,
+        agent: Optional["Agent"] = None,
+        input_text: str = "",
+        expected_tools: Optional[List[str]] = None,
+        forbidden_tools: Optional[List[str]] = None,
+        name: Optional[str] = None,
+        save_results_path: Optional[str] = None,
+        verbose: bool = False,
+    ):
+        """
+        Initialize the reliability evaluator.
+        
+        Args:
+            agent: Agent to evaluate
+            input_text: Input to provide to the agent
+            expected_tools: List of tool names that should be called
+            forbidden_tools: List of tool names that should NOT be called
+            name: Name for this evaluation
+            save_results_path: Path to save results
+            verbose: Enable verbose output
+        """
+        super().__init__(name=name, save_results_path=save_results_path, verbose=verbose)
+        
+        self.agent = agent
+        self.input_text = input_text
+        self.expected_tools = set(expected_tools or [])
+        self.forbidden_tools = set(forbidden_tools or [])
+        
+        if agent is None:
+            raise ValueError("'agent' must be provided")
+    
+    def _extract_tool_calls(self, response: Any) -> Set[str]:
+        """
+        Extract tool names from agent response.
+        
+        Args:
+            response: Agent response object
+            
+        Returns:
+            Set of tool names that were called
+        """
+        tool_calls = set()
+        
+        if hasattr(response, 'tool_calls') and response.tool_calls:
+            for tc in response.tool_calls:
+                if hasattr(tc, 'name'):
+                    tool_calls.add(tc.name)
+                elif hasattr(tc, 'function') and hasattr(tc.function, 'name'):
+                    tool_calls.add(tc.function.name)
+        
+        if hasattr(self.agent, 'chat_history'):
+            for msg in self.agent.chat_history:
+                if isinstance(msg, dict):
+                    if msg.get('role') == 'assistant' and 'tool_calls' in msg:
+                        for tc in msg.get('tool_calls', []):
+                            if isinstance(tc, dict) and 'function' in tc:
+                                tool_calls.add(tc['function'].get('name', ''))
+                    elif msg.get('role') == 'tool':
+                        tool_name = msg.get('name', '')
+                        if tool_name:
+                            tool_calls.add(tool_name)
+        
+        return tool_calls
+    
+    def run(self, print_summary: bool = False) -> ReliabilityResult:
+        """
+        Execute the reliability evaluation.
+        
+        Args:
+            print_summary: Whether to print summary after evaluation
+            
+        Returns:
+            ReliabilityResult with tool call verification results
+        """
+        self.before_run()
+        
+        result = ReliabilityResult(
+            eval_id=self.eval_id,
+            name=self.name
+        )
+        
+        try:
+            if hasattr(self.agent, 'chat'):
+                response = self.agent.chat(self.input_text)
+            elif hasattr(self.agent, 'start'):
+                response = self.agent.start(self.input_text)
+            else:
+                raise ValueError("Agent must have 'chat' or 'start' method")
+            
+            actual_tools = self._extract_tool_calls(response)
+            
+            if self.verbose:
+                logger.info(f"Expected tools: {self.expected_tools}")
+                logger.info(f"Actual tools called: {actual_tools}")
+            
+            for tool in self.expected_tools:
+                was_called = tool in actual_tools
+                result.tool_results.append(ToolCallResult(
+                    tool_name=tool,
+                    expected=True,
+                    actual=was_called,
+                    passed=was_called
+                ))
+            
+            for tool in self.forbidden_tools:
+                was_called = tool in actual_tools
+                result.tool_results.append(ToolCallResult(
+                    tool_name=tool,
+                    expected=False,
+                    actual=was_called,
+                    passed=not was_called
+                ))
+            
+        except Exception as e:
+            logger.error(f"Error during reliability evaluation: {e}")
+            for tool in self.expected_tools:
+                result.tool_results.append(ToolCallResult(
+                    tool_name=tool,
+                    expected=True,
+                    actual=False,
+                    passed=False,
+                    metadata={"error": str(e)}
+                ))
+        
+        self.after_run(result)
+        
+        if print_summary:
+            result.print_summary()
+        
+        return result
+    
+    async def run_async(self, print_summary: bool = False) -> ReliabilityResult:
+        """
+        Execute the reliability evaluation asynchronously.
+        
+        Args:
+            print_summary: Whether to print summary after evaluation
+            
+        Returns:
+            ReliabilityResult with tool call verification results
+        """
+        await self.async_before_run()
+        result = self.run(print_summary=False)
+        await self.async_after_run(result)
+        
+        if print_summary:
+            result.print_summary()
+        
+        return result
+    
+    def evaluate_tool_calls(
+        self,
+        actual_tools: List[str],
+        print_summary: bool = False
+    ) -> ReliabilityResult:
+        """
+        Evaluate a pre-recorded list of tool calls without running the agent.
+        
+        Args:
+            actual_tools: List of tool names that were called
+            print_summary: Whether to print summary after evaluation
+            
+        Returns:
+            ReliabilityResult with tool call verification results
+        """
+        self.before_run()
+        
+        result = ReliabilityResult(
+            eval_id=self.eval_id,
+            name=self.name
+        )
+        
+        actual_set = set(actual_tools)
+        
+        for tool in self.expected_tools:
+            was_called = tool in actual_set
+            result.tool_results.append(ToolCallResult(
+                tool_name=tool,
+                expected=True,
+                actual=was_called,
+                passed=was_called
+            ))
+        
+        for tool in self.forbidden_tools:
+            was_called = tool in actual_set
+            result.tool_results.append(ToolCallResult(
+                tool_name=tool,
+                expected=False,
+                actual=was_called,
+                passed=not was_called
+            ))
+        
+        self.after_run(result)
+        
+        if print_summary:
+            result.print_summary()
+        
+        return result

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

@@ -0,0 +1,463 @@
+"""
+Result dataclasses for PraisonAI Agents evaluation framework.
+
+This module provides structured result types for different evaluation methods.
+"""
+
+from dataclasses import dataclass, field, asdict
+from typing import List, Optional, Dict, Any
+from datetime import datetime
+import statistics
+import json
+
+
+@dataclass
+class EvaluationScore:
+    """Individual evaluation score with reasoning."""
+    score: float
+    reasoning: str
+    input_text: str
+    output_text: str
+    expected_output: Optional[str] = None
+    timestamp: str = field(default_factory=lambda: datetime.now().isoformat())
+    metadata: Dict[str, Any] = field(default_factory=dict)
+
+
+@dataclass
+class AccuracyResult:
+    """Result from accuracy evaluation comparing output to expected output."""
+    evaluations: List[EvaluationScore] = field(default_factory=list)
+    eval_id: str = ""
+    name: str = ""
+    
+    @property
+    def scores(self) -> List[float]:
+        """Get list of all scores."""
+        return [e.score for e in self.evaluations]
+    
+    @property
+    def avg_score(self) -> float:
+        """Calculate average score."""
+        if not self.scores:
+            return 0.0
+        return statistics.mean(self.scores)
+    
+    @property
+    def min_score(self) -> float:
+        """Get minimum score."""
+        if not self.scores:
+            return 0.0
+        return min(self.scores)
+    
+    @property
+    def max_score(self) -> float:
+        """Get maximum score."""
+        if not self.scores:
+            return 0.0
+        return max(self.scores)
+    
+    @property
+    def std_dev(self) -> float:
+        """Calculate standard deviation of scores."""
+        if len(self.scores) < 2:
+            return 0.0
+        return statistics.stdev(self.scores)
+    
+    @property
+    def passed(self) -> bool:
+        """Check if evaluation passed (avg score >= 7)."""
+        return self.avg_score >= 7.0
+    
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert result to dictionary."""
+        return {
+            "eval_id": self.eval_id,
+            "name": self.name,
+            "evaluations": [asdict(e) for e in self.evaluations],
+            "avg_score": self.avg_score,
+            "min_score": self.min_score,
+            "max_score": self.max_score,
+            "std_dev": self.std_dev,
+            "passed": self.passed,
+            "num_evaluations": len(self.evaluations)
+        }
+    
+    def to_json(self) -> str:
+        """Convert result to JSON string."""
+        return json.dumps(self.to_dict(), indent=2)
+    
+    def print_summary(self) -> None:
+        """Print a summary of the evaluation results."""
+        try:
+            from rich.console import Console
+            from rich.table import Table
+            from rich.panel import Panel  # noqa: F401
+            
+            console = Console()
+            
+            table = Table(title="Accuracy Evaluation Summary")
+            table.add_column("Metric", style="cyan")
+            table.add_column("Value", style="green")
+            
+            table.add_row("Evaluations", str(len(self.evaluations)))
+            table.add_row("Average Score", f"{self.avg_score:.2f}")
+            table.add_row("Min Score", f"{self.min_score:.2f}")
+            table.add_row("Max Score", f"{self.max_score:.2f}")
+            table.add_row("Std Dev", f"{self.std_dev:.2f}")
+            table.add_row("Status", "✅ PASSED" if self.passed else "❌ FAILED")
+            
+            console.print(table)
+        except ImportError:
+            print("Accuracy Evaluation Summary")
+            print(f"  Evaluations: {len(self.evaluations)}")
+            print(f"  Average Score: {self.avg_score:.2f}")
+            print(f"  Min Score: {self.min_score:.2f}")
+            print(f"  Max Score: {self.max_score:.2f}")
+            print(f"  Std Dev: {self.std_dev:.2f}")
+            print(f"  Status: {'PASSED' if self.passed else 'FAILED'}")
+
+
+@dataclass
+class PerformanceMetrics:
+    """Individual performance measurement."""
+    run_time: float
+    memory_usage: float
+    iteration: int
+    timestamp: str = field(default_factory=lambda: datetime.now().isoformat())
+    metadata: Dict[str, Any] = field(default_factory=dict)
+
+
+@dataclass
+class PerformanceResult:
+    """Result from performance evaluation measuring runtime and memory."""
+    metrics: List[PerformanceMetrics] = field(default_factory=list)
+    warmup_runs: int = 0
+    eval_id: str = ""
+    name: str = ""
+    
+    @property
+    def run_times(self) -> List[float]:
+        """Get list of all run times."""
+        return [m.run_time for m in self.metrics]
+    
+    @property
+    def memory_usages(self) -> List[float]:
+        """Get list of all memory usages."""
+        return [m.memory_usage for m in self.metrics]
+    
+    @property
+    def avg_run_time(self) -> float:
+        """Calculate average run time."""
+        if not self.run_times:
+            return 0.0
+        return statistics.mean(self.run_times)
+    
+    @property
+    def min_run_time(self) -> float:
+        """Get minimum run time."""
+        if not self.run_times:
+            return 0.0
+        return min(self.run_times)
+    
+    @property
+    def max_run_time(self) -> float:
+        """Get maximum run time."""
+        if not self.run_times:
+            return 0.0
+        return max(self.run_times)
+    
+    @property
+    def std_dev_run_time(self) -> float:
+        """Calculate standard deviation of run times."""
+        if len(self.run_times) < 2:
+            return 0.0
+        return statistics.stdev(self.run_times)
+    
+    @property
+    def median_run_time(self) -> float:
+        """Calculate median run time."""
+        if not self.run_times:
+            return 0.0
+        return statistics.median(self.run_times)
+    
+    @property
+    def p95_run_time(self) -> float:
+        """Calculate 95th percentile run time."""
+        if not self.run_times:
+            return 0.0
+        sorted_times = sorted(self.run_times)
+        idx = int(len(sorted_times) * 0.95)
+        return sorted_times[min(idx, len(sorted_times) - 1)]
+    
+    @property
+    def avg_memory(self) -> float:
+        """Calculate average memory usage."""
+        if not self.memory_usages:
+            return 0.0
+        return statistics.mean(self.memory_usages)
+    
+    @property
+    def max_memory(self) -> float:
+        """Get maximum memory usage."""
+        if not self.memory_usages:
+            return 0.0
+        return max(self.memory_usages)
+    
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert result to dictionary."""
+        return {
+            "eval_id": self.eval_id,
+            "name": self.name,
+            "metrics": [asdict(m) for m in self.metrics],
+            "warmup_runs": self.warmup_runs,
+            "avg_run_time": self.avg_run_time,
+            "min_run_time": self.min_run_time,
+            "max_run_time": self.max_run_time,
+            "std_dev_run_time": self.std_dev_run_time,
+            "median_run_time": self.median_run_time,
+            "p95_run_time": self.p95_run_time,
+            "avg_memory": self.avg_memory,
+            "max_memory": self.max_memory,
+            "num_iterations": len(self.metrics)
+        }
+    
+    def to_json(self) -> str:
+        """Convert result to JSON string."""
+        return json.dumps(self.to_dict(), indent=2)
+    
+    def print_summary(self) -> None:
+        """Print a summary of the performance results."""
+        try:
+            from rich.console import Console
+            from rich.table import Table
+            
+            console = Console()
+            
+            table = Table(title="Performance Evaluation Summary")
+            table.add_column("Metric", style="cyan")
+            table.add_column("Value", style="green")
+            
+            table.add_row("Iterations", str(len(self.metrics)))
+            table.add_row("Warmup Runs", str(self.warmup_runs))
+            table.add_row("Avg Run Time", f"{self.avg_run_time:.4f}s")
+            table.add_row("Min Run Time", f"{self.min_run_time:.4f}s")
+            table.add_row("Max Run Time", f"{self.max_run_time:.4f}s")
+            table.add_row("Std Dev", f"{self.std_dev_run_time:.4f}s")
+            table.add_row("Median", f"{self.median_run_time:.4f}s")
+            table.add_row("P95", f"{self.p95_run_time:.4f}s")
+            table.add_row("Avg Memory", f"{self.avg_memory:.2f} MB")
+            table.add_row("Max Memory", f"{self.max_memory:.2f} MB")
+            
+            console.print(table)
+        except ImportError:
+            print("Performance Evaluation Summary")
+            print(f"  Iterations: {len(self.metrics)}")
+            print(f"  Warmup Runs: {self.warmup_runs}")
+            print(f"  Avg Run Time: {self.avg_run_time:.4f}s")
+            print(f"  Min Run Time: {self.min_run_time:.4f}s")
+            print(f"  Max Run Time: {self.max_run_time:.4f}s")
+            print(f"  Std Dev: {self.std_dev_run_time:.4f}s")
+            print(f"  Median: {self.median_run_time:.4f}s")
+            print(f"  P95: {self.p95_run_time:.4f}s")
+            print(f"  Avg Memory: {self.avg_memory:.2f} MB")
+            print(f"  Max Memory: {self.max_memory:.2f} MB")
+
+
+@dataclass
+class ToolCallResult:
+    """Result of a single tool call verification."""
+    tool_name: str
+    expected: bool
+    actual: bool
+    passed: bool
+    arguments: Optional[Dict[str, Any]] = None
+    metadata: Dict[str, Any] = field(default_factory=dict)
+
+
+@dataclass
+class ReliabilityResult:
+    """Result from reliability evaluation checking tool calls."""
+    tool_results: List[ToolCallResult] = field(default_factory=list)
+    eval_id: str = ""
+    name: str = ""
+    
+    @property
+    def passed_calls(self) -> List[ToolCallResult]:
+        """Get list of passed tool calls."""
+        return [t for t in self.tool_results if t.passed]
+    
+    @property
+    def failed_calls(self) -> List[ToolCallResult]:
+        """Get list of failed tool calls."""
+        return [t for t in self.tool_results if not t.passed]
+    
+    @property
+    def pass_rate(self) -> float:
+        """Calculate pass rate."""
+        if not self.tool_results:
+            return 0.0
+        return len(self.passed_calls) / len(self.tool_results)
+    
+    @property
+    def status(self) -> str:
+        """Get overall status."""
+        return "PASSED" if len(self.failed_calls) == 0 else "FAILED"
+    
+    @property
+    def passed(self) -> bool:
+        """Check if all tool calls passed."""
+        return len(self.failed_calls) == 0
+    
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert result to dictionary."""
+        return {
+            "eval_id": self.eval_id,
+            "name": self.name,
+            "tool_results": [asdict(t) for t in self.tool_results],
+            "passed_count": len(self.passed_calls),
+            "failed_count": len(self.failed_calls),
+            "pass_rate": self.pass_rate,
+            "status": self.status
+        }
+    
+    def to_json(self) -> str:
+        """Convert result to JSON string."""
+        return json.dumps(self.to_dict(), indent=2)
+    
+    def print_summary(self) -> None:
+        """Print a summary of the reliability results."""
+        try:
+            from rich.console import Console
+            from rich.table import Table
+            
+            console = Console()
+            
+            table = Table(title="Reliability Evaluation Summary")
+            table.add_column("Metric", style="cyan")
+            table.add_column("Value", style="green")
+            
+            table.add_row("Total Tool Calls", str(len(self.tool_results)))
+            table.add_row("Passed", str(len(self.passed_calls)))
+            table.add_row("Failed", str(len(self.failed_calls)))
+            table.add_row("Pass Rate", f"{self.pass_rate:.1%}")
+            table.add_row("Status", "✅ PASSED" if self.passed else "❌ FAILED")
+            
+            console.print(table)
+            
+            if self.failed_calls:
+                fail_table = Table(title="Failed Tool Calls")
+                fail_table.add_column("Tool", style="red")
+                fail_table.add_column("Expected", style="yellow")
+                fail_table.add_column("Actual", style="yellow")
+                
+                for tc in self.failed_calls:
+                    fail_table.add_row(tc.tool_name, str(tc.expected), str(tc.actual))
+                
+                console.print(fail_table)
+        except ImportError:
+            print("Reliability Evaluation Summary")
+            print(f"  Total Tool Calls: {len(self.tool_results)}")
+            print(f"  Passed: {len(self.passed_calls)}")
+            print(f"  Failed: {len(self.failed_calls)}")
+            print(f"  Pass Rate: {self.pass_rate:.1%}")
+            print(f"  Status: {self.status}")
+
+
+@dataclass
+class CriteriaScore:
+    """Individual criteria evaluation score."""
+    score: float
+    passed: bool
+    reasoning: str
+    output_text: str
+    criteria: str
+    timestamp: str = field(default_factory=lambda: datetime.now().isoformat())
+    metadata: Dict[str, Any] = field(default_factory=dict)
+
+
+@dataclass
+class CriteriaResult:
+    """Result from criteria-based evaluation (agent-as-judge)."""
+    evaluations: List[CriteriaScore] = field(default_factory=list)
+    criteria: str = ""
+    scoring_type: str = "numeric"  # "numeric" or "binary"
+    threshold: float = 7.0
+    eval_id: str = ""
+    name: str = ""
+    
+    @property
+    def scores(self) -> List[float]:
+        """Get list of all scores."""
+        return [e.score for e in self.evaluations]
+    
+    @property
+    def avg_score(self) -> float:
+        """Calculate average score."""
+        if not self.scores:
+            return 0.0
+        return statistics.mean(self.scores)
+    
+    @property
+    def pass_rate(self) -> float:
+        """Calculate pass rate."""
+        if not self.evaluations:
+            return 0.0
+        passed = sum(1 for e in self.evaluations if e.passed)
+        return passed / len(self.evaluations)
+    
+    @property
+    def passed(self) -> bool:
+        """Check if evaluation passed overall."""
+        if self.scoring_type == "binary":
+            return self.pass_rate >= 0.5
+        return self.avg_score >= self.threshold
+    
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert result to dictionary."""
+        return {
+            "eval_id": self.eval_id,
+            "name": self.name,
+            "criteria": self.criteria,
+            "scoring_type": self.scoring_type,
+            "threshold": self.threshold,
+            "evaluations": [asdict(e) for e in self.evaluations],
+            "avg_score": self.avg_score,
+            "pass_rate": self.pass_rate,
+            "passed": self.passed,
+            "num_evaluations": len(self.evaluations)
+        }
+    
+    def to_json(self) -> str:
+        """Convert result to JSON string."""
+        return json.dumps(self.to_dict(), indent=2)
+    
+    def print_summary(self) -> None:
+        """Print a summary of the criteria evaluation results."""
+        try:
+            from rich.console import Console
+            from rich.table import Table
+            
+            console = Console()
+            
+            table = Table(title="Criteria Evaluation Summary")
+            table.add_column("Metric", style="cyan")
+            table.add_column("Value", style="green")
+            
+            table.add_row("Criteria", self.criteria[:50] + "..." if len(self.criteria) > 50 else self.criteria)
+            table.add_row("Scoring Type", self.scoring_type)
+            table.add_row("Threshold", str(self.threshold))
+            table.add_row("Evaluations", str(len(self.evaluations)))
+            table.add_row("Average Score", f"{self.avg_score:.2f}")
+            table.add_row("Pass Rate", f"{self.pass_rate:.1%}")
+            table.add_row("Status", "✅ PASSED" if self.passed else "❌ FAILED")
+            
+            console.print(table)
+        except ImportError:
+            print("Criteria Evaluation Summary")
+            print(f"  Criteria: {self.criteria[:50]}...")
+            print(f"  Scoring Type: {self.scoring_type}")
+            print(f"  Threshold: {self.threshold}")
+            print(f"  Evaluations: {len(self.evaluations)}")
+            print(f"  Average Score: {self.avg_score:.2f}")
+            print(f"  Pass Rate: {self.pass_rate:.1%}")
+            print(f"  Status: {'PASSED' if self.passed else 'FAILED'}")

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

@@ -0,0 +1,154 @@
+"""
+Utility functions for PraisonAI Agents evaluation framework.
+"""
+
+import json
+import logging
+from pathlib import Path
+from typing import Any, Dict, Optional, Union
+
+logger = logging.getLogger(__name__)
+
+
+def save_result_to_file(
+    result: Any,
+    file_path: str,
+    eval_id: Optional[str] = None,
+    name: Optional[str] = None
+) -> bool:
+    """
+    Save evaluation result to a file.
+    
+    Args:
+        result: The evaluation result to save
+        file_path: Path to save the result (supports {name}, {eval_id} placeholders)
+        eval_id: Optional eval ID for placeholder substitution
+        name: Optional name for placeholder substitution
+        
+    Returns:
+        True if save was successful, False otherwise
+    """
+    try:
+        formatted_path = file_path.format(name=name or "", eval_id=eval_id or "")
+        path = Path(formatted_path)
+        path.parent.mkdir(parents=True, exist_ok=True)
+        
+        if hasattr(result, 'to_dict'):
+            data = result.to_dict()
+        elif hasattr(result, '__dict__'):
+            data = result.__dict__
+        else:
+            data = {"result": str(result)}
+        
+        path.write_text(json.dumps(data, indent=2, default=str))
+        logger.debug(f"Saved result to: {formatted_path}")
+        return True
+    except Exception as e:
+        logger.warning(f"Failed to save result: {e}")
+        return False
+
+
+def load_result_from_file(file_path: str) -> Optional[Dict[str, Any]]:
+    """
+    Load evaluation result from a file.
+    
+    Args:
+        file_path: Path to the result file
+        
+    Returns:
+        Dictionary containing the result data, or None if load failed
+    """
+    try:
+        path = Path(file_path)
+        if not path.exists():
+            logger.warning(f"Result file not found: {file_path}")
+            return None
+        
+        data = json.loads(path.read_text())
+        return data
+    except Exception as e:
+        logger.warning(f"Failed to load result: {e}")
+        return None
+
+
+def format_score(score: float, max_score: float = 10.0) -> str:
+    """
+    Format a score for display.
+    
+    Args:
+        score: The score value
+        max_score: Maximum possible score
+        
+    Returns:
+        Formatted score string
+    """
+    return f"{score:.1f}/{max_score:.0f}"
+
+
+def format_percentage(value: float) -> str:
+    """
+    Format a value as a percentage.
+    
+    Args:
+        value: Value between 0 and 1
+        
+    Returns:
+        Formatted percentage string
+    """
+    return f"{value * 100:.1f}%"
+
+
+def format_duration(seconds: float) -> str:
+    """
+    Format a duration in seconds for display.
+    
+    Args:
+        seconds: Duration in seconds
+        
+    Returns:
+        Formatted duration string
+    """
+    if seconds < 0.001:
+        return f"{seconds * 1000000:.0f}μs"
+    elif seconds < 1:
+        return f"{seconds * 1000:.2f}ms"
+    elif seconds < 60:
+        return f"{seconds:.2f}s"
+    else:
+        minutes = int(seconds // 60)
+        secs = seconds % 60
+        return f"{minutes}m {secs:.1f}s"
+
+
+def format_memory(mb: float) -> str:
+    """
+    Format memory usage for display.
+    
+    Args:
+        mb: Memory in megabytes
+        
+    Returns:
+        Formatted memory string
+    """
+    if mb < 1:
+        return f"{mb * 1024:.1f} KB"
+    elif mb < 1024:
+        return f"{mb:.2f} MB"
+    else:
+        return f"{mb / 1024:.2f} GB"
+
+
+def calculate_pass_rate(passed: int, total: int) -> float:
+    """
+    Calculate pass rate.
+    
+    Args:
+        passed: Number of passed evaluations
+        total: Total number of evaluations
+        
+    Returns:
+        Pass rate as a float between 0 and 1
+    """
+    if total == 0:
+        return 0.0
+    return passed / total

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

@@ -1054,7 +1054,7 @@ class LLM:
         **kwargs
     ) -> str:
         """Enhanced get_response with all OpenAI-like features"""
-        logging.info(f"Getting response from {self.model}")
+        logging.debug(f"Getting response from {self.model}")
         # Log all self values when in debug mode
         self._log_llm_config(
             'LLM instance',
@@ -2472,7 +2472,7 @@ Output MUST be JSON with 'reflection' and 'satisfactory'.
         """Async version of get_response with identical functionality."""
         try:
             import litellm
-            logging.info(f"Getting async response from {self.model}")
+            logging.debug(f"Getting async response from {self.model}")
             # Log all self values when in debug mode
             self._log_llm_config(
                 'get_response_async',

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

@@ -19,6 +19,15 @@ from rich.console import Console
 from rich.live import Live
 import inspect
 
+# Import display_tool_call for callback support (lazy import to avoid circular imports)
+_display_tool_call = None
+def _get_display_tool_call():
+    global _display_tool_call
+    if _display_tool_call is None:
+        from ..main import display_tool_call
+        _display_tool_call = display_tool_call
+    return _display_tool_call
+
 # Constants
 LOCAL_SERVER_API_KEY_PLACEHOLDER = "not-needed"
 
@@ -978,16 +987,19 @@ class OpenAIClient:
                         function_name = tool_call.function.name
                         arguments = json.loads(tool_call.function.arguments)
                     
+                    # Always trigger callback for tool call tracking (even when verbose=False)
+                    display_tool_call_fn = _get_display_tool_call()
+                    display_tool_call_fn(f"Calling function: {function_name}", console=console if verbose else None)
+                    
                     if verbose and console:
-                        console.print(f"[bold]Calling function:[/bold] {function_name}")
                         console.print(f"[dim]Arguments:[/dim] {arguments}")
                     
                     # Execute the tool
                     tool_result = execute_tool_fn(function_name, arguments)
                     results_str = json.dumps(tool_result) if tool_result else "Function returned an empty output"
                     
-                    if verbose and console:
-                        console.print(f"[dim]Result:[/dim] {results_str}")
+                    # Trigger callback with result
+                    display_tool_call_fn(f"Function {function_name} returned: {results_str[:200]}{'...' if len(results_str) > 200 else ''}", console=console if verbose else None)
                     
                     messages.append({
                         "role": "tool",
@@ -1150,8 +1162,11 @@ class OpenAIClient:
                         function_name = tool_call.function.name
                         arguments = json.loads(tool_call.function.arguments)
                     
+                    # Always trigger callback for tool call tracking (even when verbose=False)
+                    display_tool_call_fn = _get_display_tool_call()
+                    display_tool_call_fn(f"Calling function: {function_name}", console=console if verbose else None)
+                    
                     if verbose and console:
-                        console.print(f"[bold]Calling function:[/bold] {function_name}")
                         console.print(f"[dim]Arguments:[/dim] {arguments}")
                     
                     # Execute the tool (async)
@@ -1167,8 +1182,8 @@ class OpenAIClient:
                     
                     results_str = json.dumps(tool_result) if tool_result else "Function returned an empty output"
                     
-                    if verbose and console:
-                        console.print(f"[dim]Result:[/dim] {results_str}")
+                    # Trigger callback with result
+                    display_tool_call_fn(f"Function {function_name} returned: {results_str[:200]}{'...' if len(results_str) > 200 else ''}", console=console if verbose else None)
                     
                     messages.append({
                         "role": "tool",
@@ -1299,6 +1314,10 @@ class OpenAIClient:
                                 yield f"\n[Error parsing arguments for {function_name if 'function_name' in locals() else 'unknown function'}: {str(e)}]"
                             continue
                         
+                        # Always trigger callback for tool call tracking (even when verbose=False)
+                        display_tool_call_fn = _get_display_tool_call()
+                        display_tool_call_fn(f"Calling function: {function_name}", console=None)
+                        
                         if verbose:
                             yield f"\n[Calling function: {function_name}]"
                         
@@ -1311,6 +1330,9 @@ class OpenAIClient:
                             if verbose:
                                 yield f"\n[Function error: {str(e)}]"
                         
+                        # Trigger callback with result
+                        display_tool_call_fn(f"Function {function_name} returned: {results_str[:200]}{'...' if len(results_str) > 200 else ''}", console=None)
+                        
                         if verbose:
                             yield f"\n[Function result: {results_str}]"
                         

src/praisonai-agents/praisonaiagents/main.py

@@ -231,15 +231,15 @@ def display_tool_call(message: str, console=None):
     if not message or not message.strip():
         logging.debug("Empty message in display_tool_call, returning early")
         return
-    if console is None:
-        console = Console()
     message = _clean_display_content(str(message))
     logging.debug(f"Cleaned message in display_tool_call: {repr(message)}")
     
-    # Execute synchronous callbacks
+    # Execute synchronous callbacks (always, even when console is None)
     execute_sync_callback('tool_call', message=message)
     
-    console.print(Panel.fit(Text(message, style="bold cyan"), title="Tool Call", border_style="green"))
+    # Only print panel if console is provided (verbose mode)
+    if console is not None:
+        console.print(Panel.fit(Text(message, style="bold cyan"), title="Tool Call", border_style="green"))
 
 def display_error(message: str, console=None):
     if not message or not message.strip():

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

@@ -440,7 +440,7 @@ class Workflow:
             previous_output = output
             
             if verbose:
-                print(f"✅ {step.name}: {str(output)[:100]}...")
+                print(f"✅ {step.name}: {str(output)}")
             
             # Handle early stop
             if stop:
@@ -571,7 +571,7 @@ class Workflow:
                 output = f"Error: {e}"
         
         if verbose:
-            print(f"✅ {normalized.name}: {str(output)[:100]}...")
+            print(f"✅ {normalized.name}: {str(output)}")
         
         return {
             "step": normalized.name,

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

@@ -0,0 +1,76 @@
+"""Agent Skills module for PraisonAI Agents.
+
+This module provides support for the open Agent Skills standard (agentskills.io),
+enabling agents to load and use modular capabilities through SKILL.md files.
+
+Zero Performance Impact:
+- All imports are lazy loaded via __getattr__
+- Skills only loaded when explicitly enabled
+- No auto-discovery at import time
+
+Usage:
+    from praisonaiagents.skills import SkillManager, SkillProperties
+    
+    manager = SkillManager()
+    manager.discover(["./skills"])
+    prompt_xml = manager.to_prompt()
+"""
+
+__all__ = [
+    # Models
+    "SkillProperties",
+    "SkillMetadata",
+    "ParseError",
+    "ValidationError",
+    # Parser
+    "parse_frontmatter",
+    "find_skill_md",
+    "read_properties",
+    # Validator
+    "validate",
+    "validate_metadata",
+    # Prompt
+    "to_prompt",
+    "generate_skills_xml",
+    "format_skill_for_prompt",
+    # Discovery
+    "discover_skills",
+    "get_default_skill_dirs",
+    # Loader
+    "SkillLoader",
+    # Manager
+    "SkillManager",
+]
+
+
+def __getattr__(name: str):
+    """Lazy load module components to avoid import overhead."""
+    if name in ("SkillProperties", "SkillMetadata", "ParseError", "ValidationError"):
+        from .models import SkillProperties, SkillMetadata, ParseError, ValidationError
+        return locals()[name]
+    
+    if name in ("parse_frontmatter", "find_skill_md", "read_properties"):
+        from .parser import parse_frontmatter, find_skill_md, read_properties
+        return locals()[name]
+    
+    if name in ("validate", "validate_metadata", "_validate_name", "_validate_description", "_validate_compatibility"):
+        from .validator import validate, validate_metadata, _validate_name, _validate_description, _validate_compatibility
+        return locals()[name]
+    
+    if name in ("to_prompt", "generate_skills_xml", "format_skill_for_prompt"):
+        from .prompt import to_prompt, generate_skills_xml, format_skill_for_prompt
+        return locals()[name]
+    
+    if name in ("discover_skills", "get_default_skill_dirs"):
+        from .discovery import discover_skills, get_default_skill_dirs
+        return locals()[name]
+    
+    if name == "SkillLoader":
+        from .loader import SkillLoader
+        return SkillLoader
+    
+    if name == "SkillManager":
+        from .manager import SkillManager
+        return SkillManager
+    
+    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")

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

@@ -0,0 +1,129 @@
+"""Skill discovery and directory scanning."""
+
+from pathlib import Path
+from typing import List, Optional
+
+from .parser import find_skill_md, read_properties
+from .models import SkillProperties
+
+
+def get_default_skill_dirs() -> List[Path]:
+    """Get default skill directory locations.
+
+    Returns directories in precedence order (high to low):
+    1. Project: ./.praison/skills/ or ./.claude/skills/
+    2. User: ~/.praison/skills/
+    3. System: /etc/praison/skills/ (admin-managed)
+
+    Returns:
+        List of existing skill directories
+    """
+    dirs = []
+    cwd = Path.cwd()
+
+    # Project-level directories
+    project_dirs = [
+        cwd / ".praison" / "skills",
+        cwd / ".claude" / "skills",
+    ]
+    for d in project_dirs:
+        if d.exists() and d.is_dir():
+            dirs.append(d)
+
+    # User-level directory
+    user_dir = Path.home() / ".praison" / "skills"
+    if user_dir.exists() and user_dir.is_dir():
+        dirs.append(user_dir)
+
+    # System-level directory (Unix-like systems)
+    system_dir = Path("/etc/praison/skills")
+    if system_dir.exists() and system_dir.is_dir():
+        dirs.append(system_dir)
+
+    return dirs
+
+
+def discover_skills(
+    skill_dirs: Optional[List[str]] = None,
+    include_defaults: bool = True,
+) -> List[SkillProperties]:
+    """Discover all valid skills in the given directories.
+
+    Args:
+        skill_dirs: List of directory paths to scan for skills.
+            Each directory should contain skill subdirectories.
+        include_defaults: Whether to include default skill directories
+
+    Returns:
+        List of SkillProperties for all valid skills found
+    """
+    all_dirs = []
+
+    # Add explicit directories
+    if skill_dirs:
+        for d in skill_dirs:
+            path = Path(d).expanduser().resolve()
+            if path.exists() and path.is_dir():
+                all_dirs.append(path)
+
+    # Add default directories
+    if include_defaults:
+        all_dirs.extend(get_default_skill_dirs())
+
+    # Remove duplicates while preserving order
+    seen = set()
+    unique_dirs = []
+    for d in all_dirs:
+        if d not in seen:
+            seen.add(d)
+            unique_dirs.append(d)
+
+    skills = []
+
+    for parent_dir in unique_dirs:
+        # Each subdirectory in parent_dir might be a skill
+        try:
+            for item in parent_dir.iterdir():
+                if not item.is_dir():
+                    continue
+
+                # Check if this directory contains a SKILL.md
+                skill_md = find_skill_md(item)
+                if skill_md is None:
+                    continue
+
+                try:
+                    props = read_properties(item)
+                    skills.append(props)
+                except Exception:
+                    # Skip invalid skills
+                    continue
+        except PermissionError:
+            # Skip directories we can't read
+            continue
+
+    return skills
+
+
+def discover_skill(skill_path: str) -> Optional[SkillProperties]:
+    """Discover a single skill from a directory path.
+
+    Args:
+        skill_path: Path to a skill directory
+
+    Returns:
+        SkillProperties if valid, None otherwise
+    """
+    path = Path(skill_path).expanduser().resolve()
+
+    if not path.exists() or not path.is_dir():
+        return None
+
+    skill_md = find_skill_md(path)
+    if skill_md is None:
+        return None
+
+    try:
+        return read_properties(path)
+    except Exception:
+        return None

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

@@ -0,0 +1,236 @@
+"""Progressive skill loading for Agent Skills."""
+
+from pathlib import Path
+from typing import Optional
+from dataclasses import dataclass, field
+
+from .parser import find_skill_md, read_properties, parse_frontmatter
+from .models import SkillProperties, SkillMetadata
+
+
+@dataclass
+class LoadedSkill:
+    """A fully or partially loaded skill.
+
+    Supports progressive disclosure:
+    - Level 1: metadata only (name, description, location)
+    - Level 2: full instructions (SKILL.md body)
+    - Level 3: resources (scripts, references, assets)
+    """
+
+    properties: SkillProperties
+    instructions: Optional[str] = None
+    resources_loaded: bool = False
+    _scripts: dict = field(default_factory=dict)
+    _references: dict = field(default_factory=dict)
+    _assets: dict = field(default_factory=dict)
+
+    @property
+    def metadata(self) -> SkillMetadata:
+        """Get lightweight metadata for system prompt."""
+        return SkillMetadata.from_properties(self.properties)
+
+    @property
+    def is_activated(self) -> bool:
+        """Check if skill has been activated (instructions loaded)."""
+        return self.instructions is not None
+
+    def get_scripts(self) -> dict:
+        """Get loaded scripts."""
+        return self._scripts
+
+    def get_references(self) -> dict:
+        """Get loaded references."""
+        return self._references
+
+    def get_assets(self) -> dict:
+        """Get loaded assets."""
+        return self._assets
+
+
+class SkillLoader:
+    """Loader for Agent Skills with progressive disclosure support.
+
+    Usage:
+        loader = SkillLoader()
+
+        # Level 1: Load metadata only
+        skill = loader.load_metadata("/path/to/skill")
+
+        # Level 2: Activate skill (load instructions)
+        loader.activate(skill)
+
+        # Level 3: Load specific resources
+        loader.load_scripts(skill)
+        loader.load_references(skill)
+    """
+
+    def load_metadata(self, skill_path: str) -> Optional[LoadedSkill]:
+        """Load skill metadata only (Level 1).
+
+        This is the minimal load for system prompt injection.
+        Only reads the YAML frontmatter, not the full SKILL.md body.
+
+        Args:
+            skill_path: Path to skill directory
+
+        Returns:
+            LoadedSkill with metadata only, or None if invalid
+        """
+        path = Path(skill_path).expanduser().resolve()
+
+        if not path.exists() or not path.is_dir():
+            return None
+
+        try:
+            props = read_properties(path)
+            return LoadedSkill(properties=props)
+        except Exception:
+            return None
+
+    def activate(self, skill: LoadedSkill) -> bool:
+        """Activate a skill by loading its full instructions (Level 2).
+
+        Args:
+            skill: LoadedSkill instance to activate
+
+        Returns:
+            True if activation succeeded
+        """
+        if skill.is_activated:
+            return True
+
+        if skill.properties.path is None:
+            return False
+
+        skill_md = find_skill_md(skill.properties.path)
+        if skill_md is None:
+            return False
+
+        try:
+            content = skill_md.read_text()
+            _, body = parse_frontmatter(content)
+            skill.instructions = body
+            return True
+        except Exception:
+            return False
+
+    def load_scripts(self, skill: LoadedSkill) -> dict:
+        """Load scripts from skill's scripts/ directory (Level 3).
+
+        Args:
+            skill: LoadedSkill instance
+
+        Returns:
+            Dict mapping script names to their contents
+        """
+        if skill.properties.path is None:
+            return {}
+
+        scripts_dir = skill.properties.path / "scripts"
+        if not scripts_dir.exists() or not scripts_dir.is_dir():
+            return {}
+
+        scripts = {}
+        try:
+            for script_file in scripts_dir.iterdir():
+                if script_file.is_file():
+                    try:
+                        scripts[script_file.name] = script_file.read_text()
+                    except Exception:
+                        continue
+        except PermissionError:
+            pass
+
+        skill._scripts = scripts
+        return scripts
+
+    def load_references(self, skill: LoadedSkill) -> dict:
+        """Load references from skill's references/ directory (Level 3).
+
+        Args:
+            skill: LoadedSkill instance
+
+        Returns:
+            Dict mapping reference names to their contents
+        """
+        if skill.properties.path is None:
+            return {}
+
+        refs_dir = skill.properties.path / "references"
+        if not refs_dir.exists() or not refs_dir.is_dir():
+            return {}
+
+        refs = {}
+        try:
+            for ref_file in refs_dir.iterdir():
+                if ref_file.is_file():
+                    try:
+                        refs[ref_file.name] = ref_file.read_text()
+                    except Exception:
+                        continue
+        except PermissionError:
+            pass
+
+        skill._references = refs
+        return refs
+
+    def load_assets(self, skill: LoadedSkill) -> dict:
+        """Load asset paths from skill's assets/ directory (Level 3).
+
+        Note: For binary assets, only paths are returned, not contents.
+
+        Args:
+            skill: LoadedSkill instance
+
+        Returns:
+            Dict mapping asset names to their paths
+        """
+        if skill.properties.path is None:
+            return {}
+
+        assets_dir = skill.properties.path / "assets"
+        if not assets_dir.exists() or not assets_dir.is_dir():
+            return {}
+
+        assets = {}
+        try:
+            for asset_file in assets_dir.iterdir():
+                if asset_file.is_file():
+                    assets[asset_file.name] = str(asset_file)
+        except PermissionError:
+            pass
+
+        skill._assets = assets
+        skill.resources_loaded = True
+        return assets
+
+    def load_all_resources(self, skill: LoadedSkill) -> None:
+        """Load all resources for a skill (Level 3 complete).
+
+        Args:
+            skill: LoadedSkill instance
+        """
+        self.load_scripts(skill)
+        self.load_references(skill)
+        self.load_assets(skill)
+        skill.resources_loaded = True
+
+    @classmethod
+    def load(cls, skill_path: str, activate: bool = False) -> Optional[LoadedSkill]:
+        """Convenience method to load a skill.
+
+        Args:
+            skill_path: Path to skill directory
+            activate: Whether to also load instructions
+
+        Returns:
+            LoadedSkill or None if invalid
+        """
+        loader = cls()
+        skill = loader.load_metadata(skill_path)
+
+        if skill and activate:
+            loader.activate(skill)
+
+        return skill

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

@@ -0,0 +1,189 @@
+"""SkillManager for Agent Skills integration."""
+
+from typing import List, Optional, Dict
+
+from .models import SkillMetadata
+from .discovery import discover_skills
+from .loader import SkillLoader, LoadedSkill
+from .prompt import generate_skills_xml
+
+
+class SkillManager:
+    """Manager for discovering, loading, and using Agent Skills.
+
+    This is the main entry point for Agent Skills integration.
+    Supports progressive disclosure and zero performance impact
+    when skills are not used.
+
+    Usage:
+        manager = SkillManager()
+
+        # Discover skills from directories
+        manager.discover(["./skills", "~/.praison/skills"])
+
+        # Get XML for system prompt
+        prompt_xml = manager.to_prompt()
+
+        # Activate a specific skill
+        skill = manager.get_skill("pdf-processing")
+        manager.activate(skill)
+    """
+
+    def __init__(self):
+        """Initialize the SkillManager."""
+        self._skills: Dict[str, LoadedSkill] = {}
+        self._loader = SkillLoader()
+        self._discovered = False
+
+    @property
+    def skills(self) -> List[LoadedSkill]:
+        """Get all loaded skills."""
+        return list(self._skills.values())
+
+    @property
+    def skill_names(self) -> List[str]:
+        """Get names of all loaded skills."""
+        return list(self._skills.keys())
+
+    def discover(
+        self,
+        skill_dirs: Optional[List[str]] = None,
+        include_defaults: bool = True,
+    ) -> int:
+        """Discover skills from directories.
+
+        Args:
+            skill_dirs: List of directory paths to scan
+            include_defaults: Whether to include default skill directories
+
+        Returns:
+            Number of skills discovered
+        """
+        props_list = discover_skills(skill_dirs, include_defaults)
+
+        for props in props_list:
+            if props.name not in self._skills:
+                self._skills[props.name] = LoadedSkill(properties=props)
+
+        self._discovered = True
+        return len(props_list)
+
+    def add_skill(self, skill_path: str) -> Optional[LoadedSkill]:
+        """Add a single skill from a directory path.
+
+        Args:
+            skill_path: Path to skill directory
+
+        Returns:
+            LoadedSkill if successful, None otherwise
+        """
+        skill = self._loader.load_metadata(skill_path)
+        if skill:
+            self._skills[skill.properties.name] = skill
+        return skill
+
+    def get_skill(self, name: str) -> Optional[LoadedSkill]:
+        """Get a skill by name.
+
+        Args:
+            name: Skill name
+
+        Returns:
+            LoadedSkill or None if not found
+        """
+        return self._skills.get(name)
+
+    def activate(self, skill: LoadedSkill) -> bool:
+        """Activate a skill (load its instructions).
+
+        Args:
+            skill: LoadedSkill to activate
+
+        Returns:
+            True if activation succeeded
+        """
+        return self._loader.activate(skill)
+
+    def activate_by_name(self, name: str) -> bool:
+        """Activate a skill by name.
+
+        Args:
+            name: Skill name
+
+        Returns:
+            True if activation succeeded
+        """
+        skill = self.get_skill(name)
+        if skill is None:
+            return False
+        return self.activate(skill)
+
+    def get_available_skills(self) -> List[SkillMetadata]:
+        """Get metadata for all available skills.
+
+        This is used for system prompt injection.
+
+        Returns:
+            List of SkillMetadata instances
+        """
+        return [skill.metadata for skill in self._skills.values()]
+
+    def to_prompt(self) -> str:
+        """Generate XML prompt for available skills.
+
+        Returns:
+            XML string with <available_skills> block
+        """
+        metadata_list = self.get_available_skills()
+        return generate_skills_xml(metadata_list)
+
+    def get_instructions(self, name: str) -> Optional[str]:
+        """Get instructions for a skill, activating if needed.
+
+        Args:
+            name: Skill name
+
+        Returns:
+            Skill instructions or None if not found
+        """
+        skill = self.get_skill(name)
+        if skill is None:
+            return None
+
+        if not skill.is_activated:
+            self.activate(skill)
+
+        return skill.instructions
+
+    def load_resources(self, name: str) -> bool:
+        """Load all resources for a skill.
+
+        Args:
+            name: Skill name
+
+        Returns:
+            True if resources loaded successfully
+        """
+        skill = self.get_skill(name)
+        if skill is None:
+            return False
+
+        self._loader.load_all_resources(skill)
+        return True
+
+    def clear(self) -> None:
+        """Clear all loaded skills."""
+        self._skills.clear()
+        self._discovered = False
+
+    def __len__(self) -> int:
+        """Get number of loaded skills."""
+        return len(self._skills)
+
+    def __contains__(self, name: str) -> bool:
+        """Check if a skill is loaded."""
+        return name in self._skills
+
+    def __iter__(self):
+        """Iterate over loaded skills."""
+        return iter(self._skills.values())

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

@@ -0,0 +1,95 @@
+"""Data models for Agent Skills."""
+
+from dataclasses import dataclass, field
+from typing import Optional
+from pathlib import Path
+
+
+class ParseError(Exception):
+    """Raised when SKILL.md parsing fails."""
+    pass
+
+
+class ValidationError(Exception):
+    """Raised when skill validation fails."""
+    pass
+
+
+@dataclass
+class SkillProperties:
+    """Properties parsed from a skill's SKILL.md frontmatter.
+
+    Attributes:
+        name: Skill name in kebab-case (required)
+        description: What the skill does and when the model should use it (required)
+        license: License for the skill (optional)
+        compatibility: Compatibility information for the skill (optional)
+        allowed_tools: Tool patterns the skill requires (optional, experimental)
+        metadata: Key-value pairs for client-specific properties (defaults to
+            empty dict; omitted from to_dict() output when empty)
+        path: Path to the skill directory (optional, added for PraisonAI)
+    """
+
+    name: str
+    description: str
+    license: Optional[str] = None
+    compatibility: Optional[str] = None
+    allowed_tools: Optional[str] = None
+    metadata: dict = field(default_factory=dict)
+    path: Optional[Path] = None
+
+    def to_dict(self) -> dict:
+        """Convert to dictionary, excluding None values and empty metadata."""
+        result = {"name": self.name, "description": self.description}
+        if self.license is not None:
+            result["license"] = self.license
+        if self.compatibility is not None:
+            result["compatibility"] = self.compatibility
+        if self.allowed_tools is not None:
+            result["allowed-tools"] = self.allowed_tools
+        if self.metadata:
+            result["metadata"] = self.metadata
+        return result
+
+
+@dataclass
+class SkillMetadata:
+    """Lightweight skill metadata for system prompt injection.
+    
+    This is the minimal information needed to include a skill in the
+    system prompt's <available_skills> block (~50-100 tokens per skill).
+
+    Attributes:
+        name: Skill name
+        description: What the skill does and when to use it
+        location: Absolute path to the SKILL.md file
+    """
+
+    name: str
+    description: str
+    location: str
+
+    @classmethod
+    def from_properties(cls, props: SkillProperties) -> "SkillMetadata":
+        """Create SkillMetadata from SkillProperties.
+        
+        Args:
+            props: SkillProperties instance with path set
+            
+        Returns:
+            SkillMetadata with location pointing to SKILL.md
+        """
+        if props.path:
+            # Look for SKILL.md or skill.md
+            skill_md = props.path / "SKILL.md"
+            if not skill_md.exists():
+                skill_md = props.path / "skill.md"
+            location = str(skill_md)
+        else:
+            location = ""
+        
+        return cls(
+            name=props.name,
+            description=props.description,
+            location=location
+        )

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

@@ -0,0 +1,165 @@
+"""YAML frontmatter parsing for SKILL.md files."""
+
+from pathlib import Path
+from typing import Optional, Tuple
+import re
+
+from .models import ParseError, ValidationError, SkillProperties
+
+
+def find_skill_md(skill_dir: Path) -> Optional[Path]:
+    """Find the SKILL.md file in a skill directory.
+
+    Prefers SKILL.md (uppercase) but accepts skill.md (lowercase).
+
+    Args:
+        skill_dir: Path to the skill directory
+
+    Returns:
+        Path to the SKILL.md file, or None if not found
+    """
+    for name in ("SKILL.md", "skill.md"):
+        path = skill_dir / name
+        if path.exists():
+            return path
+    return None
+
+
+def parse_frontmatter(content: str) -> Tuple[dict, str]:
+    """Parse YAML frontmatter from SKILL.md content.
+
+    Args:
+        content: Raw content of SKILL.md file
+
+    Returns:
+        Tuple of (metadata dict, markdown body)
+
+    Raises:
+        ParseError: If frontmatter is missing or invalid
+    """
+    if not content.startswith("---"):
+        raise ParseError("SKILL.md must start with YAML frontmatter (---)")
+
+    parts = content.split("---", 2)
+    if len(parts) < 3:
+        raise ParseError("SKILL.md frontmatter not properly closed with ---")
+
+    frontmatter_str = parts[1]
+    body = parts[2].strip()
+
+    try:
+        metadata = _parse_yaml(frontmatter_str)
+    except Exception as e:
+        raise ParseError(f"Invalid YAML in frontmatter: {e}")
+
+    if not isinstance(metadata, dict):
+        raise ParseError("SKILL.md frontmatter must be a YAML mapping")
+
+    # Convert metadata values to strings where needed
+    if "metadata" in metadata and isinstance(metadata["metadata"], dict):
+        metadata["metadata"] = {str(k): str(v) for k, v in metadata["metadata"].items()}
+
+    return metadata, body
+
+
+def _parse_yaml(yaml_str: str) -> dict:
+    """Parse YAML string to dictionary.
+    
+    Uses a simple parser to avoid strictyaml dependency.
+    Falls back to yaml if available.
+    
+    Args:
+        yaml_str: YAML string to parse
+        
+    Returns:
+        Parsed dictionary
+    """
+    # Try to use PyYAML if available
+    try:
+        import yaml
+        return yaml.safe_load(yaml_str) or {}
+    except ImportError:
+        pass
+    
+    # Simple fallback parser for basic YAML
+    result = {}
+    current_key = None
+    current_dict = None
+    
+    for line in yaml_str.strip().split("\n"):
+        if not line.strip():
+            continue
+            
+        # Check for nested dict
+        if line.startswith("  ") and current_dict is not None:
+            match = re.match(r"^\s+(\w+):\s*(.*)$", line)
+            if match:
+                key, value = match.groups()
+                current_dict[key] = value.strip().strip('"').strip("'")
+            continue
+        
+        # Top-level key
+        match = re.match(r"^(\S+):\s*(.*)$", line)
+        if match:
+            key, value = match.groups()
+            value = value.strip()
+            
+            if not value:
+                # Could be start of nested dict
+                result[key] = {}
+                current_key = key
+                current_dict = result[key]
+            else:
+                result[key] = value.strip('"').strip("'")
+                current_dict = None
+    
+    return result
+
+
+def read_properties(skill_dir: Path) -> SkillProperties:
+    """Read skill properties from SKILL.md frontmatter.
+
+    This function parses the frontmatter and returns properties.
+    It does NOT perform full validation. Use validate() for that.
+
+    Args:
+        skill_dir: Path to the skill directory
+
+    Returns:
+        SkillProperties with parsed metadata
+
+    Raises:
+        ParseError: If SKILL.md is missing or has invalid YAML
+        ValidationError: If required fields (name, description) are missing
+    """
+    skill_dir = Path(skill_dir)
+    skill_md = find_skill_md(skill_dir)
+
+    if skill_md is None:
+        raise ParseError(f"SKILL.md not found in {skill_dir}")
+
+    content = skill_md.read_text()
+    metadata, _ = parse_frontmatter(content)
+
+    if "name" not in metadata:
+        raise ValidationError("Missing required field in frontmatter: name")
+    if "description" not in metadata:
+        raise ValidationError("Missing required field in frontmatter: description")
+
+    name = metadata["name"]
+    description = metadata["description"]
+
+    if not isinstance(name, str) or not name.strip():
+        raise ValidationError("Field 'name' must be a non-empty string")
+    if not isinstance(description, str) or not description.strip():
+        raise ValidationError("Field 'description' must be a non-empty string")
+
+    return SkillProperties(
+        name=name.strip(),
+        description=description.strip(),
+        license=metadata.get("license"),
+        compatibility=metadata.get("compatibility"),
+        allowed_tools=metadata.get("allowed-tools"),
+        metadata=metadata.get("metadata") or {},
+        path=skill_dir,
+    )

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

@@ -0,0 +1,74 @@
+"""XML prompt generation for Agent Skills."""
+
+from pathlib import Path
+from typing import List
+import html
+
+from .models import SkillMetadata
+from .parser import read_properties
+
+
+def format_skill_for_prompt(skill: SkillMetadata) -> str:
+    """Format a single skill as XML for system prompt.
+
+    Args:
+        skill: SkillMetadata instance
+
+    Returns:
+        XML string for the skill
+    """
+    name = html.escape(skill.name)
+    description = html.escape(skill.description)
+    location = html.escape(skill.location)
+
+    return f"""  <skill>
+    <name>{name}</name>
+    <description>{description}</description>
+    <location>{location}</location>
+  </skill>"""
+
+
+def generate_skills_xml(skills: List[SkillMetadata]) -> str:
+    """Generate XML block for available skills.
+
+    Args:
+        skills: List of SkillMetadata instances
+
+    Returns:
+        XML string with <available_skills> block
+    """
+    if not skills:
+        return "<available_skills>\n</available_skills>"
+
+    skill_entries = "\n".join(format_skill_for_prompt(s) for s in skills)
+    return f"<available_skills>\n{skill_entries}\n</available_skills>"
+
+
+def to_prompt(skill_dirs: List[Path]) -> str:
+    """Generate prompt XML from skill directories.
+
+    This is the main entry point for generating the <available_skills>
+    block to inject into agent system prompts.
+
+    Args:
+        skill_dirs: List of paths to skill directories
+
+    Returns:
+        XML string with <available_skills> block
+    """
+    skills = []
+
+    for skill_dir in skill_dirs:
+        skill_dir = Path(skill_dir)
+        if not skill_dir.exists() or not skill_dir.is_dir():
+            continue
+
+        try:
+            props = read_properties(skill_dir)
+            meta = SkillMetadata.from_properties(props)
+            skills.append(meta)
+        except Exception:
+            # Skip invalid skills silently
+            continue
+
+    return generate_skills_xml(skills)

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

@@ -0,0 +1,176 @@
+"""Skill validation logic."""
+
+import unicodedata
+from pathlib import Path
+from typing import Optional, List
+
+from .models import ParseError
+from .parser import find_skill_md, parse_frontmatter
+
+MAX_SKILL_NAME_LENGTH = 64
+MAX_DESCRIPTION_LENGTH = 1024
+MAX_COMPATIBILITY_LENGTH = 500
+
+ALLOWED_FIELDS = {
+    "name",
+    "description",
+    "license",
+    "allowed-tools",
+    "metadata",
+    "compatibility",
+}
+
+
+def _validate_name(name: str, skill_dir: Optional[Path]) -> List[str]:
+    """Validate skill name format and directory match.
+
+    Skill names support i18n characters (Unicode letters) plus hyphens.
+    Names must be lowercase and cannot start/end with hyphens.
+    """
+    errors = []
+
+    if not name or not isinstance(name, str) or not name.strip():
+        errors.append("Field 'name' must be a non-empty string")
+        return errors
+
+    name = unicodedata.normalize("NFKC", name.strip())
+
+    if len(name) > MAX_SKILL_NAME_LENGTH:
+        errors.append(
+            f"Skill name '{name}' exceeds {MAX_SKILL_NAME_LENGTH} character limit "
+            f"({len(name)} chars)"
+        )
+
+    if name != name.lower():
+        errors.append(f"Skill name '{name}' must be lowercase")
+
+    if name.startswith("-") or name.endswith("-"):
+        errors.append("Skill name cannot start or end with a hyphen")
+
+    if "--" in name:
+        errors.append("Skill name cannot contain consecutive hyphens")
+
+    if not all(c.isalnum() or c == "-" for c in name):
+        errors.append(
+            f"Skill name '{name}' contains invalid characters. "
+            "Only letters, digits, and hyphens are allowed."
+        )
+
+    if skill_dir:
+        dir_name = unicodedata.normalize("NFKC", skill_dir.name)
+        if dir_name != name:
+            errors.append(
+                f"Directory name '{skill_dir.name}' must match skill name '{name}'"
+            )
+
+    return errors
+
+
+def _validate_description(description: str) -> List[str]:
+    """Validate description format."""
+    errors = []
+
+    if not description or not isinstance(description, str) or not description.strip():
+        errors.append("Field 'description' must be a non-empty string")
+        return errors
+
+    if len(description) > MAX_DESCRIPTION_LENGTH:
+        errors.append(
+            f"Description exceeds {MAX_DESCRIPTION_LENGTH} character limit "
+            f"({len(description)} chars)"
+        )
+
+    return errors
+
+
+def _validate_compatibility(compatibility: str) -> List[str]:
+    """Validate compatibility format."""
+    errors = []
+
+    if not isinstance(compatibility, str):
+        errors.append("Field 'compatibility' must be a string")
+        return errors
+
+    if len(compatibility) > MAX_COMPATIBILITY_LENGTH:
+        errors.append(
+            f"Compatibility exceeds {MAX_COMPATIBILITY_LENGTH} character limit "
+            f"({len(compatibility)} chars)"
+        )
+
+    return errors
+
+
+def _validate_metadata_fields(metadata: dict) -> List[str]:
+    """Validate that only allowed fields are present."""
+    errors = []
+
+    extra_fields = set(metadata.keys()) - ALLOWED_FIELDS
+    if extra_fields:
+        errors.append(
+            f"Unexpected fields in frontmatter: {', '.join(sorted(extra_fields))}. "
+            f"Only {sorted(ALLOWED_FIELDS)} are allowed."
+        )
+
+    return errors
+
+
+def validate_metadata(metadata: dict, skill_dir: Optional[Path] = None) -> List[str]:
+    """Validate parsed skill metadata.
+
+    This is the core validation function that works on already-parsed metadata,
+    avoiding duplicate file I/O when called from the parser.
+
+    Args:
+        metadata: Parsed YAML frontmatter dictionary
+        skill_dir: Optional path to skill directory (for name-directory match check)
+
+    Returns:
+        List of validation error messages. Empty list means valid.
+    """
+    errors = []
+    errors.extend(_validate_metadata_fields(metadata))
+
+    if "name" not in metadata:
+        errors.append("Missing required field in frontmatter: name")
+    else:
+        errors.extend(_validate_name(metadata["name"], skill_dir))
+
+    if "description" not in metadata:
+        errors.append("Missing required field in frontmatter: description")
+    else:
+        errors.extend(_validate_description(metadata["description"]))
+
+    if "compatibility" in metadata:
+        errors.extend(_validate_compatibility(metadata["compatibility"]))
+
+    return errors
+
+
+def validate(skill_dir: Path) -> List[str]:
+    """Validate a skill directory.
+
+    Args:
+        skill_dir: Path to the skill directory
+
+    Returns:
+        List of validation error messages. Empty list means valid.
+    """
+    skill_dir = Path(skill_dir)
+
+    if not skill_dir.exists():
+        return [f"Path does not exist: {skill_dir}"]
+
+    if not skill_dir.is_dir():
+        return [f"Not a directory: {skill_dir}"]
+
+    skill_md = find_skill_md(skill_dir)
+    if skill_md is None:
+        return ["Missing required file: SKILL.md"]
+
+    try:
+        content = skill_md.read_text()
+        metadata, _ = parse_frontmatter(content)
+    except ParseError as e:
+        return [str(e)]
+
+    return validate_metadata(metadata, skill_dir)

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

@@ -218,6 +218,10 @@ TOOL_MAPPINGS = {
     'crawl4ai_extract_sync': ('.crawl4ai_tools', None),
     'Crawl4AITools': ('.crawl4ai_tools', 'Crawl4AITools'),
     'crawl4ai_tools': ('.crawl4ai_tools', None),
+    
+    # Unified Web Search (auto-fallback across providers)
+    'search_web': ('.web_search', None),
+    'get_available_providers': ('.web_search', None),
 }
 
 _instances = {}  # Cache for class instances
@@ -254,7 +258,8 @@ def __getattr__(name: str) -> Any:
             'exa', 'exa_search', 'exa_search_contents', 'exa_find_similar', 'exa_answer',
             'exa_search_async', 'exa_search_contents_async', 'exa_answer_async',
             'crawl4ai', 'crawl4ai_many', 'crawl4ai_extract', 'crawl4ai_llm_extract',
-            'crawl4ai_sync', 'crawl4ai_extract_sync'
+            'crawl4ai_sync', 'crawl4ai_extract_sync',
+            'search_web', 'get_available_providers'
         ]:
             return getattr(module, name)
         if name in ['file_tools', 'pandas_tools', 'wikipedia_tools',

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

@@ -14,6 +14,7 @@ Usage:
     context, cleaned_prompt = parser.process("@file:main.py explain this code")
 """
 
+import os
 import re
 import logging
 from pathlib import Path
@@ -43,10 +44,14 @@ class MentionsParser:
         "url": re.compile(r'@url:(https?://[^\s]+)'),
     }
     
+    # Default max file chars: 500K (~125K tokens) - fits GPT-4o (128K), Claude 3.5 (200K)
+    DEFAULT_MAX_FILE_CHARS = 500000
+    
     def __init__(
         self,
         workspace_path: Optional[str] = None,
-        verbose: int = 0
+        verbose: int = 0,
+        max_file_chars: Optional[int] = None
     ):
         """
         Initialize MentionsParser.
@@ -54,10 +59,24 @@ class MentionsParser:
         Args:
             workspace_path: Path to workspace/project root
             verbose: Verbosity level
+            max_file_chars: Maximum characters to include from files.
+                           Default: 500000 (500K chars, ~125K tokens).
+                           Set to 0 for no limit.
+                           Can be overridden via PRAISON_MAX_FILE_CHARS env var.
         """
         self.workspace_path = Path(workspace_path) if workspace_path else Path.cwd()
         self.verbose = verbose
         
+        # Configure max file chars: constructor > env var > default
+        if max_file_chars is not None:
+            self.max_file_chars = max_file_chars
+        else:
+            env_val = os.environ.get('PRAISON_MAX_FILE_CHARS')
+            if env_val is not None:
+                self.max_file_chars = int(env_val)
+            else:
+                self.max_file_chars = self.DEFAULT_MAX_FILE_CHARS
+        
         # Lazy-loaded managers
         self._docs_manager = None
         self._rules_manager = None
@@ -158,9 +177,15 @@ class MentionsParser:
             
             content = full_path.read_text(encoding="utf-8")
             
-            # Limit content size
-            if len(content) > 50000:
-                content = content[:50000] + "\n... (truncated)"
+            # Limit content size (0 means no limit)
+            if self.max_file_chars > 0 and len(content) > self.max_file_chars:
+                original_size = len(content)
+                content = content[:self.max_file_chars] + "\n... (truncated)"
+                self._log(
+                    f"File {file_path} truncated from {original_size:,} to {self.max_file_chars:,} chars. "
+                    f"Set PRAISON_MAX_FILE_CHARS=0 for no limit.",
+                    logging.WARNING
+                )
             
             # Detect language for code fence
             ext = full_path.suffix.lower()
@@ -253,9 +278,9 @@ class MentionsParser:
             with urllib.request.urlopen(req, timeout=10) as response:
                 content = response.read().decode('utf-8', errors='ignore')
             
-            # Limit content size
-            if len(content) > 50000:
-                content = content[:50000] + "\n... (truncated)"
+            # Limit content size (0 means no limit)
+            if self.max_file_chars > 0 and len(content) > self.max_file_chars:
+                content = content[:self.max_file_chars] + "\n... (truncated)"
             
             # Try to extract text from HTML
             if '<html' in content.lower():