AgentGuard CrewAI Integration

Automatic trace capture for CrewAI multi-agent workflows.

Installation

pip install agentguard-crewai

Quick Start

from crewai import Crew, Agent, Task
from agentguard_crewai import AgentGuardObserver

# Define your agents
researcher = Agent(
    role="Researcher",
    goal="Research and analyze topics",
    backstory="Expert researcher with deep analysis skills"
)

writer = Agent(
    role="Writer",
    goal="Write engaging content",
    backstory="Creative writer with storytelling expertise"
)

# Define tasks
research_task = Task(
    description="Research AI trends in 2026",
    agent=researcher,
    expected_output="Detailed research report"
)

writing_task = Task(
    description="Write article based on research",
    agent=writer,
    context=[research_task],
    expected_output="Published article"
)

# Create crew with AgentGuard observer
crew = Crew(
    agents=[researcher, writer],
    tasks=[research_task, writing_task]
)

observer = AgentGuardObserver(api_key="ag_live_your_key")
observer.register_crew_agents(crew)

# Run crew - traces automatically captured!
result = crew.kickoff()

# View traces in AgentGuard dashboard
print("✅ Execution complete! View traces at http://localhost:3000/dashboard")

Features

• ✅ Automatic agent registration - Agents registered in AgentGuard on first use
• ✅ Task execution tracking - Capture every task execution with timing
• ✅ Tool call monitoring - Track all tool uses by agents
• ✅ Multi-agent coordination - Capture delegation and collaboration events
• ✅ Error tracking - Automatic error capture and reporting
• ✅ State drift detection - Monitor inter-agent state changes
• ✅ PII redaction - Sensitive data automatically redacted

Advanced Usage

Custom Configuration

observer = AgentGuardObserver(
    api_key="ag_live_your_key",
    base_url="https://api.agentguard.io",  # Production API
    capture_tools=True,  # Capture tool calls
    capture_coordination=True,  # Capture agent coordination
    auto_register_agents=True  # Auto-register agents
)

Manual Event Tracking

# Track task start
observer.on_task_start(task, agent, crew)

# Track tool use
observer.on_tool_use(agent, tool, input_data, output_data)

# Track coordination
observer.on_agent_coordination(source_agent, target_agent, message, crew)

# Track completion
observer.on_task_complete(task, agent, output, crew)

# Track errors
observer.on_task_error(task, agent, error, crew)

Monkey-Patching (Automatic)

For automatic tracing without manual observer setup:

from agentguard_crewai import monkey_patch_crewai, AgentGuardObserver

# Patch CrewAI globally
observer = AgentGuardObserver(api_key="ag_live_...")
monkey_patch_crewai(observer)

# All Crew executions are now automatically traced
crew = Crew(agents=[...], tasks=[...])
crew.kickoff()  # Automatically traced!

What Gets Captured

Task Execution

{
    "task": "Research AI trends",
    "expected_output": "Research report",
    "context": ["Previous task outputs"],
    "tools": ["WebSearchTool", "ScraperTool"],
    "duration_ms": 12500,
    "status": "success"
}

Multi-Agent Coordination

{
    "coordination": {
        "allow_delegation": True,
        "crew_size": 3,
        "agent_role": "Researcher",
        "events": [
            {
                "from": "Researcher",
                "to": "Writer",
                "message": "Here's the research data",
                "timestamp": 1737645123.45
            }
        ]
    }
}

Tool Calls

{
    "tool_calls": [
        {
            "tool": "WebSearchTool",
            "input": "AI trends 2026",
            "output": "Found 100 articles...",
            "timestamp": 1737645120.12
        }
    ]
}

Dashboard Integration

View all captured traces in AgentGuard dashboard:

• Timeline view - See task execution sequence
• Coordination graph - Visualize agent interactions
• Tool usage - Track which tools are used most
• Error analysis - Identify failure patterns
• Performance metrics - Optimize slow tasks

Support

• 📚 Docs: https://docs.agentguard.io/integrations/crewai
• 💬 Discord: https://discord.gg/agentguard
• 🐛 Issues: https://github.com/agentguard/crewai-integration/issues

License

MIT