[FEATURE] Native integration with Agentmetry for AI SIEM / Crew Audit Logging

[blitzcrieg1]

Feature Area

Integration with external tools

Is your feature request related to a an existing bug? Please link it here.

NA

Describe the solution you'd like

Hi team! Incredible work on CrewAI. As more enterprises adopt Crews for complex workflows, I'm seeing a lot of concern around security, auditability, and tracking exactly what tools the agents are executing in production.

I am the maintainer of Agentmetry, an open-source flight recorder / SIEM designed specifically for AI agents.

I would love to contribute a PR that adds an optional AgentmetryCrewCallback to the framework. This would allow users to:

1. Stream all tool executions from every agent in a Crew directly to a local Agentmetry dashboard.
2. Automatically tag agent actions with MITRE ATT&CK labels to detect risky behavior.
3. Provide a complete, cryptographically hashed audit trail of everything the Crew did during its execution.

Describe alternatives you've considered

Currently, users have to rely on standard print statements, LangSmith, or Datadog for observability, but none of those tools provide SIEM-level security monitoring or MITRE ATT&CK tagging for agentic actions out of the box.

Additional context

You can check out the Agentmetry project here: https://github.com/blitzcrieg1/agentmetry. It is fully open-source and runs locally via Docker.

Willingness to Contribute

Yes, I'd be happy to submit a pull request

[blitzcrieg1]

Update: I built this, and it needs no changes to CrewAI at all, so there is nothing here for you to merge or maintain.

It is a plain BaseEventListener on the public crewai_event_bus, subscribing to CrewKickoffStarted/Completed/Failed and ToolUsageStarted/Finished/Error:

from agentmetry_crewai import AgentmetryListener

agentmetry = AgentmetryListener() # keep the reference alive

crew = Crew(agents=[...], tasks=[...])
crew.kickoff()
Each kickoff becomes one correlation_id, so sequence detections fire across a whole Crew rather than a single tool call. Tool arguments are hashed and never transmitted unless you opt in.

Code: https://github.com/blitzcrieg1/agentmetry/tree/master/adapters/crewai

I withdraw the PR offer from my original post. I have since read #5531, where integration docs were closed as reserved for established integrations chosen by maintainers, and that is a fair line to hold. I will put this on the community forum instead. Feel free to close this issue.