CrewAI | FishMem Docs

Give a multi-agent crew shared evidence without mixing agent identity and user scope.

CrewAI can coordinate several roles around one task. Keep three identities distinct:

the authenticated customer or account (user_id);
the durable crew/role identity (agent_id);
the current kickoff/execution (run_id).

Before kickoff, search user and role scope for the current objective and load Documents for source-backed context. After kickoff, write only verified decisions or outcomes; intermediate role chatter belongs in execution state.

context = client.memories.search({
    "query": objective,
    "user_id": authenticated_user_id,
    "agent_id": "research-crew",
    "top_k": 6,
})

# Run the crew with a labeled summary of context["results"].

if verified_outcome:
    receipt = client.memories.add_async(
        {
            "content": verified_outcome,
            "user_id": authenticated_user_id,
            "agent_id": "research-crew",
        },
        idempotency_key=f"crew:{execution_id}:outcome",
    )

Persist the returned event ID with the crew execution rather than blocking an interactive response on inference completion.