colber-crewai

CrewAI integration for the Colber platform — step / task observability callbacks, long-term semantic memory backed by colber-memory, and a 5-service toolkit. Apache-2.0.

This is the second framework plugin in the "Lego" GTM lever (after langchain-colber). It depends only on crewai>=0.80 and the published colber-sdk PyPI release, so it stays lightweight and is independently versionable.

Install

pip install colber-crewai

For local development inside the Colber monorepo:

pip install -e apps/colber-crewai

Components

ColberStepCallback + ColberTaskCallback

Captures CrewAI step and task events as Colber observability spans + structured logs. CrewAI exposes two hook points (plain callables, not a BaseCallbackHandler like LangChain):

• step_callback: invoked after each agent step (LLM call OR tool call).
• task_callback: invoked after each task completes.

Wire them on either an Agent or a Crew:

from crewai import Agent, Task, Crew
from colber_crewai import ColberStepCallback, ColberTaskCallback

step_cb = ColberStepCallback(
    agent_did="did:key:z6Mk...",
    operator_id="op-demo",
    service_name="my-crewai-agent",
)
task_cb = ColberTaskCallback(
    agent_did="did:key:z6Mk...",
    operator_id="op-demo",
    service_name="my-crewai-agent",
)

agent = Agent(
    role="Researcher",
    goal="...",
    backstory="...",
    step_callback=step_cb,
)
task = Task(description="...", agent=agent, callback=task_cb)
crew = Crew(agents=[agent], tasks=[task])
result = crew.kickoff()

Each invocation emits one span (/v1/observability/traces) with traceId, spanId, name, durationMs, status, and a small attributes dict. Errors additionally emit a structured log event (/v1/observability/logs) at level=error.

Network failures to the observability backend are caught, logged at WARNING, and swallowed — the crew is never aborted because telemetry is sick. Both callbacks are thread-safe (CrewAI may run agents concurrently).

ColberLongTermMemory

Backs CrewAI's LongTermMemory with the colber-memory service (Qdrant + ACL + chiffrement). CrewAI keeps short-term and entity memory native (in-memory / SQLite); we plug Colber on long-term only — the tier where Colber differentiates (cross-agent share, semantic search, encrypted at rest).

from crewai import Agent, Crew
from crewai.memory import EntityMemory, ShortTermMemory
from colber_crewai import ColberLongTermMemory

long_term = ColberLongTermMemory(
    agent_did="did:key:z6Mk...",
    top_k=5,
    share_with=["did:key:z6MkPeer1"],
)
crew = Crew(
    agents=[...],
    tasks=[...],
    memory=True,
    long_term_memory=long_term,
)

save(value, metadata) persists the value via colber-memory.store (auto-shared with the configured peer DIDs); search(query, limit) runs a semantic top-K query via colber-memory.query.

ColberToolkit

Exposes 5 Colber services as CrewAI BaseTool instances (one per operation, 14 in total).

from colber_crewai import ColberToolkit

toolkit = ColberToolkit(agent_did="did:key:z6Mk...")
tools = toolkit.get_tools()  # list[crewai_tools.BaseTool]

# Plug into any CrewAI agent.
agent = Agent(role="Negotiator", tools=tools, ...)

Pass services=["negotiation", "insurance"] to scope down the surface for a deal-only agent.

The observability service is not exposed as a tool. Letting an LLM call log_ingest is a footgun (the agent could DoS its own log pipeline). Use ColberStepCallback / ColberTaskCallback for telemetry — they give the agent first-class observability without LLM-driven calls. Passing services=["observability"] raises ValueError with the explicit reason.

Configuration

The plugin reads three environment variables when no explicit ColberClient is passed:

If none are set, the plugin falls back to ColberClient.local() (β-VM ports on localhost).

You can always pass a pre-built client explicitly:

from colber_sdk import ColberClient
from colber_crewai import ColberToolkit

client = ColberClient.from_base_url("https://api.colber.dev", auth_token="...")
toolkit = ColberToolkit(client=client, agent_did="did:key:z6Mk...")

CrewAI version requirement

crewai>=0.80,<1. The 0.80+ line stabilised the public extension points this plugin depends on (step_callback / task_callback plain-callable convention, LongTermMemory save/search storage interface). Older versions ship different signatures.

Out of scope (Wave 2.3+ follow-up)

• CrewAI Flows integration (declarative state machines).
• Custom prompt templates pre-baked with reputation/memory context.
• Attestation flow helpers (auto-sign + verify Colber payloads inside an agent loop).

These are noted as Wave 2.3+ follow-ups in the Colber ROADMAP.

License

Apache-2.0 — same license as colber-sdk, langchain-colber, and @colber/mcp.