🐮 Calfkit

Build decentralized multi-agent systems. Agents discover each other at runtime, choreograph work, and scale as independent, event-driven services.

Calfkit agents find each other and choreograph work over a mesh — a highly-connected data streaming network they auto-discover and communicate on. Each agent runs as an independent, event-driven service, so you can build free-flowing multi-agent workflows that collaborate and react to live data streams.

Why Calfkit?

• Dynamic agent-to-agent discovery and collaboration. Agents find each other at runtime and work together — messaging each other and handing off tasks — so you build multi-agent systems without complex wiring or orchestration, and extend team capabilities at any time.
• Scalable by default. Every agent runs and scales as an independent microservice, so your agent teams are resilient and scalable from day one.
• React to realtime data streams. Agents are event-driven, so they react to realtime data streams — live market feeds, log streams, support-ticket queues — and send results wherever they're needed. Build agents that work like continuously streaming workflows, not one-off requests.
• Scale on production-ready infrastructure. The agent mesh is Kafka-compatible so you can run + scale your agents on production-ready streaming infrastructure straight out of the box.

Installation

# Recommended for getting started, includes a zero-setup in-memory dev mesh:
pip install 'calfkit[mesh]'

Quickstart

With the [mesh] extra, ck dev spins up a local in-memory mesh for you — no Docker, no CALFKIT_MESH_URL required.

Agent

Save as general.py:

from calfkit import Agent, Handoff, Messaging, OpenAIResponsesModelClient

general = Agent(
    name="general",
    description="Answers simple questions and routes requests to whoever can handle it.",
    system_prompt="You are a general assistant. Defer technical questions to other agents.",
    model_client=OpenAIResponsesModelClient(model_name="gpt-5.4"),
    peers=[
        Messaging(discover=True),  # discover and delegate to any agent at runtime
        Handoff(discover=True),    # discover and hand off to any agent at runtime
    ],
)

Run it and chat

# Starts the agent (and a local mesh if one isn't running):
#   ck dev run <file>:<agent>
ck dev run general:general

# In a second terminal, chat with the agent:
ck dev chat

Add another agent — and watch them discover each other

Save as finance.py:

from calfkit import Agent, OpenAIResponsesModelClient

finance = Agent(
    name="finance",
    description="Answers the user's personal finance questions.",
    system_prompt="You are the personal finance specialist. Answer finance-related questions.",
    model_client=OpenAIResponsesModelClient(model_name="gpt-5.4"),
)

ck dev run finance:finance

Now ask a finance question in ck dev chat — general discovers finance at runtime and hands off automatically. No wiring, no orchestrator.

Running an agent mesh

Calfkit agents discover and communicate over a mesh.

For local dev, the bundled in-memory broker (via [mesh] extra) is zero-setup — see How to run a local mesh with ck dev.

In production, the mesh is Kafka-compatible so you can drop your agent swarms into production-ready Kafka streaming infrastructure you already use.

Want a fully-managed mesh your agents can join from anywhere? Join the beta

Documentation

• Getting started: See docs/.
• Examples: See examples/ — multi-agent team and general framework API examples.

Contributing

Issues and pull requests are welcome. Please open an issue to discuss substantial changes before sending a PR.

See CONTRIBUTING.md.

License

This project is licensed under the Apache License 2.0. See the LICENSE file for details.