🐮 Calfkit

Build agents that 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 CALFKITMESHURL 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.",     modelclient=OpenAIResponsesModelClient(modelname="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.",     modelclient=OpenAIResponsesModelClient(modelname="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