🐮 Calfkit SDK

The SDK to build event-driven, distributed AI agents.

Calfkit lets you compose agents with independent services—chat, tools, routing—that communicate asynchronously. Add agent capabilities without coordination. Scale each component independently. Stream agent outputs to any downstream system.

Agents should work like real teams, with independent, distinct roles, async communication, and the ability to onboard new teammates or tools without restructuring the whole org. Build AI employees that integrate.

pip install calfkit

Why Event-Driven?

Building agents like traditional web applications, with tight coupling and synchronous API calls, creates the same scalability problems that plagued early microservices:

• Tight coupling: Changing one tool or agent breaks dependent agents and tools
• Scaling bottlenecks: Since all agents and tools live on one runtime, everything must scale together
• Siloed outputs: Agent and tool outputs stay trapped in your AI layer, streaming outputs to external dependencies is not as natural as distributed, event-driven designs

Event-driven architectures provide the solution. Instead of direct API calls between components, agents and tools asynchronously communicate. Each component runs independently, scales horizontally, and outputs can flow anywhere: CRMs, data warehouses, analytics platforms, other agents, or even more tools.

Why Use Calfkit?

Calfkit is a Python SDK that builds event-driven agents out-the-box. You get all the benefits of a asynchronous, distributed system (loose coupling, horizontal scalability, durability) without the complexity of managing event-driven infrastructure and orchestration yourself.

• Distributed agents out-the-box: Build event-driven, multi-service agents without writing orchestration code or managing infrastructure
• Add agent capabilities without touching existing code: Deploy new tool capabilities as independent services that agents can dynamically discover, no need to touch your agent code
• Scale what you need, when you need it: Chat handling, tool execution, and routing each scale independently based on demand
• Nothing gets lost: Event persistence ensures reliable message delivery and traceability, even during service failures or restarts
• High throughput under pressure: Asynchronous communication decouples requests from processing, so Calfkit agents work through bursty traffic reliably, maximizing throughput
• Real-time responses: Low-latency event processing enables agents to react instantly to incoming data
• Development team independence: Because of the decoupled design, dev teams can develop and deploy chat, tools, routing, and upstream or downstream dependencies in parallel without cross-team collaboration overhead
• Universal data flow: Decoupling enables data to flow freely in both directions.
  • Downstream, agent outputs can be streamed to any system (CRMs, customer data platforms, warehouses, or even another AI workflow).
  • Upstream, tools can wrap any data sources and deploy independently, no coordination needed.

Quick Start

Prerequisites

• Python 3.10 or later
• Docker installed and running (for local testing with a Calfkit broker)
• OpenAI API key (or another OpenAI API compliant LLM provider)

Install

pip install calfkit

☁️ Calfkit Cloud (Coming Soon)

Skip the infrastructure. Calfkit Cloud is a fully-managed Kafka service built for Calfkit AI agents and multi-agent teams. No server infrastructure to self-host or maintain, with built-in observability and agent-event tracing.

Coming soon. Fill out the interest form →

Start Local Calfkit Server (Requires Docker)

Calfkit uses Kafka as the event broker. Run the following command to clone the calfkit-broker repo and start a local Kafka broker container:

git clone https://github.com/calf-ai/calfkit-broker && cd calfkit-broker && make dev-up

Once the broker is ready, open a new terminal tab to continue with the quickstart.

Define and Deploy the Tool Node

Define and deploy a tool as an independent service. Tools are not owned by or coupled to any specific agent—once deployed, any agent in your system can discover and invoke the tool. Deploy once, use everywhere.

# weather_tool.py
import asyncio
from calfkit.nodes import agent_tool
from calfkit.broker import BrokerClient
from calfkit.runners import NodesService

# Example tool definition
@agent_tool
def get_weather(location: str) -> str:
    """Get the current weather at a location"""
    return f"It's sunny in {location}"

async def main():
    broker_client = BrokerClient(bootstrap_servers="localhost:9092") # Connect to Kafka broker
    service = NodesService(broker_client) # Initialize a service instance
    service.register_node(get_weather) # Register the tool node in the service
    await service.run() # (Blocking call) Deploy the service to start serving traffic

if __name__ == "__main__":
    asyncio.run(main())

Run the file to deploy the tool service:

python weather_tool.py

Deploy the Chat Node

Deploy the LLM chat node as its own service.

# chat_service.py
import asyncio
from calfkit.nodes import ChatNode
from calfkit.providers import OpenAIModelClient
from calfkit.broker import BrokerClient
from calfkit.runners import NodesService

async def main():
    broker_client = BrokerClient(bootstrap_servers="localhost:9092") # Connect to Kafka broker
    model_client = OpenAIModelClient(model_name="gpt-5-nano")
    chat_node = ChatNode(model_client) # Inject a model client into the chat node definition so the chat deployment can perform LLM calls
    service = NodesService(broker_client) # Initialize a service instance
    service.register_node(chat_node) # Register the chat node in the service
    await service.run() # (Blocking call) Deploy the service to start serving traffic

if __name__ == "__main__":
    asyncio.run(main())

Set your OpenAI API key:

export OPENAI_API_KEY=sk-...

Run the file to deploy the chat service:

python chat_service.py

Deploy the Agent Router Node

Deploy the agent router that orchestrates chat, tools, and conversation-level memory.

# agent_router_service.py
import asyncio
from calfkit.nodes import agent_tool, AgentRouterNode, ChatNode
from calfkit.stores import InMemoryMessageHistoryStore
from calfkit.broker import BrokerClient
from calfkit.runners import NodesService
from weather_tool import get_weather # Import the tool, the tool definition is reusable

async def main():
    broker_client = BrokerClient(bootstrap_servers="localhost:9092") # Connect to Kafka broker
    router_node = AgentRouterNode(
        chat_node=ChatNode(), # Provide the chat node definition for the router to orchestrate the nodes
        tool_nodes=[get_weather],
        system_prompt="You are a helpful assistant",
        message_history_store=InMemoryMessageHistoryStore(), # Stores messages in-memory in the deployment runtime
    )
    service = NodesService(broker_client) # Initialize a service instance
    service.register_node(router_node) # Register the router node in the service
    await service.run() # (Blocking call) Deploy the service to start serving traffic

if __name__ == "__main__":
    asyncio.run(main())

Run the file to deploy the agent router service:

python agent_router_service.py

Invoke the Agent

Send a request and receive the response.

When invoking an already-deployed agent, use the RouterServiceClient. The node is just a configuration object, so you don't need to redefine the deployment parameters.

# client.py
import asyncio
from calfkit.nodes import AgentRouterNode
from calfkit.broker import BrokerClient
from calfkit.runners import RouterServiceClient

async def main():
    broker_client = BrokerClient(bootstrap_servers="localhost:9092") # Connect to Kafka broker

    # Thin client - no deployment parameters needed
    router_node = AgentRouterNode()
    client = RouterServiceClient(broker_client, router_node)

    # Invoke and wait for response
    response = await client.invoke(user_prompt="What's the weather in Tokyo?")
    final_msg = await response.get_final_response()
    print(f"Assistant: {final_msg.text}")

if __name__ == "__main__":
    asyncio.run(main())

Run the file to invoke the agent:

python client.py

The RouterServiceClient handles ephemeral Kafka communication and cleanup automatically. You can also stream intermediate messages:

response = await client.invoke(user_prompt="What's the weather in Tokyo?")

# Stream all messages (tool calls, intermediate responses, etc.)
async for message in response.messages_stream():
    print(message)

Runtime Configuration (Optional)

Clients can override the system prompt and restrict available tools at invocation time without redeploying:

from weather_tool import get_weather

# Client with runtime patches
router_node = AgentRouterNode(
    system_prompt="You are an assistant with no tools :(",  # Override the deployed system prompt
    tool_nodes=[],  # Patch in any subset of the deployed agent's set of tools
)
client = RouterServiceClient(broker_client, router_node)
response = await client.invoke(user_prompt="Weather in Tokyo?")

This lets different clients customize agent behavior per-request. Tool patching is currently limited to subsets of tools configured in the deployed router.

Motivation

Scalable agent teams must progress beyond brittle, tightly coupled, synchronous coordination. This means embracing event-driven, asynchronous communication patterns between agents and their dependencies.

Contact

Support

If you found this project interesting or useful, please consider:

• ⭐ Starring the repository — it helps others discover it!
• 🐛 Reporting issues
• 🔀 Submitting PRs

License

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