For Python devs

Pythonic on the surface. Rust where it counts.

Type-checked, async-first, pip install-able. The runtime is compiled Rust under a PyO3 wheel — no GIL contention in the hot path.

Install

pip install agentmatic
# Prebuilt wheels for CPython 3.10–3.13 on Linux/macOS/Windows × x64/arm64.

Five-line agent

from agentmatic import Agent, OpenAI, tool

@tool
def calculator(expression: str) -> str:
    """Evaluate a math expression."""
    return str(eval(expression))

agent = Agent.build("math-helper", llm=OpenAI(), tools=[calculator])
print(agent.invoke("What is 42 * 17?"))

What's different from pure-Python frameworks

• GIL doesn't matter in the hot path. The Pregel scheduler is Rust; tool calls re-enter Python only at the boundary.
• Cold start ≈ 80 ms on the prebuilt wheel. No lazy-import-the-world tax.
• Async-first. Every method has ainvoke / astream variants. Tokio runtime under the hood, asyncio bridge.
• Type-hinted end-to-end. StateGraph[YourState] is generic; mypy + pyright catch invalid edges.

What we ship for Python specifically

• Pydantic v2 model support for state types.
• FastAPI adapter — mount any agent at a route.
• Jupyter rendering — graphs visualize inline; checkpoint timeline as a widget.
• uvloop auto-detection for max throughput.