Python-native agent runtime · 0.2

Build agents you can see, steer, and trust.

AgentMuru gives engineers an observable runtime, governed tools, durable local history, and a replayable Workspace—without moving application logic into the browser.

Install AgentMuru Operate durable sessions

AgentMuru

Define agents and typed tools in Python. The Runtime coordinates model streams, tool calls, approvals, sessions, artifacts, workflows, usage, and traces. Muru Workspace projects the same ordered event history for operators.

from agentmuru import Agent, Application, FakeModel

application = Application(
    agent=Agent(
        name="assistant",
        instructions="Help the user.",
        model=FakeModel.responses("AgentMuru is ready."),
    )
)

See

Inspect ordered events, model deltas, tools, usage, artifacts, and traces.

Steer

Pause risky actions for an explicit approval, rejection, or expiry.

Trust

Reopen SQLite-backed sessions with honest interrupted-process semantics.

Start with the verified quickstart