fluxos-agent 2.0.1

Stateful agent runtime with tool execution and event streaming

fluxos-agent

Python port of pi-mono/packages/agent [https://github.com/badlogic/pi-mono/tree/main/packages/agent].

fluxos-agent provides:

• Stateful agent runtime (Agent) with prompt/continue flows
• Agent loop primitives (agent_loop, agent_loop_continue)
• Tool execution with streaming update events
• Steering and follow-up queues
• Optional proxy stream adapter (stream_proxy)

Installation

Install from PyPI (recommended):

uv add fluxos-agent
# or: pip install fluxos-agent

Local editable install (repo development):

pip install -e packages/fluxos-agent

Quick Start

import asyncio

from flux_agent import Agent, AgentOptions
from flux_agent.types import (
    AssistantDoneEvent,
    AssistantMessage,
    Context,
    Model,
    TextContent,
    Usage,
    UsageCost,
)
from flux_agent.event_stream import EventStream


class MockAssistantStream(EventStream):
    def __init__(self):
        super().__init__(
            lambda event: event.type in {"done", "error"},
            lambda event: event.message if event.type == "done" else event.error,
        )


def stream_fn(model: Model, context: Context, options):
    stream = MockAssistantStream()
    message = AssistantMessage(
        content=[TextContent(text="Hello from Python agent")],
        api=model.api,
        provider=model.provider,
        model=model.id,
        usage=Usage(cost=UsageCost()),
        stop_reason="stop",
    )
    stream.push(AssistantDoneEvent(reason="stop", message=message))
    return stream


async def main():
    agent = Agent(
        AgentOptions(
            stream_fn=stream_fn,
            initial_state={
                "system_prompt": "You are helpful.",
                "model": Model(id="mock", name="mock", api="mock", provider="mock"),
            },
        )
    )

    agent.subscribe(lambda event: print(event.type))
    await agent.prompt("Hello")


asyncio.run(main())

Testing

cd packages/fluxos-agent
pip install -e ".[dev]"
pytest tests/unit -v

Run full suite (unit + e2e scaffolding):

pytest tests -v

Release Tracking

• Keep package changes tracked in CHANGELOG.md.
• For every release:
  1. Update pyproject.toml version.
  2. Add a new dated section in CHANGELOG.md.
  3. Publish and tag the release.