---

async-pykka is a pure asyncio framework that implements the Actor model, inspired by Pykka-style APIs and focused on high-throughput I/O workloads.

Quick Start Example | Scenarios | Performance | FAQ

[!IMPORTANT] Same-loop constraint All actor operations must run in the same event loop. Cross-thread or cross-loop calls fail fast with RuntimeError.

✨ Highlights

• Familiar Actor model API: start / proxy / ask / tell / stop
• Asyncio-first scheduling with no actor thread pool dependency
• Built-in async proxy and registry utilities for large actor sets
• Deterministic loop-bound safety model for production debugging

📥 Installation

Current latest tag: v0.1.4

From Git tag (recommended now)

pip install "git+https://github.com/yzbf-lin/async-pykka.git@v0.1.4"

From source archive (no git required)

pip install "https://github.com/yzbf-lin/async-pykka/archive/refs/tags/v0.1.4.tar.gz"

From release wheel (after release assets uploaded)

pip install "https://github.com/yzbf-lin/async-pykka/releases/download/v0.1.4/async_pykka-0.1.4-py3-none-any.whl"

PyPI publishing is planned. Until then, use GitHub tag/release installation methods. Import package name: async_pykka.

🚀 Quick Start

uv venv
uv sync --group dev
uv run python examples/quickstart_counter.py
uv run pytest -q

Expected output: counter=5

🧩 Minimal Example

import asyncio
import async_pykka


class GreeterActor(async_pykka.AsyncioActor):
    def __init__(self, name: str):
        super().__init__()
        self.name = name

    async def greet(self) -> str:
        return f"Hello, {self.name}!"


async def main() -> None:
    ref = GreeterActor.start("World")
    proxy = ref.proxy()

    message = await proxy.greet()
    print(message)

    await ref.stop()


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

📦 Core API

API	Description
AsyncioActor	Actor base class
ActorRef.tell(msg)	Fire-and-forget message send
ActorRef.ask(msg)	Request-response Future
ActorRef.proxy()	Async proxy facade
ActorProxy.set(name, value)	Safe actor state mutation
ActorRegistry.stop_all()	Graceful batch shutdown
Future.get(timeout=...)	Timeout-aware await

🧠 Common Patterns

Request/Response with timeout

future = ref.ask({"type": "query", "key": "profile"})
result = await future.get(timeout=1.0)

Notify queue + batch handling

Put high-frequency notify events into a queue, then consume in batches.

Runnable example: examples/notify_batch.py

Graceful shutdown

results = await async_pykka.ActorRegistry.stop_all(current_loop_only=True)
assert all(results)

⚡ Network I/O A/B Benchmark

Run A/B benchmark against upstream pykka:

uv sync --group dev --group bench
./scripts/fetch_pykka_repo.sh
uv run python examples/benchmark_network_ab.py --actors 50 --requests 5000 --concurrency 200 --rounds 3

Benchmark setup (2026-03-06): actors=50, requests=5000, concurrency=200, payload_bytes=256, rounds=3, localhost TCP echo server.

Metric (avg over 3 rounds)	async-pykka	pykka	Delta
Throughput (req/s)	29999.08	14514.66	+106.68%
Mean latency (ms)	5.3951	11.9221	54.75% lower
p95 latency (ms)	6.6906	13.9568	52.06% lower
p99 latency (ms)	11.8346	20.5744	42.48% lower

📚 Documentation

• Quickstart
• Scenarios
• Performance Guide
• FAQ
• Glossary

🛠 Development

uv sync --group dev
uv run ruff check .
uv run pytest -q

🚢 Publishing (Maintainers)

Option A: GitHub Actions (recommended)

This repository includes .github/workflows/publish-pypi.yml, which publishes on v* tags.

One-time setup on PyPI (Trusted Publishing):

1. In PyPI project settings, add a Trusted Publisher with:
   • Owner: yzbf-lin
   • Repository: async-pykka
   • Workflow: publish-pypi.yml
   • Environment: pypi
2. In GitHub repo settings, create an environment named pypi.

Release command:

git tag vX.Y.Z
git push origin vX.Y.Z

Option B: Local token publish

export PYPI_TOKEN='pypi-xxxxx'
./scripts/publish_pypi.sh

🙏 Attribution

This project is inspired by async proposals around Pykka and further adapted into a standalone asyncio-first framework.

• https://github.com/jodal/pykka/pull/218
• https://github.com/x0ul/pykka.git

See: ACKNOWLEDGEMENTS.md, THIRD_PARTY_NOTICES.md

📄 License

MIT. See LICENSE.