Website · Documentation · Blog · Slack · YouTube
Flyte 2 is the open-source AI runtime to scale durable workflows to production — on your own infra.
Write plain, async-native Python that orchestrates agents, data and ML pipelines, and inference — with the durability, reproducibility, and observability that production demands.
- 🤖 Dynamic, agent-native execution — Write or generate orchestration logic that adapts on the fly at runtime. Loops, conditionals, fan-outs, and LLM-driven control flow are just Python.
- 🛡️ Durable, self-healing workflows — Flyte 2 is infrastructure-aware, automatically recovering from both logic and infrastructure failures like OOM crashes. Every execution is logged, reproducible, and observable from a single UI.
- ⚡ Realtime and batch inference — Flyte 2 handles inference natively, no separate serving stack required.
| Repository | What it is |
|---|---|
| flyteorg/flyte | The Flyte 2 control plane and backend live on main — scheduling, state, recovery, and the console. |
| flyteorg/flyte-sdk | The Python SDK — pure Python with async/await. pip install flyte |
| flyteorg/flyte-sdk-go | The Go SDK — interface with Flyte from other languages. |
# pip install flyte
import flyte
env = flyte.TaskEnvironment(name="hello")
@env.task
async def hello(name: str) -> str:
return f"Hello, {name}!"
if __name__ == "__main__":
flyte.init_from_config()
run = flyte.run(hello, name="world")
print(run.url)Show us some ❤️ by starring flyteorg/flyte!
