I built a library called Duron that provides the core building blocks for a durable execution engine, built on deterministic log replay and a custom asyncio loop.
To better support interactive workflows—such as human-in-the-loop interactions, streaming progress updates, and real-time inputs—I'm experimenting with two new primitives: signals and streams.
Key features:
- Asyncio-native: Custom asyncio loop that works seamlessly with standard primitives like `asyncio.timeout`, `asyncio.gather`, and `asyncio.sleep`
- Minimal dependencies: Zero external dependencies, pluggable log architecture for easy integration
- Interactive workflows: Streams and signals enable real-time communication between workflows and host applications
- Stateful effects: Generator-based effects with automatic checkpointing and state accumulation
Example use cases:
- Building custom durable workflow engines with minimal overhead
- Interactive workflows requiring human approval or external triggers
- AI agents with long-running conversations that persist across restarts
GitHub: https://github.com/brian14708/duron
I’m still iterating on the API design and stress-testing durability guarantees. Feedback, critiques, and ideas for real-world use cases are all welcome!