Durable workflows that feel like writing normal code.
Sayiir is a fast, lightweight durable workflow engine built in Rust with Python bindings. It checkpoints after each task and resumes from the last checkpoint — no deterministic replay, no DSLs, no separate infrastructure. Just a library you import.
| Repository | Description |
|---|---|
| sayiir | Core engine (Rust) + Python bindings — durable execution, fork/join, retries, distributed workers |
- No deterministic replay — checkpoint & resume, not re-execution. Use any API, any library, zero determinism constraints
- A library, not a platform — add it as a dependency and run. No cluster, no scheduler, no separate workers
- Rust core, thin bindings — every language gets the same performance and safety guarantees
- Pluggable everything — swap storage backends, codecs, and runners without touching workflow code
- Documentation — full docs site
- Quick Start: Python
- Quick Start: Rust
- Architecture
- Comparisons — vs Temporal, Airflow, Prefect, Step Functions, Elsa
- Discord — questions, feedback, discussion
- Contributing — how to set up, build, test, and submit PRs
MIT Licensed — no lock-in, no vendor dependencies.