FerroAir is an Apache Airflow™ 3.x–compatible orchestrator in Rust. The scheduler core is Rust; Python DAGs run through a PyO3 boundary, and the static fraction of DAG parsing is handled natively in microseconds instead of importing every file through CPython on every poll. Live on AWS Marketplace.

Measured record (as of 2026-07-10)

Quality gates passed

• 100% parser parity on a 64-DAG sample against Airflow’s own parsing (Phase 0 gate, 2026-04).
• The static DAG parser is published as a standalone crate — ferro-airflow-dag-parser — extracted under the same rules as the rest of ferro-protocols, including a panic_safe shim for fuzz-found upstream parser panics.
• Conformance is tracked by running upstream Airflow pytest suites against FerroAir; that surface is being expanded release by release rather than claimed wholesale.

Honest limitations

• v0.1.0 means it: the full Airflow 3.x compat surface (the long tail of operators, AIP-72 task execution modes, provider packages) is explicitly in progress.
• DAGs whose structure depends on runtime state still route through CPython — the native fast path covers the static fraction only.