Durable workflow engine.
The Greek conception of eternal, unbounded time — distinct from Chronos, the ticking clock. A workflow that sleeps for three months and resumes is living in Aion time.
What is Aion
A system that runs long tasks reliably. If something crashes or the server restarts, the task picks up exactly where it left off — no lost work, no starting over. Think of it as a really reliable to-do list for computers.
What makes it different
BEAM-native fault tolerance
Temporal rebuilt process management, supervision, and fault isolation from scratch in Go. Aion runs on beamr — every workflow is a lightweight BEAM process with native mailboxes, supervision trees, and crash isolation. The durability layer is what Aion adds.
Type-safe workflow authoring
Workflows are ordinary Gleam modules. Activity inputs, results, signals, and queries are statically typed — wire mismatched types and the compiler stops you. Deterministic time and random sources are provided by the SDK; divergence is caught during replay.
No default timeouts
Agentic activities that run for hours are first-class. The engine imposes zero time bounds. The author chooses deadlines; the engine provides the mechanism. Workflows that sleep for months resume exactly where they stopped.
Content-hash versioned deploys
Deploy a new version at runtime; running workflows stay pinned to theirs. Hot-reload during development — edit, save, and the new version loads without restart. No coordination, no downtime.
The north star
AI agents as infrastructure — not a tool you build with, but a load-bearing primitive you can make guarantees about. Deploy one binary that survives crashes, shows its work, and runs forever. That's the pitch.
Why not just use Temporal?
Temporal proved that durable execution works. But it was built in Go, which meant rebuilding process management, supervision trees, and fault tolerance from scratch — things the BEAM has provided for thirty years. Aion starts where Temporal starts over: with a runtime that already knows how to supervise, isolate, and recover processes. The durability layer — event-sourced histories, deterministic replay, durable timers — is what Aion adds.
Kill the server mid-run. On restart, Aion replays the event history and the workflow resumes at the exact same await — without re-executing completed activities. Sleep for three months, resume exactly where you stopped.
How it compares
Runtime
Go server, goroutine-per-workflow. Fault tolerance rebuilt from scratch.
BEAM processes on beamr. Supervision, isolation, and crash recovery from the runtime.
Authoring
Go, Java, TypeScript, Python, .NET. Untyped signal and query payloads.
Gleam. Statically typed activities, signals, and queries — mismatches caught at compile time.
Versioning
Task-queue-based. Author manages version routing and patching.
Content-hash immutable packages. Running workflows pinned to their version automatically.
Timeouts
Schedules-to-close, start-to-close, and heartbeat timeouts required on activities.
No default timeouts. The author sets deadlines; the engine provides the mechanism.
AGPL-3.0 licensed. View on GitHub →