Ship AI you can prove.
Pre-alpha · Check the Status & Acceptance tracker for what works today.
FleetForge is a control plane that makes AI workflows provably safe: every step has a signed receipt, policies enforce safety and cost, and any run can be replayed exactly.
Keep LangGraph, AutoGen, CrewAI, Airflow, Argo, and creative pipelines intact while the mesh supplies receipts your risk teams can trust.
Why teams adopt FleetForge
- Provenance built-in – Every step emits signed attestations (SLSA/in-toto predicates, C2PA manifests) stored in the trust graph and, optionally, posted to SCITT.
- Deterministic delivery – Seeds, checkpoints, and backpressure honour SLOs so replays mirror production behaviour.
- Policy you can hire for – Rego→Wasm ships today; OPA bundle + Cedar compatibility, policy SDKs, and provenance for imported packs stay on the roadmap.
- Portable capabilities – Scoped capability tokens secure each hop now, with Biscuit v2 exports and W3C VCs next for offline verification.
- Telemetry that will not break dashboards – Versioned
trust.*attributes align with OpenTelemetry GenAI conventions and the documented stability flags keep collectors in sync.
Start here
- Hello Fleet demo – run the canonical demo and collect receipts end-to-end.
- Concepts – dive into Delivery, Policy, Replay, and ChangeOps pillars.
- Reference hub – CLI, SDK, API, schemas, telemetry, and attestation vault specs.
Need to understand what’s shipping next? The Roadmap & Status page consolidates M1/M2 work with the longer-term policy, capability, replay, and transparency tracks.