NovaFabric: replayable, verifiable execution evidence for AI agents

NovaFabric is self-hosted, open-source replay and evidence infrastructure for AI agents and agentic systems. It captures agent executions as portable, secret-scanned Run Capsules and provides replay, behavioral and structural diff, lineage, cryptographic provenance, assurance, and audit around those executions. The same capture layer can wrap general commands and runs entirely in the user's own infrastructure, from local environments through Kubernetes to SLURM/HPC, with no accounts and no telemetry.

It is built around five primitives — Asset Registry, Run Capsule, Replay, Lineage, and Evidence Bundle — and five explicit replay modes (forensic, mocked, semantic, exact, and the experimental intervention mode). Run Capsules can be sealed with the user’s own key; sealing is opt-in. Evidence bundles verify offline using only sha256sum and an ed25519 verifier, with no NovaFabric installation required.

Interoperates with established specifications rather than inventing formats: OpenTelemetry GenAI semantic conventions, in-toto/DSSE, RFC 3161 timestamping, W3C PROV, OpenLineage, and CycloneDX.