aevum-core

aevum-core

The independent black box for AI agents — tamper-evident, independently verifiable records of what your agents did.

aevum-core is the Aevum context kernel: five governed functions (ingest, query, review, commit, replay) over an append-only episodic ledger, gated by five unconditional barriers (crisis detection, classification ceiling, consent, audit immutability, provenance) that are hardcoded and never policy-controlled.

• SHA3-256 hash-chained sigchain — every event is RFC 8785 (JCS) canonicalized and Ed25519-signed (RFC 8032); an optional hybrid ML-DSA-65 (FIPS 204) signature runs alongside it for post-quantum defense-in-depth.
• COSE_Sign1 receipts with RFC 3161 trusted timestamps — each entry is wrapped in a portable, independently verifiable receipt.
• RFC 6962-style Merkle log — Signed Tree Heads with inclusion and consistency proofs detect any historical rewrite of the ledger.
• Cedar ABAC + OPA policy engines — consent and content/infrastructure decisions are externalised and optional, sitting behind the unconditional barriers rather than replacing them.
• Verification logic also ships as a separate package, aevum-verify, so a receipt can be checked without running the rest of the kernel.

Install

pip install aevum-core
pip install "aevum-core[cedar]"  # + real Cedar consent enforcement

Docs

Full documentation, threat model, and regulatory alignment: https://github.com/aevum-labs/aevum

License

Apache-2.0.