Open Source · BSL 1.1 · AI Audit & Governance
Provenant
AI agents now take real actions — run commands, move money, change records. When a regulator, an auditor, or a customer asks "what did your agent actually do, and can you prove you didn't edit the logs afterward?", most teams have nothing but a mutable text file. Provenant is the answer: tamper-evident, examiner-ready records of every agent action, verifiable by a third party without trusting you — or us.
Tamper-Evident by Construction
Every Record Is Cryptographically Chained to the Last
Each agent action is written as a record whose fingerprint (an HMAC hash) includes the fingerprint of the record before it. That links the whole history into a chain. Change a single character in any past record — a timestamp, an amount, an outcome — and its fingerprint changes, which breaks every fingerprint downstream. The tampering isn't hidden; it's mathematically obvious. There is no "quietly edit the log" path.
Verifiable Without Trusting the Vendor
Anchored to Public Time. Provable by Anyone.
Tamper-evidence proves the log wasn't edited — but not when it existed. Provenant closes that gap by anchoring the chain to public, independent time using OpenTimestamps, which commits to the Bitcoin blockchain. The result: an examiner can confirm your records existed at a given moment and haven't changed since, using nothing but public data — no access to your systems, no need to take your word for it. This is the difference between "a log we keep for you" and "a receipt anyone can verify."
Speaks the Regulators' Language
Mapped to the Frameworks Auditors Actually Cite
Provenant doesn't claim to make you "compliant" — no tool can. It produces the specific, defensible evidence that record-keeping and traceability controls demand across the AI-governance frameworks now landing in law and procurement.
EU AI Act — Article 12
Automatic record-keeping ("logging") over the lifecycle of a high-risk AI system. Provenant is that log, made tamper-evident.
ISO/IEC 42001
AI management system controls for operational records and traceability of AI decisions and actions.
SOC 2
Evidence for monitoring and change-management criteria (CC7 / CC8): a verifiable trail of what ran and when.
NIST AI RMF
Supports the Measure and Manage functions with traceable, auditable records of agent behavior.
What's in the Box
Four Small CLIs. Zero Dependencies. Runs Air-Gapped.
attest
Records an agent action into the tamper-evident chain, automatically.
audit
Renders a self-contained HTML report an examiner can open and verify.
policy
Checks actions against declared rules before they're allowed to proceed.
anchor
Commits the chain head to public time via OpenTimestamps → Bitcoin.
Stdlib-only, so it runs anywhere Python runs — including fully offline. Capture is automatic (git plus assistant hooks), the examiner report is a single file with nothing to install, and the whole thing is sovereign: no SaaS, no vendor in the trust path.
Prove what your agents did.
Tamper-evident. Examiner-ready. Third-party verifiable. Air-gappable.
View Provenant on GitHub →