Provared

Provared is a free, open format for proving what an AI agent was allowed to do and what it actually did. It works in four parts: a permission (Slip) that a person signs with a passkey specifying which actions the agent can take and what limits apply, receipts (Stubs) that the agent signs for each action, a sealed book (Seal) that chains and timestamps the receipts, and a show (Show) that proves an entry is in the book.

The system includes a free checker that verifies whether the agent stayed within its permission and identifies any breaches. It answers two questions separately: whether the record is intact, and whether the agent stayed within its limits. Going over a limit is not treated as a fault in the record itself, but rather as what a sound record should reveal.

Provared is built from published standards only, using JSON Web Signature, Ed25519 and ML-DSA-87 quantum-safe signatures, RFC 3161 time-stamps, and RFC 9162 trees. It is available as libraries for JavaScript (npm) and Python (pip), with integrations for LangChain and Pydantic AI agents.