Capture
Native agent streams, prompts where available, tool calls, shell commands, process exits, file changes, and git diffs.
The evidence model focuses on observable agent actions and boundary crossings, then preserves integrity through canonical hashes, signed checkpoints, and verifier-friendly exports.
Native agent streams, prompts where available, tool calls, shell commands, process exits, file changes, and git diffs.
Canonical event encoding, stable payload hashes, hash-chain links, manifest hashes, and local verification metadata.
Remote audit writer checkpoints sequence the received stream and produce server-owned signed evidence state.
Evidence packages can be inspected independently to check hashes, signatures, timestamps, custody, and run boundaries.