Offline evidence for agent-assisted code

Know what made
the patch.

Turn files, commands, tests, and delegated tasks into one readable review ledger—without recording prompts or sending telemetry.

Open schema Redacted by default Signed when you need it

A surreal paper-cut night landscape where a red thread links document monoliths and brass evidence markers beneath a watchful moon.
A patch is a destination. The ledger preserves the path.

Review field note · 01

The diff tells you what.
The ledger tells you how.

Terminal scrollback disappears. Agent narration is reconstructed after the fact. Another model review adds more guesses. A ledger keeps deterministic evidence close to the files it supports.

A three-step chain

From action stream to review artifact.

The format stays deliberately boring so any agent, hook, shell, or CI job can produce it.

  1. 01

    Record JSONL events

    Write one strict event for a file, command, test, or delegated task. Unknown fields fail loudly.

  2. 02

    Hash and connect

    The CLI hashes current files and artifacts, then links execution evidence to each change.

  3. 03

    Export and verify

    Hand reviewers clean Markdown plus structured JSON. Add an Ed25519 signature when provenance matters.

Local workbench

Try the evidence path.

Paste JSONL below. The preview runs entirely in your browser; nothing is uploaded.

Action events

One JSON object per line. Prompts and file contents are not accepted fields.

Privacy controls

Ready for JSONL events.

Review ledger

No ledger yet

Paste events or load the example. Your data stays in this browser.

Recorded workflow · 6 seconds. The operator loads four example events, builds the ledger locally, then reviews one changed file and three linked evidence entries. There is no audio.

The real instrument

A small binary for the review loop.

Run it beside any coding agent. No daemon, account, database, or runtime telemetry.

shell
$ cargo install agent-audit-ledger

$ aal build --input actions.jsonl \
    --markdown audit.md --json audit.json --root .

 4 files · 7 evidence events · paths redacted

Own the workflow

The ledger is free.
Team memory is one-time.

Core CLI, browser export, redaction, hashing, and signing stay open. The optional team kit remembers and shares consistent policy presets.

Team policy kit

$49 USD · one-time

  • Save named redaction policies locally
  • Export policy JSON for team repositories
  • Reuse review conventions across projects
  • All future v1 policy updates
Buy the team kit

Sociobot/Dodo is merchant of record. Refunds revoke the license. Read our privacy policy and terms.

Have a license? Restore it

Checking local license…

Review with evidence

“The goal is not to prove that an agent meant well. It is to make the work legible enough to judge.”
Build your first ledger