For anyone working with an AI agent

Keep your AI honest and on task.

It reads what your agents are actually doing while they work, tells you the moment one goes wrong — quoting the line it is objecting to — and stops it before the job finishes wrong. The checks are plain code running outside the model, so the agent can't talk its way past them.

Runs on your machineNothing uploadedClaude Code, Cursor, CopilotMac, Windows, Linux

Stops

  • Loops
  • Stalls
  • Acting on unverified assumptions
  • Context loss after compaction
  • The agent swapping mid‑run
  • Prompt injection
  • Runaway cost and runtime
  • Capability it was never granted
  • Risky actions without approval

Proves

  • Hash‑chained audit trail
  • Tamper‑evident by design
  • Independent verification
  • The agent cannot pass its own work
  • On doubt it blocks, never guesses
  • Your own append‑only log
The agent cannot change its own objective. It cannot expand its own permissions. It cannot certify its own success.
1 · InstallOne command. Thirty seconds.
2 · Restart your toolYour state comes back with it.
3 · Carry onIt speaks up only when it catches something.

The model reasons.
The model never decides.