Guardrails for AI Agents

What it does

OpsContext keeps a tamper-evident log of what your AI agents did. Once an hour, while OpsContext runs and only if that log grew, it makes a checkpoint on your machine and sends it to SealHour. SealHour answers with a signed receipt and keeps the checkpoint, so a lost or rewritten disk does not lose the proof.

At minute 2 of the next hour SealHour seals the checkpoints of the hour together, has that seal stamped by the official provider and writes it into Bitcoin. Your checkpoint is included in the stamped hour, verifiable by anyone: the proof comes back to your machine, and anyone can check it without OpsContext and without SealHour.

Who has it

SealHour is included in OpsContext Team and Enterprise, and open to every licence while its pilot lasts. The pilot ends on 1 March 2027 at 00:00 UTC: 28 February is its last day.

After the pilot, Pro reaches SealHour by moving to Team. Without a licence, a pilot code made on your machine (random: it names no one) takes the place of the licence key while the pilot lasts.

What leaves your machine

Once an hour, OpsContext sends SealHour a checkpoint of your audit chain: a few 32-byte fingerprints, the number of records and the time, and, if you choose, one fingerprint for all your repositories with their number; with your licence key. As with any web request, the receiver also sees this machine's address and the time, so it learns the hours you were active. Nothing else leaves: no record, no file, no code, no name. You can stop it any time.

It is off until you turn it on. npx @compr/opscontext-mcp anchor enable shows what would leave your machine and starts nothing without your yes.

SealHour keeps the checkpoints, their receipts and the sealed hours with no time limit, during the pilot and after it: a proof can be needed many years later. SealHour's privacy page says the same.