Terms of Service

agentwatch is an independent observer for AI coding agents. It captures an agent’s prompt, actions, and final summary, then produces a deterministic verdict comparing what the agent claimed against what the captured record shows. The verdict is informational. It is not a guarantee about the correctness, security, or completeness of any code, and it does not replace your own review and testing.

You are responsible for safeguarding your account and your API keys, and for activity that occurs under them. Keys are org-scoped and individually revocable; rotate any key that is exposed. You retain ownership of the data you submit. You grant us the limited license needed to host, process, and display it to you in order to provide the service. How we collect, retain, and delete that data is described in our Privacy Policy, which is part of these terms.

You agree not to:

• use the service to capture data from people or systems you are not authorized to monitor;
• attempt to breach, overload, or circumvent the ingest API, rate limits, or access controls;
• reverse engineer the hosted platform (the open-source client excepted, per its license);
• use the service unlawfully or to infringe the rights of others.

You are responsible for getting any consent required to monitor your team’s agent activity before you install repo-level capture.

The capture client is published under its open-source license in the public repository and is governed by that license, not by these terms. These terms cover your use of the hosted service (dashboard + ingest API). You can run the client in --local mode with zero egress and no account; in that mode no data is sent to us.

Paid plans are billed per the price and run limits shown on our pricing page. Free tiers have hard caps and no overage. Enterprise terms are set by separate written agreement, which prevails over these terms where they conflict. You can stop capture at any time with agentwatch logout or by revoking your key.

The service is provided “as is,” without warranties of any kind, to the maximum extent permitted by law. A verdict is an automated, best-effort assessment of an agent’s self-report against a captured record; it can be wrong in either direction and must not be the sole basis for shipping, blocking, or any high-stakes decision. To the fullest extent permitted by law, agentwatch is not liable for indirect, incidental, or consequential damages, and our aggregate liability is limited to the amount you paid us in the twelve months preceding the claim.

We may update these terms; material changes will be reflected by the “last updated” date above and, where appropriate, an in-product notice. Continued use after a change means you accept the updated terms. Questions: security@agentwatch.run or sales@agentwatch.run.