A clear route to an answer.

Cairn sits between an AI agent and its connected tools. It evaluates tool calls against policy, allows or denies them, or holds them for a human decision. It records those decisions in a hash-chained ledger.

Cairn has a working prototype and synthetic evaluation evidence. We are preparing the managed product for pilots. The first step is to agree a focused workflow, the tools it can use, the people who approve actions and the evidence you need. Discuss a pilot.

Allow lets a request continue within policy. Hold pauses it for a human decision. Deny blocks it. An approval is part of the record, so the decision can be traced afterwards.

The managed product is being built with customer owners and administrators who manage their own team. Operators, approvers and auditors will have distinct permissions. The roles and access boundaries must be verified before your pilot starts.

No. The examples on this website are illustrative and use synthetic agents, organisations and transactions. Clicking a demonstration control does not move money or operate a customer's systems.

Each ledger entry includes a hash linked to the entry before it. Verification detects inconsistent changes to the retained chain. Detecting a complete rewrite also requires a trusted checkpoint held separately. The assurance page explains the limits.

The prototype ledger uses JSON Lines and includes an Apache-licensed verifier. Pilot requirements include evidence export and recovery, so a product interface is not the only way to read your record. Access and retention are agreed as part of setup.

Include the page or workflow, time of the issue, browser or integration version, steps to reproduce, expected result and actual result. A redacted screenshot or error reference helps. Keep API keys, passwords and customer records out of an initial email. Prepare a bug report.

Contact the team privately with a brief description and a way to reach you. We will arrange an appropriate channel for sensitive technical details.