Why we built this
In February 2024, a tribunal held Air Canada responsible for inaccurate information its own support chatbot gave a customer (Moffatt v. Air Canada, 2024 BCCRT 149). The argument that the chatbot was a separate entity was rejected.
Every company that deploys a customer-facing AI agent is now exposed to the same reasoning. Yet most have no systematic way to check whether their agent stays within their contractual commitments — SLAs, warranties, pricing, terms — or to prove it did at a given date.
ClaimSentry is our answer: extract your commitments, test your agent against each one, countersign every verdict, and seal it into an Evidence Pack you can verify independently.
Made in France, and why it matters
We build in France because data sovereignty is a value, not a feature. Your commitments and your agents' responses are sensitive. They belong to you and should remain in infrastructure you control.
ClaimSentry is available fully on-premise. Each customer runs in an isolated instance with a dedicated database. Secrets are encrypted at rest, and the LLM judge can run entirely inside your perimeter.
Built on open standards
We publish your verified commitments through the Model Context Protocol so any MCP-capable agent you operate can ground its answers in your official facts. No proprietary lock-in.
The judge LLM is configurable: bring your own OpenAI-compatible key, or run a model inside your own network. Your data never has to touch a shared LLM budget.
Evidence by design
Knowing what an AI agent said is one thing; proving it later is another. Each ClaimSentry verdict is hashed (SHA-256) and carries an RFC 3161 electronic timestamp, delivered as a .tsr token you can verify yourself with standard tools.
Under Article 41 of the eIDAS Regulation (EU 910/2014), an electronic timestamp is admissible as evidence in legal proceedings; with a qualified timestamping authority, it enjoys a presumption of the accuracy of its date and the integrity of the data. ClaimSentry supports qualified TSAs. It builds the diligence file that supports your obligations — it is not a certification.
Accuracy first
Every verdict is grounded in your reference documents and countersigned by a human, with the exact source excerpt that defines correctness.
Sovereign by design
Isolated instances, on-premise deployment, internal judge. Your data stays where you decide it stays.
Provable, not just plausible
Every verdict is RFC 3161-timestamped and independently verifiable. Your compliance team gets a diligence file — not just a report.
Get in touch
For sales, partnerships, or press inquiries: contact@claimsentry.io