// INDUSTRIES · CUSTOMER SERVICE AI
In 2024 a court held Air Canada to a refund policy its chatbot invented. The airline argued the bot was a separate entity and lost. Every insurer, hospital, bank and airline putting AI in front of customers now carries that question: what exactly did the bot say, does it bind, and can either side prove the transcript was not edited afterwards. Fidacy answers it the way it answers payments, with proof instead of trust.
// HOW IT WORKS
One line of @fidacy/session in your chat loop hashes every message into a tamper-evident chain. The transcript never leaves your infrastructure, so a hospital or an insurer keeps its PII and Fidacy only ever sees a hash. The session closes and the digest is anchored to Bitcoin, third-party verifiable, that not even you can rewrite after the fact.
The conversation ends and the customer gets a verify link, no account, no key. They confirm what was said, byte for byte, against a neutral third party that holds no custody. Proof only settles a dispute when the other side can check it too, and a company cannot issue trustworthy proof of itself. Every receipt carries a quiet line of distribution: verified by Fidacy.
We do not judge what the bot said. Moderation is a detection arms race and we chose not to be detection. We gate commitments. A refund, an approved claim, a quote above a threshold, a document sent, each is an action that needs a signed grant. A promise made without a grant does not bind, and your terms of service say so, pointing at the open spec. The bot can talk freely. It cannot obligate you outside what you authorized.
// WHO SHIPS THIS
Selling a hospital or an insurer direct is a twelve-month procurement. The fast path is the company that already ships the chatbot. One integration covers a thousand end customers, and audit-proof AI support becomes their sales differentiator. Same distribution playbook we run with MCP, ClawHub and CrabTrap.
One integration covers every customer you serve. Auditable, dispute-proof AI support becomes the reason an enterprise picks you over the vendor that cannot prove what its bot said.
An AI agent that discusses coverage carries the same question a claims adjuster does: what was represented, and does it bind. Anchor the conversation, gate the commitment, keep selling faster than the incumbent.
Patient conversations are heavy PII. Fidacy never sees the transcript, only the hash. You keep the record inside your walls and still hold a proof any auditor can verify without trusting you.
See it, don't take our word
Send an agent action. Get back approve / review / deny and a cryptographically signed verdict you can check against our public JWKS. No trust in us required.
>=-/\ SHIP /\-=<
The market says this layer has to exist. It's running, self-serve: spin up your org, connect an agent in one MCP install, and verify the verdict yourself. No sales call.
Live on UCP · AP2 · A2A. Built for the audit-trail and accountability duties regulators are writing worldwide, EU AI Act, and the wave following it.
