No. It records, and the answer says so: every response carries observation_only: true, so an agent honouring it keeps working unchanged the day blocking becomes available. What you get today is the register, a refusal on claims that are not usable, and a complete record of what went out.
Only if you connect one, and then only what the connector asks for. A mailbox connector reads sent items only and asks for metadata rather than message bodies — the body is not requested rather than requested and discarded. A CRM connector asks only for the fields you have mapped.
The sentence a promise was made in, capped at 1,000 characters. That is the only customer message text the platform keeps, and the cap is in the record's schema so the claim is checkable rather than a policy. We do not keep the rest of the message.
In transit, everything. At rest, your data sits on encrypted storage provided by our hosting provider — that encryption is theirs, with their keys, and we do not claim it as ours. What we encrypt ourselves, with our own keys, is the credentials for any system you connect, and every backup before it is written. /privacy.md says this in full, including the six things about it that are not settled yet.
On your own deployment — one application and one volume per customer, in London. That is a decision about blast radius rather than cost: a credential belonging to one customer is refused at the door of another.
Yes. It can never accept a claim or approve its own mandate. Lines an agent drafts stay proposed, and a check against one answers not yet confirmed by a person rather than allowed. Paying does not change this, and a paid agent's tool list is identical to an unpaid one's.
It escalates and tells the buyer it is checking. Answering anyway is not a setting — silence and a confident wrong answer are the two failure modes the whole design exists to prevent, so neither is expressible.
They check what an agent says. This checks what it commits your company to, and keeps the record. Most teams that arrive here already run one of those and have just found the gap. /compare.md is the detail, with sources.
£29 a month for a standalone agent, £149 per agent a month for connected agents sharing one register and record, and enterprise is priced by conversation. Cancel any time, no card to start. /pricing.md.
One person, for now, and that is on /founder.md rather than hidden.
Early access, and the honest answer is: parts of it. The register, the record, the checks and the agent connection are built and running. Nothing has been run in anger by a customer yet, there are no case studies, and there is nobody we can point you at — we would rather say that than imply otherwise.