DIN for enterprise
Governance you can read. DIN records what your AI agents actually did at the tool-call boundary, grades every claim by how it was obtained and whether it held, and keeps that record on infrastructure you control.
It runs on your infrastructure
- Self-host is the default, and hosting is never forced. Every tier listed hosted is also sold self-host. Hosted is set up with you: an isolated tenant on our hosted service with a store cap of 100 GB per org.
- Self-hosted. The installer starts one server on loopback behind a token it mints on your
own machine. A paid licence is verified when you run
din activate: one call to our licensing service, or none on a machine with no network:din activate --filewith a saved reply, which lasts 14 days too, so an offline install needs a fresh reply carried to it every 14 days. The programs then check its signature on your machine. A paid install checks in withapi.dinctrl.com, with a random install id, when you rundin activate --renewand at most once a day when the field or the console starts (or withdin autostart); each signed check-in lasts 14 days. A free install never checks in. A key stays valid for up to 72 hours past each billing date. - Your inference is yours. Grading runs on your assistant CLI, your own API key or a local model, never on our account. Nothing about your work is shipped to us to be graded, and there is no metered inference line anywhere on the bill.
- Your records are never deleted on a self-hosted install, on a downgrade or a lapse.
The audit trail is the product
DIN grades; it does not just search. Anything can find the conversation where a decision was made. What this does is tell you whether the claim in it was ever borne out.
- How a claim was obtained is one mark: receipted, logged, inferred. What it showed is another: proved, observed, asserted, refuted. The two are never collapsed into one badge, because that is exactly how an assertion ends up looking like a measurement.
- Model provenance on every record, and the drift view over it, so "which model said this, and when" is a question with an answer rather than an investigation.
- Nothing is overwritten and nothing is deleted. A superseded capture is kept as the audit trail, and it is not counted twice anywhere else.
- The arithmetic behind a number is printed on the screen the number is on, so a figure can be audited rather than taken.
The access model, one line each
Enterprise is a conversation, not a checkout
The field maths continues past Agency and the shape does not change: Enterprise starts at 180 fields, with more client orgs, self-hosted or hosted. It is not on a price list and there is no checkout for it: the company that pays for the AI seats talks to a person, who answers with a figure for what you need.
Enterprise starts where Agency stops (90 fields, 5 orgs, $399 a month self-hosted). A message takes a name and an address, and a person answers anything this page does not.