Your firm’s documents stay your firm’s.
The work is sensitive and the buyers are skeptical — as they should be. So here’s the plain version of how your data is handled, with mechanisms, not badges.
Your data is yours
Every firm's files and standards are logically separated from every other firm's at the data layer. No cross-firm access at any tier.
Never used to train AI
We don't use your documents to train, fine-tune, or improve any AI model — ours or anyone else's. That's a commitment, not a setting, and it holds end-to-end.
Encrypted in transit and at rest
Your documents are encrypted on the way in and while stored. We describe what's encrypted and when — no “military-grade” theater.
A complete record
Every value and every flag carries a full audit trail: who decided what, against which source, and when. Defensible by design.
You control retention
You decide what's kept and for how long. When you delete, it's gone — not archived behind your back.
Citations can't be fabricated
Because the cited text is fetched from your document and never written by the model, the citation itself can't be invented. The AI points; it doesn't author the proof.
Every flag links to the exact line — fetched, not written.
The single most important integrity property: the AI emits only a location, and the system pulls the actual text from your document. No paraphrase, no invented citation. Anyone reviewing the work can check it against the source in seconds.
The honest part.
We’re in private beta, and we’d rather be precise than impressive. So, plainly: we don’t hold a SOC 2 report or compliance certifications yet, and we won’t display badges we haven’t earned. We don’t publish an uptime SLA. We don’t claim “end-to-end encryption” as a slogan — if you need exact encryption scope for your review, we’ll state it in writing.
What we will do is sit with your IT and compliance team, answer the hard questions directly, and put the answers in your security review. If we can’t do something you need, you’ll hear it from us first.