Which legal AI vendors offer on premise or private cloud deployment?

11 of the 130 legal AI vendors in the AI Legal Index publish their deployment options and data residency to a standard an outsider can check: the regions available, what changes between tiers, and where processing happens as distinct from where data is stored. They are Ankar, DeepIP, DeepJudge, Draftwise, LEGALFLY, Lexroom, Noxtua, Reveal, Solve Intelligence, SpotDraft and Vesence. A further 36 state the deployment model clearly with partial residency detail, or offer residency without addressing where processing occurs. 44 imply cloud delivery without stating the tenancy model or the region, and 39 publish nothing about where the software runs or where client data sits. This axis grades what is published rather than what is sold, so a vendor may offer single tenant or on premise deployment and grade lower here because none of it is stated in public.

Firms ask this question to establish control over client data, which makes the useful answer larger than a list of tenancy options. The AI Legal Index grades Deployment Model and Data Residency on whether an outsider can establish where the software runs and where the data sits, and then puts that next to the two records that decide whether the control is real: who else touches the data on the way through, and what may be done with it once it arrives.

Residency control is not a training commitment, and the top deployment tier proves it. Of the 11 vendors publishing deployment and residency at the highest standard, 3 prohibit training on customer content in the published agreement, 3 state it in policy only, 4 publish no located term or policy addressing training either way, and 1 reserve the right to train. Choosing where data sits and establishing what may be done with it are two separate diligence steps, and the second does not follow from the first.

DeepIP, LEGALFLY, Reveal and SpotDraft publish deployment and residency at the top of the axis and have no located term or policy addressing training either way. That combination is worth naming precisely because it looks like the opposite from the outside: a vendor that has clearly thought about where data lives reads as a vendor that has settled what happens to it, and the record does not support that inference. The full training picture across the index is on do legal AI vendors train their models on client data.

Whether a vendor will build you a private deployment. This axis measures published material, so a vendor that routinely delivers single tenant or on premise installations under a negotiated agreement and has never said so in public grades on the silence. That is a finding about disclosure and not about capability, and it is dated: a vendor that publishes a deployment page tomorrow is redated the day it does.

The index also cannot audit where processing actually happens. It records what the vendor states and whether the statement distinguishes storage from processing, which is the distinction most residency claims blur. A firm with a genuine sovereignty requirement should treat the published position as the start of the diligence rather than the end of it, and should ask the supply chain question separately, because that is the one the deployment page is least likely to answer.