Security
Sub-processor List
Which third parties can process customer data on our behalf. For the Vault AI appliance, the short answer is none.
No sub-processor receives client content from the appliance. We notify customers before adding any sub-processor that could change that, so you have time to object under the DPA.
Model inference
- None. Inference runs on open-weight Llama 3 and Mistral models that ship on the appliance itself.
- No prompt, document or response is transmitted to a model vendor at any point.
- Model weight and legal-database updates are signed packages we publish; the appliance pulls them outbound-only and sends no content back.
Infrastructure
- Marketing website hosting — serves tokengrill.com. No matter data is ever present there.
- Transactional email — device reservation and account correspondence only.
- Everything else — application, database, index and audit log — runs on the appliance you own.
What no sub-processor receives
- Your firm's client list, matter names, documents, prompts or responses — none of it leaves the appliance.
- Any copy of your matter index. TokenGrill holds nothing to hand on.
- Marketing or analytics vendors are not given access to workspace data.
Need this in writing for your review?
We share the underlying documentation with security and risk teams under NDA, and will walk your counsel through anything on this page.
Request documentation