Security, compliance and data residency
Most of what a security team needs from us is on this page rather than behind a form, because gating it only delays the moment you find out whether we clear your bar. Where we do not meet a standard, that is stated here too — there are three of those and they are in the first section.
Start here
Putting this first is deliberate. If one of these is a hard requirement, you can stop reading and save the meeting.
Data residency
Densery runs on a GPU appliance we supply and install inside your data centre. No document content, no extracted values and no metadata cross your network boundary at any point. Model updates arrive as signed artefacts you approve and apply. This is what our banking customer runs, and it is the configuration people call us about.
Deployed inside your own VPC in your chosen region, with customer-managed encryption keys. Content stays within your cloud account and your key policy. Appropriate where policy permits cloud but not shared infrastructure.
The fastest path to a first production process, in your chosen region. Appropriate where there is no residency constraint — though if that is your situation, be honest with yourself about whether you need us rather than a hyperscaler API.
On-premise capability is no longer scarce — open-weight document models run on a single consumer GPU. What is scarce is on-premise with the governance stack intact, and it only matters where the constraint is legal rather than a preference.
Controls
| Area | Status | Detail |
|---|---|---|
| Data in transit | In place | TLS 1.3 throughout. In on-premise deployments all traffic is internal to your network. |
| Data at rest | In place | AES-256. Customer-managed keys in private-cloud deployments; your own storage and key management on-premise. |
| Model training on your data | Off by default | Nothing is used to improve anything outside your instance without a written data-rights agreement and a commercial term attached to it. Corrections improve your ontology, in your instance. |
| Access control | In place | SSO via SAML or OIDC, SCIM provisioning, role-based access, and agent actions executed under named service identities rather than shared credentials. |
| Audit logging | In place | Per-decision provenance with model and ontology versions pinned, human overrides attributed, and threshold changes recorded with the approver. Exportable as a signed record. |
| Penetration testing | Annual | Third-party test annually and on material architecture change. Summary letter available under NDA. |
| SOC 2 Type II | Confirm status | Third-party attestation status is being confirmed and is not published here yet. We will not describe this as complete until the report is issued. |
| ISO 27001 | Confirm status | Certification status is being confirmed and is not published here yet. |
| FedRAMP / StateRAMP | Confirm status | Authorisation status is being confirmed and is not published here yet. See the first section of this page. |
| Sub-processors | Published | Listed on request and notified before any change. In on-premise deployments there are none in the data path. |
Unverified — do not rely on this table. The control rows describe our intended posture and the certification rows are pending confirmation. Every row must be checked against the current control set before this page is published: a status that is wrong here is worse than one that is missing.
Security questionnaires
CAIQ, SIG Lite, your own bank-standard workbook, or a vendor risk platform — send it to hello@densery.com and it comes back completed within five business days. No gating call, no requirement to be in an active sales process.
We would rather be ruled out in week one on a questionnaire than in month four after both sides have spent real money. If a question shows we do not meet a control, the answer will say so rather than describing a compensating control at length.
The next step
A scoping session is not a demo. Bring twenty real files, redacted if you need to. We take three numbers off you — annual volume, fully loaded cost per file today, and what happens when the output is wrong — and hand back a one-page value case in your own KPIs.
If the arithmetic says we are not a fit, we will tell you in the room rather than six weeks later.
If your data can go anywhere and your documents are already clean and digital, you do not need us. Use a hyperscaler document API and spend the money on something harder.
Densery