Home/Security

Isolated by default. Grounded by design.

Continuity knowledge is some of the most sensitive a firm holds. So every client runs on its own server, the model stays grounded in your data, and nothing you capture is ever used to train anyone’s model.

At a glance:SOC 2 — in progressSingle-tenant isolationNo training on your dataOn-box embeddingsFull audit log
01

The four guarantees.

Isolation

Your own server

One dedicated instance per firm. Your knowledge, embeddings, and audit trail live on hardware that serves only you — no shared data plane.

Grounding

Cite-or-decline

Answers are generated only from your retrieved knowledge, with sources shown. Below the retrieval floor, the Coach declines rather than inventing.

Ownership

No training on you

The frontier model is used under no-training, zero-retention terms. Your captured knowledge is never used to train any model.

Governance

Full audit trail

Every question, answer, drafted action, and approval is written to an exportable log — the record that survives a handover or a diligence request.

Data handling

Where your knowledge lives — and where it never goes.

Retrieval runs on-box: documents are embedded locally on your instance, so the raw text never leaves the server to be indexed elsewhere. Only the minimal, relevant context needed to answer a question is sent to the reasoning model, under no-training terms.

  • ✓ Local, on-box embeddings — knowledge isn’t shipped out to be indexed
  • ✓ Per-client object storage on the instance
  • ✓ Nightly encrypted backups with restore drills
  • ✓ Optional sovereign mode — a fully local model, nothing leaves the box
Request path
1 Question asked on your instance
2 Local retrieval + embeddings stays on box
3 Minimal context → model no-training
4 Cited answer + audit entry logged
02

Security questions, answered.

Is our data shared with other clients?
No. Each firm gets a dedicated single-tenant instance. There is no shared database or shared retrieval index across clients.
Is our knowledge used to train AI models?
No. The reasoning model is accessed under no-training, zero-retention terms, and your captured knowledge is never used for training.
Can the AI make something up?
It’s engineered not to. Answers must clear a retrieval floor and pass citation validation; when they can’t, the Coach declines and the gap is logged for capture.
Who can see what?
Role-based access separates admins, subject-matter experts, and staff. Sensitive items can be restricted, and every access and action is auditable.
What about compliance?
A SOC 2 Type II program is in progress. Single-tenant isolation, audit logging, and data-handling controls are built in today; we’re happy to walk your security team through the details.
For your security team

Bring your questionnaire.

We’ll walk your team through isolation, data handling, and the audit trail.

Talk to us →
Security review welcome