AI that never sees your client's matter.

Run open-weight models in a workspace your firm governs: zero-retention inference that never stores or trains on your prompts, and a complete audit trail of who asked what, when. Enterprise: deployed inside your firm's own cloud tenant, where matter data never crosses your perimeter.

Why it matters

Why law firms choose PrivateStack

The risk

Pasting client facts into a public AI tool is a disclosure decision. Under your duty of confidentiality, that's a decision you're making for your client, every time, often without telling them.

With PrivateStack

PrivateStack gives every prompt and draft a governed, logged home: inference is zero-retention and never used for training, with a disclosed subprocessor and contractual terms. On an Enterprise deployment inside your firm's own cloud tenant, nothing crosses your perimeter at all.

The risk

Courts are increasingly issuing standing orders on AI use in filings, and opposing counsel is learning to ask what you fed the model.

With PrivateStack

Full request logging means you can certify exactly how AI was used in a matter: who queried it, what was asked, and what came back. An answer you control, from logs you own.

The risk

A closed AI vendor's terms of service can change, and their retention practices are theirs, not yours.

With PrivateStack

Your retention terms are contractual, not a vendor's changeable policy. On an Enterprise deployment, open-weight models run on infrastructure you own, so if we disappeared tomorrow your stack keeps running and your data never moved.

Solo / Entrepreneur is $49/month for one seat, by invitation. Team is $99/user/month for up to 9 seats, sales-assisted. Flat seat pricing avoids a metered bill that scales with every associate's usage.

At a glance

PrivateStack is a governed AI workspace for law firms: AI that never trains on privileged communications.
Models: open-weight families including Llama, Mistral, DeepSeek, GLM, Qwen, and Gemma.
Every request is logged and exportable: who asked what, when, and what came back. The firm can answer “how was AI used in this matter” from records it owns.
Hosted-tier inference is processed by a disclosed US-based subprocessor under contractual zero-retention terms: prompts are never stored and never used for training.
Enterprise BYOC deploys inside your firm's own cloud tenant. On a BYOC deployment, matter data never crosses your perimeter.
Pricing: Solo / Entrepreneur is $49/month for one seat, by invitation. Team is $99/user/month for up to 9 seats, sales-assisted.

Plan the evaluation

Bring the workflow, the owner, and the evidence.

Synthetic example — not a customer story or measured result.

Compare two invented contract clauses with no client or matter information. Ask a lawyer to review both the draft and the proposed handling of future client inputs.

Who decides: Responsible lawyer, firm risk partner, and IT/security lead
Data risk to resolve: Confidentiality, privilege, provider terms, and retention require separate judgments; a synthetic exercise proves none of them for a real matter.
Deployment question: Who can access prompts, outputs, and logs in the proposed configuration, and what must the firm authorize before client information is used?
Evidence to request before approval: Dated provider terms and a processor/data-flow map
A redacted access and retention configuration
The firm's written decision on permitted inputs and lawyer review

Use this checklist to prepare your conversation. It is not evidence that a deployment passed review. Confirm current controls against the Trust Center and the configuration being proposed.

Review the law-firm confidentiality questions

Ready to see PrivateStack in your environment?

Open models running in your own cloud, with the logs and controls to prove it. We'll walk your team through the exact architecture.