Blog/Private AI for e-discovery document review
Legal

Private AI for e-discovery document review

By Samuel Seidel · September 9, 2026 · 7 min read

A dedicated review model can triage a discovery set for responsiveness and likely privilege without sending it to a cloud AI service. That matters because the set gathers internal emails, draft contracts, board minutes and communications with outside counsel in one place. Through a shared AI service, the other side's confidential records and your own privileged material pass through systems neither party agreed to.

What a document set contains

Discovery collections are broad by design. A request for "all communications related to the termination" pulls in years of unrelated email threads, HR files, calendar invites, and whatever else got swept up by the search terms. Some of that material is privileged. Some belongs to a third party who was never a party to the litigation and has no idea their emails are sitting in a review set. A single leaked batch, even briefly, is the kind of incident that ends up in a motion for sanctions.

First-pass review at scale is a natural task for AI: classify documents as responsive or not, flag likely privilege, summarize long threads before an attorney opens them. Through a general-purpose cloud model, the whole set, privileged material included, goes to the AI vendor to be classified.

What changes when the review model is self-hosted

On a dedicated DGX Spark, every document and every classification goes to your own review model on a machine assigned only to you, not to an AI vendor's API. The team still gets triage by responsiveness, flags on documents that mention outside counsel or use legal-advice language, and a short summary of each thread.

A rented Spark is still a GPUwerk-run machine (its own in Czechia, or a partner's in the country the console shows), operated under its published DPA, so whether it fits a protective order or the client's outside-counsel guidelines is a call for counsel. A firm that needs the set to stay in its own building should run the model on its own hardware.

Open WebUI handles the review interface, and to check documents against the matter's search terms or privilege log format, use the retrieval approach in on-premise RAG: load the search terms and privilege criteria as reference material so the flags are grounded in the matter, not a generic sense of what looks sensitive. The same logic applies to other confidential document work; see private AI contract review.

What the model is useful for, and what stays with counsel

A review model is a reasonable first pass for sorting a large set into responsive, non-responsive and needs-a-closer-look, and for flagging likely privileged documents so an attorney sees them before anything is produced. The privilege call, the responsiveness determination on a close document and every privilege log entry stay with counsel. That is the legal standard: a machine classification doesn't replace the professional judgment discovery obligations require.

What this doesn't solve

Self-hosting the review model doesn't replace a review platform's workflow and audit-trail tooling, and it doesn't make a privilege call for you. What it removes is privileged and confidential material passing through an AI vendor's service to get a first pass done. See pricing for what a dedicated Spark costs for a litigation team.

Related pages

First top-up: pay $20, get $30 in credit

Run first-pass review without an AI vendor.

A dedicated Spark for document review, starting at $0.79/hour, deployed in minutes from EU-Central. Read the Open WebUI setup →

Your data stays on a machine assigned only to you. GPUwerk processes it under its published DPA.

Create account See private LLM hosting