Your work product is privileged. Your AI stack should act like it.

Northwood deploys and operates open-source inference for legal teams: the best-fit model for each task, lower cost per matter, and a complete record of every call, all inside your tenancy.

Learn more→Read the security brief
CostLower, attributable by matter
TenancySingle, in your environment
RecordDefensible, exportable

Your risk committee is right.

Legal teams run on material they are obligated to protect: client files, deposition transcripts, investigation notes, discovery productions, draft pleadings, settlement posture, and attorney work product. Once that context leaves your control, you cannot un-disclose it, and you may not be able to argue it was ever protected.

Northwood deploys and operates an open-source inference layer around that material: the best-fit model for each task, the cost evidence to prove it, and a record of every call, on infrastructure you control.

↗Client confidences have no place in a third-party prompt. Generic tools can’t tell you where a query went, who could read it, or what retained it. Northwood deployments run in your tenancy; outbound routes to provider APIs don’t exist unless you create them.
↗Ethical walls must hold inside the AI too. Permissions flow from your identity provider. A timekeeper screened off a matter cannot retrieve what the screen restricts, enforced by topology, not a policy memo.
↗A fabricated citation is a sanctions problem. Answers are grounded in the documents you supply and cite the controlling source on every claim. Where the record doesn’t support an answer, the system says so instead of inventing one.
↗Opposing counsel will ask how it was produced. Every prompt, response, source document, model, and human approval is logged append-only and exportable, so a review protocol or production methodology can be defended on the record.

Where legal teams put open-source inference to work.

Discovery

Document review at volume

First-pass responsiveness and issue coding on validated lower-cost models, with privilege calls and dispositive judgments routed to frontier models. You see the model, the cost, and the citation behind every decision.

Privilege

Privilege and work-product screening

Surface candidate privileged material with the controlling document cited on each call, then route it to a named reviewer for the final designation. Documents are processed entirely inside your environment.

Case development

Transcripts, chronologies, and fact development

Build case chronologies from depositions, exhibits, and productions. Every fact traces back to a bates number or transcript line. Nothing about the matter leaves your tenancy.

Work product

Research and drafting from your own precedent

Memos, briefs, and correspondence drafted from prior work product and approved sources, grounded in documents you control. Human approval before anything is filed or sent, and every draft is logged.

Common workflows

First-pass document review
Privilege and work-product screening
Deposition and transcript analysis
Case chronology and fact development
Contract review and abstraction
Precedent and work-product search
Memo and brief drafting support
Relevant Northwood research
AI token economics →Inference cost and routing →Private AI for legal teams →Model portability for legal workflows →
Read the research

The matter stays in the building.

Providers, models, routing policy, and the open-source control layer, deployed inside your environment, documented for your team, and accountable to your professional obligations.

Learn more→Read the research