Four steps.
Nothing leaves your laptop.
From install to answer, your documents never touch a server.
Install. Once.
Download Privileged for macOS. The AI itself installs onto your machine — the actual model, not a connection to one. There's no server on the other end.
Create a matter. Add documents.
Each matter gets its own document pool — drag in the PDFs that belong to it. They stay right there on your computer, never leaving, no third parties.
Review it. Ask it.
Run a Matrix Review — your documents as rows, your questions as columns, a page-anchored citation behind every cell — or ask a question across the whole matter and get an answer that cites the exact passage. Every answer is generated on your machine. No server, no third party.
Privilege, intact.
No telemetry. No prompt logging. No third-party processor. Documents, extractions, answers, and history all stay on your machine.
Know exactly what it will and won’t do.
Privileged never does legal research and never drafts. It reports what’s in the documents you give it — with the page number to prove it. Every answer is anchored to a passage you can click and read yourself. And when it can’t verify an answer against the text, it says so: the cell is flagged UNVERIFIED instead of guessing.
We publish the numbers. In our benchmark — 1,310 answers across 150 real contracts, scored against expert annotations — 97.2% of verified answers were correct, and the handful that weren’t are hand-reviewed and described, one by one, on the accuracy page.
AI on your laptop, built for one job: document review.
- Model
- Two local tiers — Standard and Heavy — powered by Ollama. Match inference depth to your hardware's available memory.
- Inference
- Apple Silicon via Metal. M1 or later.
- Storage
- Stored right there on your laptop, protected the same way everything else already is.
- Network
- No document or prompt data, ever. The only network traffic is license validation and optional model-update checks.
My job is making small, local models do serious legal document work — testing them against real matters, hardening the pipeline, and re-benchmarking every release before it reaches you. Your job doesn’t change: serve your clients. When a better model ships, I do the evaluation, and your app quietly gets better — without your documents ever moving.
— Akash, founder