Legal AI your team actually owns.
LQAI is a self-hosted AI platform for legal teams. Bring your own keys, run it where you want, own your data — conversational chat, verified citations, playbooks and tabular review, with every skill open and readable.
Run it in three commands
You need Docker and one model key, or Ollama for fully offline. The stack comes up at localhost:3000 with a first-run admin account.
$ git clone https://github.com/LegalQuants/lq-ai.git $ cd lq-ai && cp .env.example .env # add one provider key $ docker compose up -d # → http://localhost:3000
Prefer the long version? Read the 20-minute quickstart — clone to first cited NDA review.
Bring your own model — plus Ollama for fully local inference. Run privileged matters air-gapped on Tier 1.
Pre-release · v0.4.x · M1–M4 shipped — deployable today; some edges are deferred and listed in HONEST-STATE.md.
Built for the work lawyers actually do
Every artifact that shapes an answer is visible work product.
Most legal-AI tools are closed SaaS: you pay per seat for a hidden system prompt and send privileged files to someone else’s cloud. LQAI inverts that. The skills, the playbooks, the citation logic — all open. Click view this skill on any answer and read the exact file that produced it.
You do not take our word for a single claim on this page. The verification path runs through code you can read — including the parts not wired yet, which we list honestly in HONEST-STATE.md.
See it before you install
Shipped, with an honest roadmap
M1 through M4 are shipped. Forward-looking work — in-Word surfaces, the contract relationship graph, the MCP client — is community-driven and listed with verification paths.
| Milestone | Status | What landed |
|---|---|---|
| M1 — Foundation | ✓ Shipped | Self-hostable release, 10 starter skills, Projects, Org Profile, tier awareness, hybrid retrieval. |
| M2 — Citations + privacy | ✓ Shipped | Four-stage Citation Engine and the anonymization layer with custom legal recognizers. |
| M3 — Playbooks + review | ✓ Shipped | Playbooks and tabular review end-to-end; Word add-in scaffold; Slack/Teams intake plumbing. |
| M4 — Autonomous layer | ✓ Shipped | Opt-in background agents under hard cost, halt and phase-gate brakes. Off by default. |
Help build it
LQAI is built in the open. Open jobs are read live from the repo — filter by what you are and open the one-page spec. You do not need to be a coder; many tasks are made for lawyers and compliance people.
EASIEST-CONTRIBUTIONS.md · live · 8 open
Effort: S = under a day · M = a few days · L = more than a week. Read the source — EASIEST-CONTRIBUTIONS.md on GitHub.
- 01Pick a task.Browse the live board above — each shows who it is for, how big it is, and how ready it is. Many need no code at all.
- 02Claim it.Open the spec, file a GitHub issue named after the job, and comment “I’d like to take this.” A maintainer replies in about a week.
- 03Build and hand it in.Each task links to a one-page mini-PRD with a clear “done”. Open a pull request; once it passes the checklist it ships — with your name on it.
Stuck? Ask on GitHub. People will help — there are no dumb questions.
Run legal AI you control.
Free, open source, self-hosted. Bring your own keys and have it running in about fifteen minutes.