Legal research shouldn't mean digging through hundreds of pages
Searching contracts is slow, important clauses get missed, and online AI tools mean shipping confidential documents to someone else's server. Lawyer Assistant sits in between: a desktop app that does real retrieval over your files — 100% private, runs on your machine, no cloud required.
It's an open-source project (MIT license, haal-lab/Lawyer-Assistant on GitHub) built with the same local-first philosophy as GGUF Loader: your data never leaves your computer.
What Lawyer Assistant actually does
1. Ask questions in plain English — get answers with page numbers
Type "What are the payment terms?" or "Can I terminate this contract early?" and the app searches every document in your workspace, then answers with the exact source: document name, page number, and a preview of the relevant text. No guessing — every claim is backed by a passage you can open and verify.
2. Scan contracts for risky clauses automatically
One click flags the clauses lawyers care about: unlimited liability, automatic renewal, broad indemnification, vague payment terms, missing termination rights. Each finding is rated High / Medium / Low risk and explained in plain English.
3. Works with the documents you already have
- PDFs — including scanned/image PDFs (OCR)
- Word documents (.docx)
- Text files and images with text
4. Keeps everything organized by project
Different cases live in different folders, each with its own documents and search history. No mixing between matters, full isolation.
Two modes: fully local, or your own API key
Local Mode (100% private) API Mode (faster answers)
────────────────────────── ──────────────────────────
Ollama on your GPU/CPU Claude, ChatGPT, DeepSeek,
Works fully offline Groq, Mistral, Kimi, Qwen +
Works with no subscription Bring your own key, your choice
Nothing ever leaves the PC Only the needed text is sent,
never your full documents
In local mode everything runs on your hardware — ingest, index, search, and answering — so no question, document, or answer ever touches a third-party server. In API mode, only the small retrieved passage needed for the answer goes to your chosen provider, and that's shown clearly in the app.
That makes it a genuinely practical legal document chatbot for solo lawyers, small firms, and legal teams that care about attorney–client privilege — a design goal, not a checkbox.
Under the hood: a real RAG pipeline
Lawyer Assistant is a production-grade retrieval system, not a chat wrapper:
- Hybrid search — dense semantic embeddings (BGE-M3) fused with keyword/BM25 matching, so legal shorthand, typos, and paraphrases all find their target
- Cross-encoder reranking — the best candidates get re-ranked before the answer is generated
- Vector store — ChromaDB with HNSW indexing for fast retrieval over large document sets
- Visual pipeline editor — a drag-and-drop node editor (Search, Rerank, Filter, Merge) so you can see and reshape exactly how your questions are answered
- Honest answers — when the answer isn't in your documents, the app says so instead of hallucinating. Critical for legal work.
Hardware and setup
- Recommended: 16 GB RAM + 8 GB GPU. Works with or without a graphics card, and it auto-manages memory.
- Tested on laptops with 8–16 GB RAM.
- Setup: download the installer, click Install (first-time model download takes 5–10 minutes), then Launch. Under five minutes to your first question.
- Platforms: Windows, macOS, Linux.
Since it pairs with Ollama for fully-offline local models, it fits naturally alongside a local-first stack — the same idea behind running GGUF models on your own hardware.
Why it matters for legal AI
Generic AI chatbots answer from memory and can guess. Cloud "legal AI" tools hold your documents hostage and charge per-document subscription fees. Lawyer Assistant answers only from your own documents, cites every source, runs offline, and is free forever — no account, no cloud, no lock-in.
If you do legal research, contract review, or compliance work, it's worth a look. The website has full docs, installers for all three platforms, and a deep dive into how the retrieval pipeline works: lawyers-assistant.github.io.
Frequently Asked Questions
Is Lawyer Assistant really private?
Yes. In local mode, everything — ingesting, indexing, searching, and answering — runs on your own computer. No question, document, or answer ever leaves your machine. In API mode, only the small retrieved passage needed for the answer goes to your chosen provider, and the app shows you exactly what's sent. Your full documents never go anywhere.
Does it work completely offline?
Yes — with a local Ollama model, Lawyer Assistant is fully functional with no internet connection. Embeddings, search, chat history, and compliance scans all run locally. You only need the internet if you choose an optional cloud API provider instead of a local model.
Which AI models can I use with it?
Any you trust. Use a fully local Ollama model for maximum privacy, or bring your own API key from Claude, ChatGPT, Groq, Grok, DeepSeek, Mistral, Kimi, Qwen, Gemini, OpenRouter, or Together — 13+ providers in total. Switch anytime from one dropdown without changing how you work.
Does it work with my existing documents?
Yes. Lawyer Assistant reads PDFs (including scanned or image PDFs via OCR), Word documents (.docx), text files, and images with text. Drop files into your project folder and it indexes them automatically.
How much does it cost?
Nothing. It's open source (MIT license) and free forever — no subscriptions, no per-document fees, no account required. Local mode has zero running cost; API mode only costs what your chosen provider charges for your own key.
Can it spot risky clauses in a contract?
Yes. The compliance scanner flags unlimited liability, automatic renewal, broad indemnification, vague payment terms, and missing termination rights — each rated High, Medium, or Low risk with a plain-English explanation. You can also define your own playbook rules.