doc.haus

doc.haus is a self-hosted legal AI agent for lawyers and firms reviewing or drafting contracts. It's a fork of OpenCode that edits Word documents through actual tracked changes, so reviewers can accept or reject each edit in Word.

• Cited answers and redlines: Ask questions with quoted clause references or request edits to .docx files; the tool checks quotations against the current document and supports several governing jurisdictions within a matter.
• Drafting and templates: Create documents from scratch or reusable firm templates, with guided questions about terms and placeholders for details the assistant can't know.
• Contract reviews and comparisons: Run reviews with a reviewer and adversarial challenger, compare answers across contracts in a grid, compare document versions, or extract obligations and dates; firms can define repeatable review workflows.
• Document redaction: Review proposed removals before the tool deletes sensitive text and metadata from the document, verifies removal and records it in the matter's log; it also flags hidden instructions in documents as suspicious content.

Each matter has separate documents, conversations and a search index. The free, open-source project uses the MIT license and runs through a browser on macOS, Windows or Linux, with Docker also supported. It can read scanned PDFs.

Files and indexes stay on your infrastructure, and there's no doc.haus cloud. Models can run locally, or you can connect Anthropic, OpenAI or Google Vertex. Cloud providers receive the agents' prompts, so choosing a local model matters for keeping processing on your own hardware. AI output can be wrong and requires review by a licensed attorney before reliance.