Legal Research Assistant

Part of the worlds-biggest-software-project initiative.

An open-source, AI-native legal research assistant that delivers hallucination-resistant case law search, citation analysis, and brief drafting on top of public-domain legal data.

The Legal Research Assistant is a retrieval-augmented research platform built for attorneys, legal aid organisations, law students, and solo practitioners who are priced out of Westlaw, Lexis+, and Harvey. It combines openly licensed primary law sources (CourtListener, Public.Resource.Org, GPO) with grounded LLM reasoning to answer plain-English legal questions, validate every citation against the source corpus, and produce Bluebook-formatted research memos.

Why Legal Research Assistant?

Incumbent pricing excludes most of the profession. Harvey AI runs ~$1,200/lawyer/month with a 20-seat minimum; Westlaw + CoCounsel All Access reaches $500–$800/user/month; Lexis+ with Protégé runs $250–$475/user/month. Solo practitioners, legal aid organisations, and law students cannot access frontier legal AI at these prices.
Citation hallucination is still unsolved. The Stanford CodeX 2024 study measured a 34% citation error rate in Westlaw AI-Assisted Research and 17% in Lexis+ AI — unacceptable for a domain governed by ABA Model Rule 1.1 (Competence).
The data layer is open; the application layer is not. CourtListener, Juriscraper, and Eyecite (all BSD or public domain) provide tens of millions of indexed court opinions and citation extraction tooling, but no AI-native research assistant has been built on top of them as of 2026.
Closed databases create lock-in. Thomson Reuters' acquisition of Casetext, Clio's $1B acquisition of vLex, and the LexisNexis–Harvey partnership concentrate legal AI inside proprietary ecosystems with opaque models and rising prices.
Underserved buyer segments exist. Mid-market litigation firms, in-house legal teams, legal aid, and solo practitioners are explicitly the segments where incumbents either overprice or under-serve.

Key Features

Grounded Research and Citation Validation

RAG-based legal research over CourtListener's case law corpus: natural-language question to retrieved cases to AI-synthesised answer with inline citations
Post-generation citation validation: every cited authority is checked against the corpus to flag hallucinated or fabricated citations
Basic citator: track whether a cited case has been overruled, distinguished, or criticised in subsequent decisions within the corpus
Full-text and semantic (embedding-based) search across the case law corpus
Bluebook and ALWD citation formatting in AI-generated output

Jurisdiction-Aware Reasoning

Jurisdiction filter constraining results to a specified federal circuit or state
Authority hierarchy awareness: distinguishes binding from persuasive precedent and surfaces circuit splits
Ranking of retrieved cases by binding-authority status in the user's jurisdiction
Recency-aware prioritisation of controlling decisions

Brief Drafting and Document Analysis

Brief drafting assistant that generates argument sections with inline, page-referenced citations traceable to source documents
Word-format export with Bluebook citation formatting
Document upload and issue-spotting from briefs, contracts, or statutes
Research history organised by matter or topic for client-linked workflows

Regulatory Monitoring

Configurable watch on Federal Register dockets and state regulatory sources
AI-generated summaries of regulatory changes mapped to affected practice areas
Continuous monitoring designed for in-house legal teams and corporate counsel

Open Corpus and Extensibility

Built on CourtListener (public domain US court opinions), Juriscraper (BSD), and Eyecite (BSD)
Expandable corpus including Public.Resource.Org statutes, GPO Code of Federal Regulations, and openly available state codes
Public REST API for programmatic access and embedding in matter management or CLM platforms

AI-Native Advantage

The project is designed around hallucination elimination as its primary differentiator: a retrieval-augmented architecture grounds every answer in version-controlled primary sources, and a post-generation validator checks each citation against CourtListener before returning results. Beyond grounded retrieval, the assistant applies jurisdiction-aware reasoning — understanding controlling versus persuasive authority, circuit splits, and federal versus state hierarchy — and generates briefs with citation provenance traceable to specific pages and paragraphs of the cited source. A regulatory monitoring agent extends the same grounded approach to ongoing surveillance of the Federal Register and state dockets.

Tech Stack & Deployment

The system is built on open legal data infrastructure: CourtListener's REST API and bulk data, Juriscraper for court website scraping, and Eyecite for citation extraction. RAG retrieval runs over a verified, version-controlled corpus with embedding-based semantic search and post-generation citation validation. Deployment targets include hosted SaaS for general use and on-premises / locally-hosted LLM deployment for firms and legal aid organisations with strict data residency, attorney-client privilege, or GDPR constraints. Outputs are formatted to Bluebook and ALWD citation standards; an API layer is planned to enable embedding in matter management, CLM, and document automation platforms.

Market Context

The LegalTech AI market is $2.82B in 2025, projected to reach $3.7B in 2026 at 31.4% CAGR (The Business Research Company), with legal research the highest-revenue AI segment at 29.1% of legal AI software spend. Incumbent pricing ranges from ~$1,200/seat/month (Harvey AI) and $500–$800/seat/month (Westlaw + CoCounsel) at the high end, down to $99–$299/month (Paxton AI) and free–$195/month (Fastcase) at the accessible end. Primary buyers include BigLaw associates and partners, mid-market litigation firms, in-house legal teams, law students and legal aid organisations, and solo practitioners — with the latter three groups most underserved by current pricing.

Project Status

This project is in the research and specification phase.
Contributions, feedback, and domain expertise are welcome.

Contributing

We welcome contributions from developers, domain experts, and potential users. See CONTRIBUTING.md for guidelines.

Important: All contributions must be your own original work or clearly attributed open-source material with a compatible licence. Copyright infringement and licence violations will not be tolerated and will result in immediate removal of the offending contribution. If you are unsure whether a piece of code, text, or other material is safe to contribute, open an issue and ask before submitting.

Licence

Licence to be determined. See discussion for context.

lexpeer