Show HN: Judicex – Open-source legal AI that abstains instead of hallucinating

Judicex, an open-source legal AI platform, launched today with a system that abstains from answering when evidence is insufficient rather than generating false information. The software operates on a local SQLite knowledge base, allowing lawyers to ingest official sources and private files, run deterministic workflow checks, and generate drafts while keeping all matter data on the user's own hardware. Unlike closed SaaS products or generic RAG demos, Judicex enforces a structured evidence model where citations are bound only to retrieved documents, ensuring the AI cannot improvise legal answers.

Open-source legal AI softwarefor evidence-grounded drafting, matter management and AI-assisted legal work — with an answer contract that fails closed instead of hallucinating. Judicex is a workspace for lawyers and legal teams. You can ingest official sources and private matter files, run deterministic workflow checks, generate drafts in a Word-style split-view editor, and chat with an LLM whose answers are bound to the evidence in a SQLite knowledge base you control. The project ships with a Flask web UI, a CLI, and an MCP stdio server, and talks to Ollama, OpenAI, Anthropic, OpenAI-compatible providers — or runs with no LLM at all. Open-source vs. cloud.This repository is the open-source build of Judicex: install it on your laptop, on a workstation, or on a private server inside the firm. A managed cloud version with hosted multi-tenant deployment, SSO and audit features is on the roadmap as a separate offering; this codebase is and will remain Apache-2.0. Most legal AI products today are either: Closed SaaS that ship your client data to a vendor Harvey, Legora, … , or Generic RAG demos that paste retrieved chunks into a prompt and hope. Judicex takes a different bet. In legal work, the value is not in the prompt — it is in the structured, verifiable evidence the answer stands on . So the codebase is organised around that idea: Two evidence layers, never confused. Legal sources documents , document versions , legal atoms , entities , edges are the only thing that can produce citations . Operational notes agent memories store preferences, decisions and lessons — they shape how the agent works, but are never cited as law. Versioned official sources. Real ingestion from Normattiva with URN extraction, versioned over time effective from / effective to so you can ask “what did this article say on date X?” instead of pretending the law is timeless. Answer contract that fails closed. answer contract.py + numeric verifier.py + confidence.py enforce the states grounded / limited / abstain / chat with JSON-schema validation, citations bound to retrieved document id s only, and temperature 0. If evidence is insufficient, Judicex abstains. It does not improvise. Workflow packs as data. Matter-analysis profiles e.g. civil debt recovery, opposition to injunctions, generic file review are JSON workflow packs, versioned and swappable per vertical / firm — not application logic. You control where it runs. SQLite, Flask, vanilla JS, no external build step. The same codebase runs on a single workstation or on a private server inside the firm. Matter data lives in the SQLite database and the attachment directory you point Judicex at — nothing leaves the host until you explicitly point a chat at a hosted LLM. This repository is v0.2.0-alpha : a usable prototype, not production SaaS. Sources & evidence - Local SQLite knowledge base with versioned document history and legal atoms. - One-click import of official source bundles Normattiva-driven . - Two-tier evidence model: legal sources citable and operational notes style and decisions . Matters & files - Private matters, folders, versioned matter documents. - Upload PDF, DOCX, text, markdown, CSV, JSON, images. - PDF / image preview, stored-text viewer, OCR via Ollama. - Matter facts, timelines, parties, amounts and deadlines extracted deterministically. Workflows & analysis matter analysis against a thesis: explicit proof profile, present / partial / missing elements, next actions.- Built-in workflow packs and a custom workflow builder. - Tabular review with editable cells, filtering, sorting, saved views and CSV / XLSX / DOCX export. Drafting - Built-in templates plus a custom template editor. Split-view drafting page Claude- / ChatGPT-artifact style : instruction column on the left, live Word-style document preview on the right.- Generate, save into the matter, copy, export to DOCX / PDF. AI surface - Persisted chat sessions sidebar, deletable . - Multi-provider: Ollama, OpenAI, Anthropic, OpenAI-compatible, no-LLM. - Grounded answer engine with citations bound to evidence. - MCP stdio server for integration with MCP-aware clients. - CLI utilities judicex , judicex-mcp , judicex-agent , judicex-web . Privacy & ops - Local password gate, backup and restore, optional matter-file encryption. - API keys stay in .env / shell, never in the browser. Drop PNGs into docs/screenshots/ and reference them here once you publish: docs/screenshots/drafts-split-view.png , docs/screenshots/matter-analysis.png , etc. They are intentionally not committed yet — the repository is currently text-only to keep diffs clean. ┌──────────────────────────────┐ ┌──────────────────────────────┐ │ Web UI Flask + vanilla JS │ │ CLI / MCP stdio / Agent │ └──────────────┬───────────────┘ └──────────────┬───────────────┘ │ │ ▼ ▼ ┌─────────────────────────────────────────────────┐ │ agent runtime + answering │ │ intent router → tools → answer contract → out │ └────────────┬───────────────────────┬────────────┘ │ │ ▼ ▼ ┌──────────────────────┐ ┌────────────────────────┐ │ llm provider.py │ │ store.py SQLite │ │ ollama / openai / │ │ documents, atoms, │ │ anthropic / oai-c / │ │ entities, edges, │ │ none │ │ matters, agent notes │ └──────────────────────┘ └────────────────────────┘ See docs/ARCHITECTURE.md /JustVugg/judicex/blob/main/docs/ARCHITECTURE.md for the full layer description and database tables, and docs/JUDICEX PRODUCT STRATEGY.md /JustVugg/judicex/blob/main/docs/JUDICEX PRODUCT STRATEGY.md for the product thesis. git clone https://github.com/JustVugg/judicex.git cd judicex python3 -m venv .venv source .venv/bin/activate pip install -e ". crypto " cp .env.example .env judicex init-db --db ./memory.db judicex web --db ./memory.db --area civile --port 5051 Then open http://127.0.0.1:5051 http://127.0.0.1:5051 . For OCR on scanned PDFs: pip install -e ". crypto,ocr " Judicex reads provider settings from the in-app Settings page and from environment variables. API keys are kept out of the frontend on purpose. Local Ollama default, fully offline : ollama serve ollama pull qwen2.5:7b JUDICEX LLM PROVIDER=ollama OLLAMA HOST=http://127.0.0.1:11434 OLLAMA MODEL=qwen2.5:7b JUDICEX OCR MODEL=glm-5:cloud OpenAI: JUDICEX LLM PROVIDER=openai OPENAI API KEY=... OPENAI MODEL=gpt-4.1-mini Anthropic / Claude: JUDICEX LLM PROVIDER=anthropic ANTHROPIC API KEY=... ANTHROPIC MODEL=claude-3-5-sonnet-latest OpenAI-compatible LocalAI, vLLM, hosted compatible APIs : JUDICEX LLM PROVIDER=openai compatible OPENAI COMPATIBLE API KEY=... OPENAI COMPATIBLE BASE URL=http://127.0.0.1:8000/v1 OPENAI COMPATIBLE MODEL=local-model No-LLM mode also works: deterministic matter analysis, draft templates, search and audit still run. Synthetic demo database: python scripts/create demo.py --db ./memory.demo.db judicex web --db ./memory.demo.db --area civile --port 5051 CLI examples: judicex matter-create --db ./memory.db --title "Recupero credito Beta" \ --client-name "Alfa S.r.l." --area civile judicex matter-add-doc --db ./memory.db --matter-id "matter:..." \ --file ./examples/demo matter.md --kind memo judicex matter-context --db ./memory.db --matter-id "matter:..." \ --query "fattura pagamento" judicex list-workflow-packs --db ./memory.db judicex matter-analyze --db ./memory.db --matter-id "matter:..." \ --thesis "ricorso per decreto ingiuntivo" judicex ask --db ./memory.db --provider ollama --model qwen2.5:7b \ --area civile --matter-id "matter:..." \ --question "Che elementi mancano?" Operational notes are separate from legal sources. Legal sources hold statutes, official documents, legal atoms and citations; operational notes hold how the assistant should work — preferences, decisions, lessons, recurring strategies, and matter-specific notes. judicex memory-add --db ./memory.db --kind preference \ --title "Stile recupero crediti" \ --content "Per recupero crediti B2B usa tono pratico." \ --tag recupero crediti --importance 0.9 judicex memory-search --db ./memory.db --query "recupero crediti" The open-source build is designed to run wherever you install it: - On a single workstation, with a local LLM via Ollama, for a sole practitioner who wants zero recurring cost. - On a private server inside the firm, behind a real WSGI stack, VPN and backups see SECURITY.md /JustVugg/judicex/blob/main/SECURITY.md for the hardening checklist . - On any cloud you operate yourself — the codebase has no hard dependency on a specific provider. A managed cloud version of Judicex multi-tenant, SSO, hosted upgrades is on the roadmap as a separate offering. This repository remains Apache-2.0; the cloud product is built on top of it, not in place of it. judicex/ ├── judicex memory os/ Python package engine, store, agent, web │ ├── store.py SQLite schema, persistence, search │ ├── web app.py Flask routes & JSON APIs │ ├── agent runtime.py tool-calling agent loop │ ├── answering.py grounded answer pipeline │ ├── answer contract.py claim/citation/numeric validation │ ├── matter analysis.py deterministic matter workflow analysis │ ├── llm provider.py Ollama / OpenAI / Anthropic / compatible │ ├── mcp stdio.py MCP stdio server entry point │ ├── cli.py judicex CLI │ ├── workflow packs/ JSON workflow packs │ ├── templates/ Flask + draft templates │ └── static/ Vanilla JS UI ├── tests/ unittest suite ├── benchmarks/ public benchmark fixtures ├── docs/ architecture, strategy, installers └── scripts/ dev / test / installer / demo helpers The Python package is named judicex memory os for historical reasons; the product is just Judicex . For local development, install the package in editable mode and run the standard test script. The tests use Python's built-in unittest ; no pytest dependency is required. pip install -e ". crypto,test " scripts/dev.sh scripts/test.sh python scripts/run public benchmark.py --db ./memory.benchmark.db The web UI is server-rendered Flask plus static JavaScript. There is no TypeScript, Next.js, Supabase or external frontend build step — and there is no plan to add one. Contributions that require one will be politely declined. - Windows: powershell -ExecutionPolicy Bypass -File .\scripts\install windows.ps1 - macOS: bash scripts/install macos.sh See docs/INSTALLERS.md /JustVugg/judicex/blob/main/docs/INSTALLERS.md . The included .gitignore excludes the things that must not end up on GitHub. Before your first push, double-check that none of these are tracked: .env memory.db and any other .db / .sqlite files memory files/ , uploads/ , private/ - API keys, real client documents, screenshots with private data A pre-publish checklist is documented in SECURITY.md /JustVugg/judicex/blob/main/SECURITY.md . See ROADMAP.md /JustVugg/judicex/blob/main/ROADMAP.md . Highlights: - Better PDF viewer with page thumbnails and annotation coordinates v0.3 - Rich document editor with version diff and accept/reject v0.4 - Optional auth layer + desktop packaging v0.5 - Multi-user workspace + RBAC + audit export later Contributions are welcome — see CONTRIBUTING.md /JustVugg/judicex/blob/main/CONTRIBUTING.md . The general rules: - Keep the stack Flask / Python / HTML / JavaScript / SQLite. - Never commit private legal documents, real databases, uploads or API keys. - Add or update tests for changes touching API behavior, persistence or workflows. - Keep provider integrations behind llm provider.py . Apache License 2.0 — see LICENSE /JustVugg/judicex/blob/main/LICENSE and NOTICE /JustVugg/judicex/blob/main/NOTICE . Apache-2.0 was chosen because it is permissive and includes an explicit patent grant, which matters for a domain that intersects regulated work. Judicex is a software assistant for legal drafting, document organization and evidence-grounded answering. It does not provide legal advice and does not replace a qualified legal professional. Outputs may be incomplete, incorrect, outdated or unsuitable for a specific jurisdiction. See DISCLAIMER.md /JustVugg/judicex/blob/main/DISCLAIMER.md . To report a vulnerability, please follow SECURITY.md /JustVugg/judicex/blob/main/SECURITY.md . Do not open public issues with exploit details. If you build something on top of Judicex — a vertical pack for criminal law, a richer drafting UI, an integration with case-management software — please open an issue or a PR. The point of an open-source legal AI is that the verticals are built in the open, not behind a sales NDA.