A fork of Mike, extended for Swiss sovereign legal AI.

Named after Emilie Kempin-Spyri (1853–1901) — the first woman in Europe to earn a law degree, who was then denied the right to practice it in Switzerland. She deserves to be in the stack.

Emilie is a document assistant for legal work. You upload legal documents (DOCX, PDF) and work with them through a chat interface powered by a local or managed LLM. Core capabilities:

• Document chat: ask questions, extract clauses, summarize, compare versions
• Projects: organize documents into matters or workspaces, share with colleagues by email
• Tabular review: run structured clause extraction across a set of documents simultaneously
• Workflows: define reusable AI workflows that run against documents automatically
• Version tracking: upload revised documents and track changes across versions
• Swiss case law search: query federal and cantonal court decisions mid-conversation via MCP

All processing runs on infrastructure you control. No document content leaves your environment unless you configure a cloud model as fallback.

Emilie adds three capabilities on top of Mike's core document assistant:

Mike relies on Supabase for user authentication. Emilie replaces this with custom JWT + bcrypt directly against Postgres — no third-party auth service, no data leaving your infrastructure. Users and sessions are stored in your own database.

Emilie connects to any Model Context Protocol server and exposes its tools directly to the LLM. Configure servers in MCP_SERVERS and they are available in every conversation without code changes.

The following Swiss legal data sources are open, free, and MCP-ready:

Emilie routes to any OpenAI-compatible inference endpoint — no OpenAI dependency. Point VLLM_BASE_URL at a local or managed server and select "Local Model" in the UI.

Recommended model: Apertus, the open-weights LLM developed by ETH Zurich, EPFL, and the Swiss National Supercomputing Centre. Apache 2.0 licensed. Trained across 1,000+ languages with strong coverage of Swiss national languages.

Two deployment paths:

Cloud providers (Anthropic, Google) remain available as fallback but are not required.

Install dependencies:

npm install --prefix backend
npm install --prefix frontend

Create and fill in env files:

cp backend/.env.example backend/.env
cp frontend/.env.local.example frontend/.env.local

Generate a JWT secret and add it to backend/.env:

openssl rand -base64 48
# → paste result as JWT_SECRET=<output>

Run the schema migration against your Postgres database:

psql "$DATABASE_URL" -f backend/migrations/000_one_shot_schema.sql

Start backend:

npm run dev --prefix backend

Start frontend:

npm run dev --prefix frontend

Open http://localhost:3000.

• Postgres: Any Postgres database. Auth is handled by Emilie directly using JWT + bcrypt — no third-party auth service required. For a sovereign deployment, run Postgres on your own infrastructure or an Infomaniak VPS.
• Object storage: Any S3-compatible store. Infomaniak Object Storage (Switzerland) is the recommended option.
• Model: A local inference endpoint via VLLM_BASE_URL (Apertus via vLLM, or Infomaniak AI Tools). Anthropic and Gemini API keys are supported as a fallback but route documents through US cloud servers.
• LibreOffice: Required for DOC/DOCX to PDF conversion. Runs entirely locally — no data leaves your machine. Maintained by The Document Foundation (German non-profit, open-source).

Add to backend/.env:

# Option A: self-hosted Apertus via vLLM
VLLM_BASE_URL=http://localhost:8000/v1
VLLM_MAIN_MODEL=<apertus-model-id>

# Option B: Infomaniak AI Tools (Swiss managed, includes Apertus)
VLLM_BASE_URL=https://api.infomaniak.com/2/ai/<product_id>/openai/v1
VLLM_API_KEY=<infomaniak-api-key>
VLLM_MAIN_MODEL=apertus

Select "Local Model" in the chat model picker once VLLM_BASE_URL is set.

Add to backend/.env:

MCP_SERVERS=[
  {"name":"entscheidsuche","url":"https://mcp.entscheidsuche.ch/mcp"},
  {"name":"opencaselaw","url":"https://mcp.opencaselaw.ch/mcp"},
  {"name":"fedlex","url":"https://fedlex-connector.ch/mcp"}
]

Multiple servers are supported. Each server's tools appear automatically in every conversation. API keys are optional — omit the apiKey field for servers that require no authentication.

npm run build --prefix backend
npm run build --prefix frontend
npm run lint --prefix frontend

AGPL-3.0-only. See LICENSE.

This project is a fork of Mike by Will Chen, used under AGPL-3.0.