LegisNote

A web app for law students and lawyers to study and navigate (Czech) legislation — tagging, annotating, commenting, linking, and versioning laws at every level — plus a Python ingestion pipeline that converts official Czech laws into clean, versioned Markdown.

Status: early scaffold. The ingestion pipeline runs end-to-end for the PoC law (91/2012 Sb.); the web app is a working skeleton (reader page + tRPC API + importer endpoint) with DB schema in place. See docs/ for the design.

Documentation

Doc	What
requirements.md	Requirements + decisions log (D1–D11)
docs/architecture.md	System architecture, stack, deployment
docs/data-model.md	PostgreSQL schema + versioning model
docs/research-czech-legislation-data.md	Where/how to get Czech law data

Repository layout

apps/web/            Next.js app (frontend + tRPC API + Prisma)
services/export/     Print/electronic export worker (placeholder)
tools/ingestion/     Python ingestion pipeline (Czech law -> Markdown + manifest)
packages/shared/     Cross-language contract: manifest JSON schema + TS types
source/              Law artifacts: pdf/ (raw) · md/ (clean) · manifest/ · cache/ (gitignored)
infra/               docker-compose, Caddyfile, db/schema.sql, .env.example
docs/                Design docs

The TS app and the Python tool meet at one contract: clean.md + manifest.json, validated against packages/shared/schema/manifest.schema.json.

Prerequisites

• Node 20 + pnpm (corepack enable)
• Python 3.11+
• Docker (for Postgres/MinIO)

Quickstart

1. Ingest a law (works today, no API key — uses the LawGPT.cz proxy)

cd tools/ingestion
python -m venv .venv
.venv/Scripts/python -m pip install -e ".[dev]"     # Windows; use .venv/bin on *nix

# Fetch + parse + emit clean Markdown and a validated manifest into source/:
.venv/Scripts/legisnote-ingest ingest \
  --citation 91/2012 \
  --title "Zákon o mezinárodním právu soukromém" \
  --effective-from 2023-09-23

# Or parse a local file fully offline:
.venv/Scripts/legisnote-ingest ingest --citation 91/2012 \
  --title "…" --effective-from 2023-09-23 --from-file path/to/text.md

pytest    # 5 tests: parser, emitter, schema validation

Outputs: source/md/91-2012.md and source/manifest/91-2012.json.

2. Bring up the database

cp infra/.env.example infra/.env   # then edit secrets
pnpm db:up                          # postgres with schema from infra/db/schema.sql

3. Run the web app

pnpm install
cp apps/web/.env.example apps/web/.env
pnpm --filter @legisnote/web db:pull       # regenerate Prisma schema from the live DB
pnpm --filter @legisnote/web db:generate
pnpm dev                                    # http://localhost:3000

4. Import an ingested law into the app (importer endpoint is a stub)

LEGISNOTE_IMPORTER_URL=http://localhost:3000/api/import \
LEGISNOTE_IMPORTER_TOKEN=<IMPORTER_TOKEN from apps/web/.env> \
  tools/ingestion/.venv/Scripts/legisnote-ingest import-manifest source/manifest/91-2012.json

The importer currently validates auth + manifest and returns 501 Not Implemented. The DB upsert (stable-node matching across snapshots) is the next step — see docs/data-model.md §2.

Secrets (D10/D11)

Your own Anthropic Claude API key is used only by the ingestion PDF-fallback path. The eSbírka API key (per-request, from the Ministry of Interior) is optional and arrives later. Both live in infra/.env (gitignored) — see infra/.env.example.

License

Code: TBD. Czech law texts are public domain (§ 3(a) of Act 121/2000 Sb.).