Touchline

One page of football a day. No scroll, no bait.

Live: touchline-chelsea.pages.dev · why this exists

Touchline is a self-hostable football companion built on a simple model (borrowed from antifeed): a static site whose only database is a JSON file, plus a brain — headless Claude Code, run locally each morning — that gathers the facts, reads the day's discourse, and writes one calm, opinionated page: what happened, what's on today, and what's worth your click.

The boundary rule: Python produces facts, the brain produces prose, the site produces pixels.

touchline.config.json   your club, competitions, timezone, feeds, voice
src/touchline/          facts CLI: `touchline facts` -> JSON bundle (ESPN by default)
brain/                  prompt + sources + curate.sh (headless Claude Code)
site/                   static reader; site/data/digests.json is the database

Daily use

./brain/curate.sh              # with morning coffee: facts -> brain -> validate -> commit -> deploy
./brain/curate.sh --no-deploy  # same, but stop before deploying

Self-hosting

1. Fork this repo.
2. Edit touchline.config.json — club name + club code, plus competitions, your timezone, feeds, and the voice you want the digest written in. The code must match your configured source (step 3): the ESPN abbreviation for espn (default — e.g. MAN for Manchester United, not football-data's MUN), the football-data.org TLA for football-data, or nothing at all for api-football — it has no club codes, so the club name must match its team name exactly instead.
3. Pick a data source in touchline.config.json ("source"): espn+thesportsdb (recommended, no key needed — ESPN for league data plus TheSportsDB's free per-team feed, which catches tour friendlies ESPN misses; set club.thesportsdb_id to your club's numeric TheSportsDB team id), espn (no key needed), api-football (set API_FOOTBALL_KEY, from api-football.com), or football-data (set FOOTBALL_DATA_TOKEN, from football-data.org).
4. Install Claude Code (the brain runs claude -p), plus uv and node.
5. First deploy: npx wrangler login, then npx wrangler pages project create <your-project> --production-branch main (once — deploy runs non-interactively and cannot create the project), then ./deploy.sh. After that, run ./brain/curate.sh each morning.

Nothing runs centrally: the brain runs on your machine, with your preferences, and publishes to your Cloudflare Pages project.

Development

uv run pytest -q                     # Python suite
node brain/validate.mjs              # check the database
cd site && python3 -m http.server    # local preview

Deliberately not built (yet)

Match pages, standings pages, PWA install, push/Telegram/email delivery, preference sync, automated scheduling. Each may return if the habit sticks.

License

MIT — see LICENSE.