v0.6.0-beta — Admin-configurable pipeline

Configuration moves out of the source and into the database. Topics, RSS feeds, LLM prompts, and the pipeline run schedule are all editable from the admin UI now, so running your own instance no longer means editing Python to change what gets fetched or how it gets summarized.

Also in this release: Groq and Gemini join Ollama as LLM providers, a fast model tier cuts the cost of the pipeline's ~1,100 mechanical LLM calls per run, the app is safe to run with multiple gunicorn workers, and five security issues are closed.

Highlights

• Admin-configurable pipeline (#1) — topics, RSS feeds, LLM prompts, and the run schedule are now DB-backed with full admin CRUD
• install.sh — one-command setup
• More LLM providers — LLM_PROVIDER now accepts groq and gemini alongside ollama
• Fast model tier — route high-volume mechanical calls to a smaller model while summaries stay on the main one
• Multi-worker gunicorn with process-safe task claiming
• Container restart from Admin Tools, via a separate proxy service so the web app never holds the Docker socket

Security

CSRF protection app-wide; a real SECRET_KEY is now required at startup; open redirect fixed; Postgres and the admin app are no longer bound to all interfaces by default; .env backups are now gitignored.

Upgrading

docker compose up -d --build
docker compose exec app flask db upgrade

Restart the scheduler container afterwards. If you customized the RSS feed lists in news_fetcher/rss_fetcher.py, back them up first — feeds now come only from the database and your file edits will not be carried over.

Thanks to @scumola and @Sanjays2402 for contributions in this release (#10, #12, #16, #19, #20).

Full notes: CHANGELOG.md