Feeds Fun

Feeds Fun is a self-hosted news and RSS reader that automatically assigns tags to feed entries and lets you define rules to score, filter, and sort news. It separates API/frontend from background workers: a lightweight ASGI HTTP API serves the UI while workers handle feed loading and tag analysis.

Key Features

• Automatic tag assignment via multiple processors (domain, native feed tags, uppercase-title, and LLM-based processors).
• Configurable LLM processors with support for OpenAI (ChatGPT) and Google Gemini; custom API endpoints are configurable.
• Rules-based scoring system to rank and surface important items by tags and scores.
• Filtering and sorting by score, date, tags, and read/unread state.
• Worker-based architecture with dedicated loader (fetches/parses feeds) and librarian (analyzes entries) workers.
• Tag normalization pipeline to merge and clean raw tags (splitters, replacers, blacklists).
• Docker-based examples for single-user and multi-user deployments; backend package available on PyPI and frontend as an npm package.

Use Cases

• Reduce noise from many RSS subscriptions by scoring and surfacing only high-relevance items.
• Curate tagged collections of feeds for teams or multi-user environments with shared collections.
• Enrich and organize large feed corpora using LLM-generated tags for discovery and automated workflows.

Limitations and Considerations

• LLM tagging requires external API keys (OpenAI or Google Gemini) by default; using these incurs API costs and raises privacy considerations unless custom endpoints or providers are configured.
• Local/self-hosted LLM support is not available out-of-the-box; current LLM processors rely on hosted providers.
• Running the librarian worker with full features may require additional components (recommended SpaCy model) and a PostgreSQL database, adding operational complexity.
• Tag normalization and some features are actively evolving; configuration may require reading example config files and tuning.

Feeds Fun is suited for users who want powerful, tag-driven control over large or noisy feed collections. Its worker-based design and configurable LLM processors make it flexible for personal and team deployments.