Algroveon-News – Self-hosted AI Newsreader
Local RSS Aggregator with Local LLM: crawls feeds, summarizes articles, evaluates journalistic quality and sentiment – entirely on your own hardware.
A self-hosted news service that automatically aggregates RSS feeds, summarises articles using a local LLM and evaluates journalistic quality.
FastAPI backend with SQLite (FTS5) for full-text search, HTMX web interface and Ollama integration for LLM-powered article processing.
Algroveon-news is a standalone news service for home servers. The system subscribes to RSS feeds, continuously processes new articles in the background, and provides the results via a web cockpit and a REST API. The content processing of the RSS articles—such as for summaries, ratings, and queries—is handled by a locally running LLM on the home server.
What Algroveon-News does
Classic RSS readers primarily show raw articles. Algroveon-News goes one step further and processes the content:
- Crawler – loads feeds on a schedule (configurable interval per feed, httpx async)
- Summarizer – summarizes each article in 2–3 sentences (Ollama, local)
- Scorer – evaluates sentiment (positive / neutral / negative / alarm / satire), quality (0–1), and clickbait flags according to versioned editorial guidelines
- Category Briefings – an editorial update per topic area every 6 hours
- Narrative Tracking – shows how a topic evolves over time
- Framing Comparison – compares how different sources report on the same topic
- Argument Maps – summarizes pro and con positions for topics with mixed sentiment
- Mini-RAG – allows questions to be asked of the news database with answers including source citations
- Feed Advisor – continuously identifies low-quality feeds and suggests measures
Architecture
RSS-Feeds → Crawler (httpx + Algroveon-Parser)
↓
SQLite (FTS5) via aiosqlite
↓
┌───────────┼─────────────┐
Summarizer Scorer Briefing
(Ollama) (Ollama) (Ollama)
↓
Web-Cockpit (HTMX/Jinja2) │ REST-API (FastAPI)
↓
Algroveon-Agent
Scheduler: pure asyncio loop – no external library.
All frontend assets (HTMX, Inter font, icons) are checked into static/ – no CDN.
Editorial Guidelines: Curation based on Substance
The scoring system follows publicly documented and versioned rules. Sentiment categories and quality criteria are therefore traceable rather than a black box.
Algroveon-Parser as a Foundation
The feed parser is decoupled as its own library: Algroveon-Parser – without external dependencies, under complete internal control. This separation facilitates independent versioning and reuse, for example in Algroveon-Agent.