Horizon: GitHub repo for AI news briefings

Horizon is an open-source news radar that builds daily briefings in English and Chinese.

Thysrael’s Horizon repo packages a self-hosted news pipeline that fetches, deduplicates, scores, enriches, and summarizes stories into daily briefings. The project is on GitHub, with 3.9k stars, 540 forks, and 159 commits listed on the repository page.

What changed

Horizon is built to give developers a first-pass briefing over noisy sources instead of a single-model summary. It pulls from GitHub releases and user activity, Hacker News, Reddit, RSS, Telegram, Twitter/X, and OpenBB financial news watchlists.

The pipeline does more than collect links. It can deduplicate the same story across feeds, score items with Claude, GPT, Gemini, DeepSeek, Doubao, MiniMax, Ollama, or any OpenAI-compatible API, add web context, summarize comments, and publish the result to GitHub Pages, email, chat apps, webhooks, or MCP.

• Source setup is driven by one JSON config.
• The wizard can generate a personalized source list from user interests.
• English and Chinese briefings come from the same inputs.
• Output targets include Pages, SMTP email, Feishu/Lark, DingTalk, Slack, Discord, and custom webhooks.

Why it matters

For teams that track fast-moving tech or market news, Horizon reduces the manual work of scanning feeds and copying items into a daily digest. The repo also makes the scoring layer adjustable, so users can set thresholds and model choices instead of accepting a fixed ranking.

That matters for builders who want a private briefing system they can run locally or in Docker. The project also ships a GitHub Actions cron example, which makes it easier to turn the repo into an automated daily pipeline without building the orchestration from scratch.

The main question is not whether AI can summarize news, but which sources, filters, and delivery channels you want in the loop.