A news radar that actually reads the comments

What it does Horizon is a Python pipeline that ingests Hacker News, Reddit, RSS feeds, Telegram, Twitter/X, GitHub activity, and OpenBB financial watchlists. It deduplicates overlapping stories across platforms, prompts an LLM to score each item 0–10 against your interests, and discards anything below a configurable threshold. Survivors get web-researched background context and summaries of community comments, then ship as structured Markdown to GitHub Pages, email, chat webhooks, or an MCP server.

The interesting bit Most summarizers treat comments as noise; Horizon treats them as signal. It pulls top comments from Hacker News and Reddit into the briefing, capturing how people are reacting, not just what happened. The entire fetch-score-filter-enrich workflow is also exposed as an MCP server, so another AI assistant can call pipeline steps on demand.

Key highlights

• Cross-platform deduplication merges the same story from HN, Reddit, and RSS before it reaches your inbox.
• Bilingual generation produces both English and Chinese daily briefings from the same source set.
• Pluggable LLM backends include Claude, GPT, Gemini, DeepSeek, Doubao, MiniMax, Ollama, and generic OpenAI-compatible APIs.
• Delivery channels span GitHub Pages, self-hosted SMTP/IMAP newsletters, Feishu/Lark, Slack, Discord, DingTalk, and custom webhooks.
• An MCP server exposes the pipeline as callable tools for external AI assistants.

Caveats

• OpenBB financial news support requires an optional extra and may need manual binary-only installation if pre-built wheels are unavailable.

Verdict Grab it if your morning routine involves manually checking Hacker News, Reddit, and GitHub releases. Skip it if tuning a JSON config file to set an LLM scoring threshold feels like more work than just scrolling X.