// the find
Thysrael/Horizon
📡 Your own AI-powered news radar. Generates daily briefings in English & Chinese. | 用 AI 构建你专属的新闻雷达
Horizon is a Python pipeline that pulls from RSS, Hacker News, Reddit, Telegram, X, GitHub, and several financial and trend sources. An LLM scores and filters each item against a profile, then writes a daily briefing in English and Chinese. It suits people who want a self-hosted personal digest and are willing to edit prompts and JSON config to tune what gets through.
Profiles keep the editorial rules in markdown prompts plus a JSON file, so changing what counts as worth reading means editing match.md and analysis.md, not Python. Analysis, filtering, and dedup run before enrichment, so the web search and comment fetching only touch items that survived. The balanced digest caps each category group after profile filtering, which keeps one busy feed from filling the whole briefing. A single Markdown output fans out to Pages, SMTP email, webhooks, and WeChat, and the MCP server exposes the pipeline stages to an AI assistant.
Output quality is whatever the scoring prompt and model produce. The README has no evaluation or sample of picks judged against a reference, so the example threshold of 7.0 is a starting guess until you tune it on your own feeds. Every surviving item gets LLM calls for analysis and enrichment, which makes a daily run a recurring API bill, and the README does not state a per-run cost. The Twitter source depends on an Apify account or on Playwright with exported browser cookies, which breaks when the platform changes, and the Dockerfile does not install the Playwright browser for that extra. The README opens with badges, sponsor tables, and QR codes before any substance, and the daily GitHub Actions workflow ships disabled, so scheduling is left to you.