Horizon is an open-source, AI-powered news aggregation pipeline that collects content from RSS, Hacker News, Reddit, Telegram, X (via Apify), GitHub, OpenBB financial feeds, OSS Insight, GDELT, and Google News, then applies profile-driven scoring, deduplication, and enrichment to produce daily bilingual (English/Chinese) Markdown briefings. Users define reusable "profiles" — such as tech-news, tech-blog, and ai-creator — each specifying custom scoring rubrics, prompts, and JSON output blocks to tailor summaries, context, solutions, or takeaways per content type. Items are routed to profiles explicitly or via AI matching, filtered by configurable thresholds (e.g., 7.0 for tech-news), and deduplicated by topic before enrichment with web-researched background and community comment summaries. A balanced digest caps total items (default 20) and per-category group limits (e.g., 5 for AI, 5 for finance) to prevent topic dominance. The pipeline runs via CLI (uv run horizon --hours 24) or Docker, with an interactive wizard (horizon-wizard) that auto-generates data/config.json from declared interests. Configuration supports OpenAI, Gemini, and ChatGPT-plan providers via environment-variable-referenced API keys, plus full-text extraction via trafilatura. Outputs publish to GitHub Pages (docs/), email (SMTP/IMAP), webhooks (Feishu, DingTalk, Slack, Discord), and WeChat (iLink Bot). An MCP server exposes pipeline stages to AI assistants, and an npx-installable agent skill (Thysrael/Horizon) enables natural-language configuration. Installation requires Python (uv/pip) or Docker (Node.js 22.20+ for the skill); GitHub Actions scheduling template included.