Horizon (github.com/Thysrael/Horizon, MIT license) is an open-source Python 3.11+ information aggregation system that lets AI act as a taste engine rather than a content generator. It pulls from eight sources—Hacker News, Reddit, RSS/Atom, GitHub, Telegram, Twitter/X (via Apify), OpenBB finance, and OSS Insight—through an async, stage-based pipeline: concurrent fetching, URL deduplication, AI scoring on five dimensions (0-10), threshold filtering, semantic deduplication, background enrichment via DuckDuckGo search with verified citations, and programmatic daily digest generation. It natively produces bilingual English/Chinese output, supports nine AI providers (Claude, GPT, Gemini, DeepSeek, Ollama local, etc.), and delivers via Markdown, GitHub Pages, email, Feishu, Slack, Discord, and custom webhooks. An MCP server exposes 12 tools and 7 resources so AI assistants can inspect and run pipeline stages. The review praises its graceful degradation, Pydantic-validated ContentItem core, and deliberate choice of deterministic (non-AI) final rendering for traceability, while honestly noting untested scoring quality, Apify/Telegram and DuckDuckGo single-point risks, and JSON-file-only persistence with no database.
Horizon Deep Dive: An AI-Powered Personal Information Radar That Reads the Internet For You
Summary
Horizon (github.com/Thysrael/Horizon, MIT license) is an open-source Python 3.11+ information aggregation system that lets AI act as a taste engine rather than a content generator. It pulls from eight sources—Hacker News, Reddit, RSS/Atom, GitHub, Telegram, Twitter/X (via Apify), OpenBB finance, and OSS Insight—through an async, stage-based pipeline: concurrent fetching, URL deduplication, AI scoring on five dimensions (0-10), threshold filtering, semantic deduplication, background enrichment via DuckDuckGo search with verified citations, and programmatic daily digest generation. It natively produces bilingual English/Chinese output, supports nine AI providers (Claude, GPT, Gemini, DeepSeek, Ollama local, etc.), and delivers via Markdown, GitHub Pages, email, Feishu, Slack, Discord, and custom webhooks. An MCP server exposes 12 tools and 7 resources so AI assistants can inspect and run pipeline stages. The review praises its graceful degradation, Pydantic-validated ContentItem core, and deliberate choice of deterministic (non-AI) final rendering for traceability, while honestly noting untested scoring quality, Apify/Telegram and DuckDuckGo single-point risks, and JSON-file-only persistence with no database.
This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/177980399