A daily automated AI ecosystem aggregation and digest tool. Crawls 10+ data sources including GitHub, HN, ArXiv, and HuggingFace, generates bilingual (CN/EN) reports via LLM analysis, and distributes through GitHub Issues, Web UI, RSS, and MCP Server.
agents-radar is a daily information aggregation tool for AI practitioners. It runs on GitHub Actions on a schedule, automatically crawling AI signals from 10 data sources, analyzing and classifying them via LLM, and generating bilingual (CN/EN) daily reports.
Data sources cover GitHub repository activity (Issues/PR/Releases of 17+ CLI tools, Trending repos), Hacker News Top 30 AI stories, ArXiv papers (cs.AI/CL/LG), Hugging Face Trending models, Product Hunt AI products, Dev.to and Lobste.rs community content, Claude Code Skills, and Anthropic/OpenAI official site updates.
The LLM analysis layer supports four backends — Anthropic (default claude-sonnet-4-6), OpenAI, GitHub Copilot, and OpenRouter — with an extensible Provider abstraction layer in src/providers/. Trending repos are classified by dimension (Infrastructure / Agent / Application / Model / RAG) with trend signal extraction.
Reports are split into 9 independent topic-based files (CLI tools, Agent ecosystem, Web updates, Trending, HN, Product Hunt, ArXiv, HuggingFace, Community), each with both CN and EN versions.
Distribution channels include GitHub Issues (with labels), Markdown file commits, a dark-themed GitHub Pages Web UI, RSS Feed (latest 30 reports), Telegram Bot notifications, Feishu Webhook card messages, and a Cloudflare Workers-based MCP Server exposing four tools (list_reports, get_latest, get_report, search) compatible with Claude Desktop and OpenClaw.
The recommended deployment is to Fork and configure GitHub Actions Secrets. Local runs via pnpm start are also supported. All tracked repositories are configured through config.yml without code changes. Web content uses web-state.json for incremental detection.
Unconfirmed: No license file found in README or repo root; daily LLM token cost not quantified; weekly/monthly report aggregation strategy not detailed.