Agent Reach: giving an agent read access to Reddit, X and YouTube
Open source CLI (MIT, Python 3.10 and up) that reads nothing itself: it picks, installs and health-checks the tools that let an agent read the web and social platforms, and the agent then calls those tools directly. Every platform has an ordered list of backends, the first one that works gets used, and `agent-reach doctor` tells you which one is live and which one broke. With zero setup: web pages (Jina Reader), YouTube transcripts (yt-dlp), RSS feeds, public GitHub repos, web search (Exa). With your own browser session or an exported cookie: X, Reddit, LinkedIn, Instagram, Facebook. The rest of the catalogue mostly targets Chinese platforms (Bilibili, Xiaohongshu, V2EX, Xueqiu).
Strengths
- YouTube transcripts, RSS, web pages and public GitHub usable with no key and no setup
- A backend that breaks gets replaced on the project side, and `agent-reach doctor` shows which one is live
- The agent calls the underlying tools directly: no wrapper filtering or rewording what it reads
Limitations
- Reddit, X and the other social platforms are only readable with your own accounts’ session or cookies
- Installed through a remote instruction file your agent executes: reread it before every update
- A large share of the channels targets Chinese platforms of no use for Western monitoring
Best for
- Monitoring that has to read Reddit or X threads, not just blogs and announcements
- Having an agent summarise a YouTube video from its transcript, with no paid service
- Giving a fresh agent environment web access in one install, then checking what works