umerkay gives Mike a web-research arm

umerkay wires Mike up to Olostep, a third-party service that fetches and cleans web pages into tidy text. The result is a research toolkit the assistant can call on: hand it a URL and get readable text back, pose a question and get an answer with its sources listed, or point it at a site to discover and pull every page. Every route sits behind login and checks the address before it reaches out.

It's an early build. The scrape and question-answer paths work today, but the crawl feature only hands back a job ticket - there's no way yet to fetch the finished results, so that piece isn't usable end to end. There's also no rate limiting, caching, or logging on these outbound paid calls, which matters once real matter work runs through them.