jmclark-lab gives Mike a live line to the open web

The motivation is grounding. Regulatory questions, LATAM filing timelines, and current legal-pathway queries all turn on facts that go stale fast, and a model answering from memory quietly gets them wrong. jmclark-lab hooked Mike into SerpApi, a service that returns live search results, so before Mike answers it fetches current hits and folds them into the prompt.

The build is cautious in the right places. Search runs on a short timeout and never blocks the conversation, so a slow or missing response just degrades gracefully. And with no key configured, Mike behaves exactly as before, which keeps the feature opt-in per deployment. The notable turn came at the end: jmclark-lab first limited live search to messages that looked like regulatory queries, then scrapped that gate and turned it on for every question. That widens fact-checking coverage across the board, at the cost of a search call, plus its latency and spend, on essentially all traffic.