Playground.

A small tool-using research agent built on Google’s Gemini 2.5 Flash and the Vercel AI SDK. Give it any public URL and a focus question. It fetches the page, optionally distills it through a summarize step, and returns three findings with a two-sentence verdict.

It’s the same pattern I use day-to-day at Tallysight: bounded agentic loops with explicit tool calls, an SSRF-guarded fetch, and rate-limiting at the edge. More on how I think about agentic systems in writing.