The browser becomes a tool
The early work made browser control practical: sessions, actions, and the repeated loop of inspecting a page before acting on it.
Selected from the development archiveThinkRun began with a stubborn idea: an agent should not have to guess what already happened in a browser.
A screenshot is a postcard from a crime scene. A prompt is a witness statement written after the fact. Both leave an agent to reconstruct the moment from fragments.
By the time the agent arrives, the click, the page state, and the useful evidence have all moved on.
ThinkRun captures the moment while it exists, then turns it into something an agent can inspect: a time-linked trail of what happened and why it mattered.
ThinkRun was not sketched into existence in a deck. It was assembled in the work of building, breaking, reviewing, and correcting a browser tool until it could carry context forward.
Record snapshot: repository memory wiki, July 9, 2026. These are history counts, not marketing projections.
The product’s history is full of the same discipline it offers users: save the evidence, name what failed, then make the next attempt less blind.
The early work made browser control practical: sessions, actions, and the repeated loop of inspecting a page before acting on it.
Selected from the development archiveThinkRun gives an agent a better starting point: not a vague report of a browser moment, but the moment itself—captured, organized, and ready to reason from.
issue: checkout request failed · evidence: click → state → response → error · next step: inspect the captured context