This is the story of a tool built so agents could see the web—then rebuilt so humans could see what their agents saw.
Software agents had hands. They needed eyes, memory, and a receipt.
ThinkRun began as browser automation infrastructure: a service, a queue, a remote session. But every failure revealed the same deeper problem. A browser action without context is just an event. A recording without analysis is just footage. The product became the connective tissue between action and evidence.
How do you give an agent browser access that is useful, inspectable, resilient, and safe? Drag the development reel.
The first service learned to create sessions, take instructions, and survive deployment.
Authentication, retries, artifacts, real-time task plans—the machine grew a nervous system.
The agent crossed from a remote browser into the tab already open on your machine.
Sessions became objects people and agents could exchange, embed, inspect, and continue.
The screen recorder moved into the product. A gesture became a bug report an agent could use.
CLI, MCP, remote grants, connectors, and onboarding turned infrastructure into an invitation.
March 16–22, 2026 was the peak. Not a launch spike—a systems spike. Extension, local bridge, sharing, recovery. The week where ThinkRun stopped being a remote browser and started becoming a way of working.
An early timeline claimed six estimates were verified—and invented a plausible caveat to explain the math. The memory wiki challenged it. Git recomputed it. The record was corrected to 1,571.
Record what happened. Give it context. Hand the evidence to a person or an agent—and keep moving.
▶ Start with ThinkRun