T ThinkRun / origin story
Read the build log ↓
A build log, told in public

The browser
became a teammate.

ThinkRun began with a simple belief: agents should not merely ask browsers for answers. They should be able to work inside them — visibly, safely, and alongside the people who care about the result.

Scroll to inspect
BUILT IN PUBLIC
2025—2026
01 / The proposition

Browsers are where work actually happens.

The interesting problem was never “can an AI click a button?” It was making every click legible: a task, a plan, an action, a result — and a human who can see the whole chain.

thinkrun › research the market

PLAN  Open sources and compare claims
NAVIGATE  3 sources opened
THINK  “Need a primary source here.”
EXTRACT  Findings structured for review

✓ Session visible. Work inspectable.
1,571commits
572unique merged PRs
323days of development
155commits in peak week
02 / Six chapters

Not a straight line.
A useful one.

The development record is unusually complete: the decisions, reversals, outages, reviews, and release receipts survived alongside the code.

AUG—NOV 2025

Genesis

The first browser automation service established the core: a browser that could be operated through a simple agent interface.

BROWSER AUTOMATIONPR #1
DEC 2025—JAN 2026

Core engine

Reliability became product work. Authentication, resilient tasks, and a video player made the system more accountable when real work was on the line.

CLEARAUTHTASK RESILIENCEVIDEO PLAYER
FEB—MAR 2026

Extension & local

The browser could now meet people where they already worked: the Chrome extension, local bridge, and shareable sessions brought the agent out of the server room.

CHROME EXTENSIONLOCAL BRIDGELOOM-STYLE SHARING
APR—MID-MAY 2026

Sharing & MCP

A run became something a team could pass around. oEmbed, MCP parity, and hybrid upload turned execution into a portable artifact.

OEMBED@THINKRUN/MCPHYBRID UPLOAD
MID-MAY—JUN 2026

Recording pipeline

Work needed a replay. Web recording, the Activity Hub, and subscription billing made the operating layer visible and sustainable.

WEB RECORDINGACTIVITY HUBSUBSCRIPTIONS
LATE JUN—JUL 2026

Distribution

The work became easier to find and adopt: SPA SEO, a connector gateway, and a single source of truth for pricing prepared ThinkRun for the open web.

SPA SEOCONNECTOR GATEWAYPRICING SSOT
03 / The unusual part

We kept the receipts.

ThinkRun’s memory wiki is not a victory lap. It holds the hard parts too: auth outages, billing pivots, deployment regressions, and the reasoning behind the fixes. That is what lets progress compound instead of merely accumulate.

development recordVERIFIED
PRD thread pages62
first commit2025-08-20
peak build week2026-W12
principleVERIFY THE ARTIFACT
04 / A different kind of browser automation

Don’t just get an answer.
See the work.

Tap any event in this tiny session to follow the thread. ThinkRun treats a browser run as a readable record, not a black box.

T
thinkrun.ai / share / market-research
LIVE SESSION REPLAY

Sources, not guesses.

A research run starts with the web — and leaves behind enough context for someone else to decide what to trust.

OPENED 3 SOURCES
EXECUTION LOG / 04 EVENTS
The next useful step

What if your browser could show its work?

ThinkRun is building the operating layer for agents on the web: capable enough to act, transparent enough to trust, and collaborative enough to share.

Meet ThinkRun
FACT CHECK: all numbers on this page come from docs/thinkrun-timeline-notes.md, refreshed against raw git on 2026-07-09. No lifetime token or cost statistics are shown because the repository does not have verified data for them.