Aug — Nov 202538 commits
Genesis
The first browser automation service arrives. The premise is small enough to fit in a command, and large enough to change the relationship between an agent and a web page.
38 documented commits
A browser story, still running
What if the browser could show its work?
ThinkRun began with a stubborn idea: an agent should be able to use a browser—and leave behind a trace a human can actually trust.
thinkrun.ai / session / build-the-browser
$ thinkrun open --profile=your-real-chrome
✓ connected browser context available
THINK A browser task should be inspectable.
THINK The trace should outlive the task.
RECORD clicks, navigation, artifacts, context
$
✓ connected browser context available
THINK A browser task should be inspectable.
THINK The trace should outlive the task.
RECORD clicks, navigation, artifacts, context
$
Scroll to replay
The ledger, not the legend
Every number on this page has a trail.
This story is built from the project’s git history and memory wiki. It deliberately leaves out lifetime cost and token claims: no verified data exists for them.
1,571commits
572unique merged PRs
323days of development
155peak-week commits
01 — The six acts
Not a straight line.
More like a browser with history.
The arc is not “idea → launch.” It is a chain of constraints, recoveries, and sharper questions about what agents need from the web.
Dec 2025 — Jan 2026279 commits
Core engine
Resilience becomes product work: authentication, task recovery, and a video player turn raw browser activity into something people can revisit.
279 documented commits
Feb — Mar 2026520 commits
Extension & local
The browser stops being an abstract remote surface. The Chrome extension, local bridge, and Loom sharing make the work legible where the work actually happens.
520 documented commits
Apr — mid-May 2026280 commits
Sharing & MCP
Recorded work becomes portable: oEmbed, MCP parity, and hybrid upload make a browser session useful beyond the moment it ran.
280 documented commits
Mid-May — Jun 2026306 commits
Recording pipeline
Web recording, the Activity Hub, and billing infrastructure turn traces into a product surface that can be reviewed, shared, and returned to.
306 documented commits
Late Jun — Jul 2026148 commits
Distribution
The final stretch makes the work discoverable: SPA SEO, the connector gateway, and one source of truth for pricing.
148 documented commits
The product is not an agent that clicks.
It is the evidence of what happened after.
02 — The command becomes a trace
A browser
with a memory.
Click a line to replay the product idea. ThinkRun makes browser control available to agents, then turns the resulting session into a trail of actions, media, and context.
SESSION / 001571READY
01 / OPENconnect to real browser context02 / ACTnavigate, inspect, interact03 / RECORDpreserve the trail04 / SHAREsend the evidence, not a recapOpen the browser that already contains the user’s working context. No clean-room simulation necessary.
03 — Open the evidence drawer
Believable is
not enough.
1,571 commits.
Six phases.
The chronology and phase counts are verified in the project timeline notes, including a correction that replaced stale figures with strict git counts.
Strict phase boundaries total exactly 1,571 commits. The timeline notes explicitly mark the earlier 1,570 count as corrected.
docs/thinkrun-timeline-notes.md264 pages
drawn from.
The narrative layer is sourced from PRD threads, daily logs, and indexed project memory—not reconstructed from terse commit titles alone.
60 thread files + 64 daily logs + 140 MEMORY.md entries supplied the interpretive record.
memory/threads · memory/daily · memory/MEMORY.mdNo invented
cost story.
Lifetime token and dollar data is not available for this development history. So it is not dressed up as a growth metric here.
The timeline notes say metering exists only for some sessions from July 2026 onward; no historical lifetime figure is verified.
docs/thinkrun-timeline-notes.md