An origin story with receipts · 2025—2026

The evidence
has a memory.

ThinkRun began with a stubborn idea: when software breaks in the real world, the person who saw it should not have to become its historian.

Open the flight recorder
01 — THE PREMISE
The bug report was never supposed to be the second job.Someone already watched the failure happen. ThinkRun was built to preserve that moment—in enough detail for a person or an agent to do something useful with it.
02 / the moment

A dead domain. A real inbox. A two-minute handoff.

One of ThinkRun’s own production bugs made the case better than any pitch could: a verification link still pointed to an old domain. Code review could not see it. A recording could.

Recorded browser sessionevidence / 01 of 03

The email arrives.

A verification message is opened in a real inbox. The bug is not in an abstract test case—it is about to happen in front of a user.

00:03

Tap through the evidence

The recorded artifact later carried 12 screenshots, narration, observations, and a fix direction for the agent that read it.

03 / the handoff

Not a video file. A handoff any agent can read.

A recording becomes a portable account of the event. The same share token resolves to a human-readable page and structured formats for an agent.

session / verification-link / complete

One event.
Four ways to understand it.

The evidence is not trapped inside pixels. A recipient can inspect the replay, while an agent can start from structured context instead of asking a human to translate what they saw.

$ agent review session.md
✓ observation: verification link points to retired domain
✓ likely class: production configuration / DNS
→ next: trace verification-email origin and re-run the real flow
04 / the build log

The development story is a receipt, too.

ThinkRun’s own record is deliberately auditable: raw Git history for what changed, a memory wiki for why, and a correction trail when the numbers disagree.

First commit → story snapshot

323

days spent turning browser activity into something an agent can understand, act on, and hand back for review.

2025-08-20 → 2026-07-09
Computed from raw Git history

Git history

1,571

commits in the verified snapshot.

Raw Git log

Merged work

572

unique merged PR numbers in commit subjects.

Git merge subjects

Peak week

155

commits in the week of March 16.

Timeline audit

“No mythology.
Just receipts.”

That is not copywriting. It is the operating principle: a product for preserving real-world context should be able to show its own work.

Snapshot verified

05 / the correction

Then the timeline failed its own review.

An early version of this story had plausible numbers that did not add up. Independent recomputation caught the drift.

That correction belongs in the story. ThinkRun is built around the idea that a claim is only as useful as the evidence behind it.

MetricEarlierRecomputed
Total commits1,5701,571 ✓
PRD threads5962 ✓
Phase 1 PRs~254 ✓
Phase 3 PRs~350147 ✓
Why show the miss? Because “verified” should mean a number can survive being checked—not merely sound credible in a headline.

The next chapter is yours

Stop explaining what the agent could have seen.

Record the moment once. Let the evidence carry the screen, voice, clicks, console, and network forward—so your agent can understand before it acts.

See what ThinkRun sees