Never hurt the host app
If in doubt, drop our data, never theirs. No jank, no OOM, no blocking the main thread.
Without compromising a single user's privacy. That sentence is our whole company.
Backstory started in healthcare software, where the tools we tried either crashed our browsers, leaked things they should never have seen, or showed us pixels when we needed answers. We wanted a replay that was correct or honest about not being correct, a recorder that could not hurt the app it was watching, and an AI that cited its evidence.
So we wrote down seven principles and built to them. Never hurt the host app. Privacy is a capture-time property. Correctness over completeness. Degrade in steps, recover with hysteresis. AI cites its evidence. Open standards at the edges. Everything auditable.
The product that came out is a session replay platform where the replay is the mechanism and the why is the product. It records what users saw, including canvas, rebuilds it with DevTools fidelity, detects friction automatically, explains root causes with receipts, and lets a human step in live, one consent at a time.
We run it on the same Helm chart we sell to self-hosted customers, and we record our own website with it.
If in doubt, drop our data, never theirs. No jank, no OOM, no blocking the main thread.
Data that must not be seen must never leave the device. Playback-time masking is not masking.
A replay that silently drifts from reality is worse than a replay with an honest gap marker.
Every subsystem has a ladder of quality levels and a state machine, not a boolean on/off.
An explanation points at the moments it came from, and a claim without a citation is never shown.
OpenTelemetry for correlation and export, W3C Trace Context, S3 API for storage, Helm for delivery.
Who watched what, who controlled whose screen, what the AI concluded and from which evidence.