Session Replay

When the agent fails at step 47, see step 47.

Every session is recorded as video, DOM diff stream, HAR network log, and console output — all stitched to the agent's step ID. Scrub to the moment things broke and reproduce it.

The replay viewer

A video scrubber your engineers will actually use.

Open any session in the dashboard and drag through it. The DOM tree, network waterfall, and console log update in lockstep with the video. Hover a node, see the agent's reasoning for the click that touched it.

Share a link with your support team. Share it with the customer. Share it with your auditor.

[ replay viewer — video + synced DOM tree + HAR ]
[ deterministic replay — original vs rerun diff ]
Deterministic replay

Rerun the exact session. Hit the exact bug.

Replay mode fixes the network responses, the clock, the random seed, and the fingerprint to what they were at recording time. The same agent, on the same recording, produces the same steps — so you can bisect a regression in minutes, not days.

Replay mode in the docs →
What we capture

Observability, not print statements.

Video at 8 fps

Lossless WebM recording of every pixel the agent saw, compressed to about 120 KB per minute.

Full DOM diff stream

Not snapshots — an event log of every mutation. Scrub to any timestamp and rehydrate the exact DOM.

HAR network log

Every request, response body, and header. Redacted for secrets, pinned for reruns.

Step-level timeline

Agent reasoning, tool calls, and prompt IDs plotted on the same timeline as the browser events.

Console & errors

Every console line and page error captured and indexed. Grep across a million sessions.

Retention you control

7 days on free, 90 on Team, 7 years on Enterprise. Export to your own S3 at any time.

From a failed run to a fixed bug in four steps.

Page fires

Your alerting catches a failed agent run and links to the replay by session ID.

Scrub the video

Find the moment the DOM diverged from the agent's plan.

Rerun deterministically

Attach a candidate fix and rerun against the pinned recording.

Ship the fix

If the replay passes, the regression is fixed. Add the case to your test suite.

Plug into your stack

Where your team already looks for answers.

Datadog

Replay link attached to every agent span.

Sentry

Every agent error auto-links to the session video.

Slack

Paste a replay URL and it unfurls with a scrubbable preview.

S3 export

Mirror replays into your own bucket in your own format.

The first time a failure happened in production and we had the replay open in 30 seconds, I knew we were never going back.

— Marcus Hale, Head of Platform, Operator Co

Privacy by default

Record everything. Leak nothing.

Vaulted credentials are masked at capture time. Custom redaction rules let you block specific selectors, URL patterns, or header values across all sessions.

  • Vaulted values never appear in video or DOM diff
  • Regex-based custom redaction for PII and PHI
  • Replay access tied to SSO roles
  • Region-pinned storage — replays never leave your data residency boundary
  • BYO-KMS encryption on Enterprise

Replay FAQ

How much storage does a replay take?

Typical 10-minute session: ~1.5 MB compressed. Included in your session-minute price.

Can I download the raw artifacts?

Yes — video (WebM), HAR, DOM diff stream (NDJSON), and console log are all downloadable from the dashboard or API.

Does replay work with my custom LLM?

Yes. The replay captures what happened in the browser. Your agent's reasoning timeline is attached via a simple SDK hook.

How long can I keep replays?

Up to 7 years on Enterprise. Configurable per-account and per-session.

Stop guessing why the agent broke.

Every session on BrowserAnvil is recorded. It is not a paid add-on. It is the default.

Start free See a live demo