It watches. It explains. · Chrome extension
The AI screen recorder
that explains what happened.
Record a bug, a workflow, a messy debugging session in Chrome. SceneRecap reads the pixels and the clicks, then writes the bug report with repro steps, the step-by-step guide, or the recap. No narration needed.
Checkout — Payment
Activity0 events
Watching the pixels and the interaction stream. Say nothing — it doesn't need you to.
Not on Chrome? No account?
Try it with your own recording.
Drop any screen recording — QuickTime, OBS, a Loom export — and SceneRecap explains it the same way, no extension needed. Keep the result by creating a free account, or let it expire.
Try it now — no account
≤ 3 min · ≤ 200 MB · deleted after 24 h unless you keep it
Why it’s different
Built for the recordings where nobody talks.
Silent-first
Other tools transcribe your voice. SceneRecap understands the screen itself — sampled frames, every click, every navigation, every console error. Record without saying a word.
Evidence-linked
Every sentence the AI writes carries the exact second it happened and becomes a seek link. No unanchored claims — you can always check.
Nothing typed is ever logged
Keystrokes are never recorded as text, and a password or payment field logs only that it was used. The video is pixels, so you review the clip before anything leaves your machine — or save it to disk instead.
Four modes
One recording.
Four kinds of understanding.
Pick a mode when you record — or re-analyze later in a different one.
Retry after 402 returns 500 and disables the CTA
- 1.Add item, open checkout0:08
- 2.Click “Pay now” → 4020:52
- 3.Retry → 500, button disabled1:03
How it works
From record to explained in about a minute.
Record
One click in Chrome records the tab you are on — with or without your voice. Pick a mode, hit record, do the thing.
SceneRecap watches
The pixels, the clicks, the navigations, the console. Not a transcript — the whole story of the session, timestamped.
Get the explanation
A titled, chaptered, searchable page: narrative, steps or bug report. Share it, export it, or file it as an issue.
Repro Agent · Pro and above
It doesn’t just describe the bug.
It re-runs it and tells you whether it reproduces.
Ask it to check a bug report and the AI reconstructs the flow and the one observable signal that means the bug — a POST /api/cart that returns 500, say. A headless browser re-drives that flow against an origin you control, several times, and the verdict is tied to whether the real signal fired. Not an opinion about the video.
Confirmed, or not
“3/3 runs” means the signal fired every time. A flaky bug shows as ambiguous rather than reproduced — we don’t claim determinism your bug doesn’t have.
“Couldn’t reproduce” is the point
We say it out loud, and why — usually a state or environment difference. A tool that only ever confirms is a tool nobody believes.
Blocked is an answer too
A login wall stops the run and says so. We never enter credentials; hand it a staging session instead.
You trigger it per report and each run is metered — it is not something that quietly runs against every bug you file. Opt in further and it can strip the flow to its minimal repro, or try an alternate path to show which branch is actually broken.
How the Repro Agent works, and what it can’t drive →
Reproduction runs the recorded flow headlessly against an origin you control; results reflect that environment, and we report when it can’t reproduce.
Takes · Pro and above
Record once.
The agent does the retakes.
Nobody demos cleanly on the first try. Demo it once — messily — and ask for a take: the agent re-performs your recording’s flow against a site you control and records a fresh, watchable version. Steady pacing, no fumbling, no wrong turns. Say “again, but through the settings page” and it performs the variation instead.
From your recording, never a prompt
Every take is planned from a real recording's interaction trail. The agent performs a flow a human already did — it never invents one from a description.
A draft until you approve it
You watch the take before it becomes a recording. Approve it into the normal pipeline, or discard it and the video is deleted. Nothing is shared or analysed before you say so.
Labelled as performed, forever
Approved takes carry a “performed by agent” badge on the player, share pages, embeds and every export — on every plan. Machine output never passes as human capture.
Shipped a fix? Re-record the repro against your fixed build and get a short video of the flow now passing — evidence to close the issue with. Takes run on origins you control, share one monthly agent-run allowance with the Repro Agent, and never enter credentials — a login wall ends the take honestly, as blocked.
Watch a real take — badge and all →How Takes work, and what they refuse to do →
A take is performed from your own recording’s real interaction trail against an origin you control, stays a private draft until you approve it, and is permanently labelled as performed by the agent.
After the record button
The recorder is the easy part.
Search that understands
Hybrid full-text + semantic search across summaries and timeline; results deep-link to the second.
Share, five ways
Public, unlisted, password, workspace-only, or specific emails. Comments at timestamps.
Exports
Markdown, HTML, PDF, DOCX, GIF highlight — and copy-as for GitHub, Jira, Linear, Slack.
One-click issues
File to GitHub, Linear or Jira with the AI-written body. Push tutorials to Notion.
AI voiceover
Turn a silent Explain or Tutorial into a narrated video in one of three voices.
API & webhooks
Scoped keys, HMAC-signed webhooks for recording.ready, recording.shared, comment.created.
In one paragraph
What SceneRecap is
SceneRecap is an AI screen recorder for Chrome. It records a tab, watches the pixels and the interaction events — clicks, navigations, console errors, failed requests — and writes the explanation: a bug report with repro steps and evidence, a step-by-step guide with screenshots, a plain explanation, or a recap.
It works for silent recordings, links every claim to its second in the video, never logs what you type, files issues to GitHub, Linear and Jira, exports to PDF, DOCX and Notion, and is hosted in the EU. On Pro, its Repro Agent re-drives a bug report headlessly against an origin you control and reports whether the bug actually reproduces. Takes let the agent re-perform a recorded flow and record the clean version — reviewed by you before it is saved, and permanently labelled as agent-performed. It is an alternative to Loom, Scribe, Tango and Jam when the point of the recording is what happened on screen.
FAQ
Questions people ask.
What is SceneRecap?
SceneRecap is an AI screen recorder for Chrome. You record a tab — with or without talking — and the AI watches the pixels and the interaction events (clicks, navigations, console errors, network failures) and writes the explanation: a bug report with reproduction steps, a step-by-step tutorial with screenshots, a plain-language explanation, or a recap of decisions and action items.
Do I have to narrate my screen recording?
No. SceneRecap is built for silent recordings. It reads the screen and the interaction log directly, so it does not depend on a transcript. If you do speak, the audio is transcribed and used as extra context.
How is this different from Loom, Scribe or Jam?
Loom is async video with transcription; Scribe captures click-through guides; Jam captures bug context. SceneRecap records the whole session and then explains it in four modes — bug report, tutorial, explain, recap — with every claim timestamped to the video, and it exports to GitHub, Linear, Jira, Notion, PDF and DOCX. And on Pro, the Repro Agent re-runs a bug report's flow headlessly against an origin you control and reports whether it actually reproduces. See the comparison pages for details.
Can SceneRecap turn a screen recording into a bug report?
Yes. Bug Report mode produces numbered reproduction steps, expected vs actual behaviour, the first bad moment, and console/network/visual evidence, each linked to a second in the video — plus a paste-ready issue body you can file to GitHub, Linear or Jira in one click.
Can it create step-by-step guides or SOP documentation?
Yes. Tutorial mode writes an ordered guide with a screenshot per step and exports to PDF, DOCX, Markdown, HTML or a Notion page.
What about passwords and sensitive data?
Keystrokes are never logged as text: when you type into a password or payment field we record only that a sensitive field was used, never its value. URLs are stored path-only, with query strings dropped. The recording itself is pixels — we do not alter what was on your screen — so you review the clip before anything uploads, and nothing leaves your machine until you do. Data is hosted in the EU (Google Cloud, Belgium/Frankfurt).
Is there a free plan?
Yes. The Free plan includes 1 free recording of up to 1 minute — enough to see the AI write a real report from your own screen, no credit card. Pro adds unlimited recordings, all four modes, PDF/DOCX exports, integrations, API and AI voiceover.
Can my own users report bugs on my site?
Yes — that is the reporting widget, a $39/month add-on. You paste one script tag; visitors click “Report a problem”, record what went wrong, and watch it back before anything is sent. The report arrives in your Inbox already written, with the JavaScript error and the failed request that caused it attached, because the widget is on the page when things break. Reports about the same bug on the same page are grouped, so five visitors hitting one problem is one row marked ×5 rather than five videos. It works on phones too, by asking for a recording from the device’s own screen recorder. See the reporting widget page for how it installs and what it collects.
Does it work without the Chrome extension?
Yes, two ways. You can record a tab directly in the web app (Dashboard → Record) with no install at all, or upload a recording you already have — QuickTime, OBS, a Loom export, a clip from your phone — and it goes through the same pipeline. The extension is still the richest path, because only it captures the click and console trail the AI uses as evidence.
Stop explaining your screen recordings.
Let them explain themselves.