Watch only the sessions worth it, from the second it went wrong.
Sequential judges every session against the questions you set for your app. Your agent does the rest.
↻ and again, with every release
The demo
A 54-second session from a fictional shop, recorded through posthog-js. The verdict, the moments and the narrative were in the sidebar before the first frame played — click a moment to jump there and check the work.
Watch · Ask · Verdict
Every session is judged as it ends. What you do with that is up to you:
Every judged session, ranked by what deserves attention — and a lane for the ones too close to call.
Describe what you're looking for in plain words; the matches come back ranked.
“someone who couldn't finish paying” 3 of 10 judged sessions came back ranked — the blocked checkout first, opened at the moment it gave up.
Every replay opens with the verdict in the sidebar and a heat lane over the scrubber.
Your rubric
Every project starts with twelve questions — intent, outcome, friction, where they struggled. Rewrite them, switch them off, or add your own, up to 24: “Did they send an invite?”, “Where did the plan change stall?” Try a draft on a real session before you save. Changes apply to sessions judged from then on; earlier verdicts keep the questions they were judged with.
Your agent reads the same rubric, so its answers use your words.
With your agent
Sequential groups what goes wrong into findings, each with the sessions that prove it. Your agent reads them, starts the change wherever your team works, and checks the result after the release.
Example
Let your agent work the list every morning. Use an agent that runs on a schedule. It reads new findings, files the work, opens the fix and checks last week's fixes. You review what it proposes, and Sequential keeps the record.
failed request Checkout breaks
your part: 1 review Everything else ran on its own.
error after release A release breaks search
your part: 1 approval Everything else ran on its own.
dead clicks People click what isn't a button
your part: the design itself Everything else ran on its own.
missing data A question the data can't answer
your part: asking the question Everything else ran on its own.
Prefer to stay in the loop? The same agent answers .
Example
What's stopping people at checkout this week?
from Sequential38 people got stuck at the shipping address: the postcode field rejects a space. Three sessions show it.
File it for the web team, with those sessions.
your agent · in JiraCreated WEB-231 with the three sessions, and linked it to the finding.
Did the fix work?
from SequentialSince release 2026.10.09: 41% → 12% of sessions on the shipping page hit it.
Works with: Claude Code · Cursor · Codex · Claude Desktop · any MCP client
Or start from a session in Sequential: Ask your agent opens Claude Desktop, Claude Code or Cursor with a prompt about it already filled in. You review it and send it.
The same finding sits in Sequential for your whole team, with the ticket and the result on it.
The verdict
Every recording is turned into a short digest of meaningful events and put to a judgment model — not a chatbot. It answers the same twelve questions for every session, each with a probability. Sequential turns those into a verdict, the moments that matter, and a heat lane on the timeline.
E001| 0:06 click "Trail Pack 38 L" ⋯ 18 more eventsE020| 0:32 click "element"E021| 0:34 click "Apply code"E022| 0:34 custom coupon_rejected — code=WELCOME20 · reason=not_applicableE023| 0:35 rage_click "Apply code" ×4E024| 0:35 custom coupon_rejected — code= · reason=not_applicableE025| 0:35 custom coupon_rejected — code= · reason=not_applicable
What Sequential reads: numbered meaningful events, pauses spelled out, no raw DOM, no user id.
Each answer is a probability over fixed options, the same for every session, so sessions compare. A moment question also asks where it happened, so the request holds fifteen answers. Every project can edit these questions.
The narrative is assembled in code from those numbers, hedged by confidence. Answers are fixed choices, not free text.
The struggle distribution becomes the heat lane above the scrubber; the key moments become jump points. Open the Insights tab in the player to see the same verdict where the work happens.
Product analytics
Every event your SDK already sends, classified at ingest into nine classes — so each day's bar reads as what people did, not a wall of event names — and any bar breaks down by page, country or browser.
| Page | User actions | Unique users |
|---|---|---|
| /shop | 122,653 | 5,708 |
| / | 120,432 | 4,893 |
| /checkout | 64,765 | 2,700 |
| /product/trail-pack | 59,090 | 2,443 |
| /cart | 49,257 | 2,038 |
| /product/ridge-tent | 42,402 | 1,803 |
| /help | 37,984 | 1,626 |
| /journal | 37,257 | 1,654 |
posthog.init('YOUR_KEY', { api_host: 'https://api.usesequential.com' });<script src="https://api.usesequential.com/rec/recorder.js/YOUR_REPLAY_KEY"></script>
Change two values in the posthog.init call you have. The second line adds session replay. The recorder reads the SDK’s session id. Setup guide
Sequential is free while it is in preview. You need no card. Each organisation has a monthly session allowance, and we email you at 80 % and at 100 %.
We keep your events and replays for 7 days.
We change features often while we build. Tell us what you need.
Paste this into your coding agent, for example Claude Code or Cursor.
Set up Sequential (session analytics and replay) in this app with a free trial. Ask me for my email address. Then send POST https://api.usesequential.com/api/agent/provision with the JSON body {"owner_email": "<my email>", "platform": "web"} (or "ios", "android" or "react-native", to match this app). Add the answer's install.snippet to the app, run the app on localhost, and keep mcp.url and mcp.token to read the sessions. Do not put the token in source control. Everything you need is in the answer. Then tell me to claim the trial from the email Sequential sends me.