01UX research
Find the silent dead ends.
Spot the hidden consent box or the disabled button that leaves a new user stuck.
/sign-upCreate account stays disabled
Session replay, built for agents.
Every user session
tells a story
Ask your agent to review captured sign-up, onboarding, and other key journeys on a schedule. Get findings backed by the moments that explain them.
Illustrated session paths
Search across sessions ↳ Review the moments
01 A daily practice
Put product insights
on autopilot.
Schedule a sign-up review. Let your agent bring the friction, the evidence, and the next step into your team’s channel.
Your scheduleDelivered to your team
onboarding-watchExample message
02 Questions worth repeating
Turn your agent into
a 24/7 analyst.
On your schedule. Focused on the journeys that matter.
02Error investigation
Explain the failed sign-up.
Connect a failed request to what the user saw and what they tried next.
/api/accountsAccount request · 5xx
03Release follow-up
Catch the new friction.
Compare onboarding journeys before and after a release. Investigate what changed.
/onboardingBefore release → After release
03 Attention has a budget
Spend tokens
where they count.
Your agent starts with a digest, then pulls the details where it matters.
Explore the map of attentionSame captured sessionSame time range
Full detail950,931 tokens
All signal kinds + errors
A selective digest14,045 tokens
Navigation, interactions + errors
Observed in a separate debugger session. Same time range; different kinds and grain.
Put it to work
Start with
tomorrow’s sign-ups.
Give your agent the prompt. It’ll help you connect Subtext, review a journey, and set up a daily routine.
An instruction for your agent
“Review sign-up journeys each day. Tell me where new users got stuck, what the evidence supports, and what to investigate next.”
Read the full instruction
Help me get started with Subtext, review sign-up and onboarding journeys, and set up a daily routine. First check whether the Subtext MCP tools are available and authenticated. If they are missing or not connected: - Read https://subtext.fullstory.com/llms.txt and https://subtext.fullstory.com/docs for current setup instructions. The documentation index is at https://subtext.fullstory.com/docs/llms.txt. - Help me sign up at https://app.fullstory.com/subtext/signup if needed, then run npx @subtextdev/subtext-wizard from my project root to install capture and connect the MCP server. Follow the current documentation; guide me through any interactive sign-in or setup steps that need my input. - Verify the connection before calling review tools. If there are no captured sessions yet, help me verify capture and collect a session first. Once connected, confirm the app, journey route (default: /sign-up), and review budget, then run an initial review using the workflow below. Before scheduling repeat runs, confirm the schedule, timezone, and reporting destination with me. Use the scheduling features in my agent environment; if scheduling is unavailable, give me a reusable manual workflow. Retain each completed brief for comparison. On each run: 1. Use review-search with since: "24h" and a navigate URL containing the agreed journey route. Cap the search at 10 returned sessions. This is a bounded selection, not an exhaustive audit or a random sample. 2. Triage the candidates with review-summary. Open the relevant sessions with review-open and read their maps before choosing detail. 3. Investigate friction with review-zoom. Choose the time window, signal kinds, and grain for the question. At ambiguous moments, use review-snapshot for the image, tree, and boxes; use our Sightmap definitions when available. 4. Separate visual obstruction, validation, and network failures. A matching route alone does not establish a problem or its cause. Close each opened session when finished. 5. Compare supported findings with the previous brief. Deduplicate known findings by cause and component. Include session references, timestamps, the supporting evidence, uncertainties, and the next useful action. 6. Report the time window and number of sessions searched, returned, and actually reviewed when available. State any limit reached or missing capture. Do not infer a site-wide error rate, causality, or that a fix is complete from this selection. If nothing new is found, say only that nothing new was found in the reviewed sessions.