UX analytics for teams that ship with agents
Watch the session. Trust the event.
Replay shows where people hesitate. A shared catalog keeps engineers, agents, and the dashboard using the same words for checkout, search, and pay.
In production
FastTrack BKK
Catalog
Shared with agents
Replay
Before React hydrates
Privacy
Consent-aware capture
Built for the moment an agent writes your instrumentation
Most analytics tools assume a human typed every event name. FastTrack Analytics assumes the next pull request was drafted by a model — and that you still need the recording, the funnel, and a name you can defend in a standup.
Replay that explains the stall
Watch the session, not a chart of averages. Rage clicks, dead ends, and funnel steps sit on the same timeline so a product change has a reason.
A catalog agents cannot invent around
LESP is the dictionary. Name checkout once. Humans approve it. Models reuse it. Unknown semantic events are refused — they do not quietly poison next month’s funnel.
Instrumentation from the editor you already use
Connect MCP, generate the snippet, embed the SDK. You do not need a separate analytics IDE to keep names honest.
Why LESP exists
Give agents a dictionary — or they will invent one
LESP (LLM Event Schema Protocol) is the shared catalog of product events for a site. You name checkout once, in language a person and a model can both read. Every later track() call, replay marker, and funnel step uses that name. If an agent invents CheckoutStart next week, ingest refuses it.
- 01
Name it once
Register the event in Event Studio with a short description of the moment that matters — “customer started payment,” not a leftover CSS class.
- 02
Reuse it everywhere
The SDK, data-insight-id tags, and any MCP client read the same glossary. The next agent session does not get a blank page and a guess.
- 03
Refuse the drift
Strict mode drops unknown semantic events. Behavioral noise (page views, rage clicks) still captures automatically. Product vocabulary stays intentional.
Replay, funnels, and the standup use the same words. You stop arguing whether checkout_started and CheckoutStart are the same conversion.
- checkout_startedactivev3
- booking_completedactivev2
- payment_faileddraftv1
- rage_clickauto—
- page.viewauto—
Event Studio — humans approve what agents propose
Already in production
FastTrack BKK
Live on a real booking site. Replay showed where people stalled; the catalog kept funnel names from drifting after each agent-assisted change.
Agencies & brands
Growth and Scale are for more than one property: compare sites, keep seats bounded, and export what a client actually asked for.
Any MCP client
Cursor, VS Code, Claude Code, Devin — the agent reads the same glossary it writes. You review Event Studio. Nothing ships as an invented name.
From a named event to a fix
Four steps. No second schema for the agent, no rented vocabulary for the dashboard.
01
Agree the events
Catalog in Event Studio
02
Capture the session
SDK + early recorder
03
Ingest with rules
Unknown names are refused
04
Watch and fix
Replay, funnels, alerts
See the next abandoned checkout
Open the dashboard to create a site, or compare plans first. Self-serve checkout is not live yet — we will not pretend it is.