Session replay captures what happened. A video scanner makes selected behaviors queryable: you define an observable pattern, run it against a recording, and receive typed observations tied to exact evidence ranges.
The scanner workflow
- Create a saved
monitor,classifier, orfind_momentsscanner. - Export an authorized replay recording and complete a private upload.
- Start one run with its ready
upload_idand attach metadata such as workspace, cohort, or user ID. - Poll the run, then retrieve its observations or query observations across runs.
POST /v1/scanners/{scanner_id}/runs
{
"upload_id": "upl_YOUR_READY_UPLOAD",
"metadata": {"workspace": "acme", "cohort": "new-user"}
}Write for what the source can support
For replay video, describe the event you want: “Find moments where the user repeats the same setup action, receives a visible error, or leaves without advancing.” If you submit audio-only research, ask about spoken statements instead; an audio answer cannot prove a visible interface state.
A supported monitor no-match is one explicit full-source boolean false observation after completed review. A completed zero-observation run is unanswerable, not false and not a processing failure.
Treat observations as evidence records
Every observation names its semantics and value type, carries a relevance score, and points back to its run, scanner version, analysis, source metadata, and timestamped evidence. Relevance is a ranking signal—not calibrated statistical confidence.
How this relates to replay-native AI
Replay-native products can search recordings already inside their analytics system and may add cohorts, alerts, and continuous scanning. ChatClipThat is a general recorded-video intelligence API: you privately upload the authorized video and control the surrounding metadata. Scanner runs return JSON; video find_moments results may be explicitly rendered. If you need continuous ingestion today, orchestrate exports, uploads, and run creation in your own service.