Analysis Outputs
Build your own frontend experience from Nandemo's raw analysis outputs: source names, scene understanding, transcript words, people, speakers, metadata, and search results.
List Videos
This is the low-level footage inventory. Use returned names as the source identifiers in later calls.
A source can exist before every analysis job has completed, so list results should be treated as inventory plus status. Use this endpoint to populate source pickers, check whether a file is known to the project, and decide which source name to pass into source-scoped endpoints.
curl "$BASE_URL/projects/$PID/videos?q=interview" \-H "x-api-key: $API_KEY"
Scene Understanding
Use project understanding for timeline browsing, shot search UIs, scene cards, or timeline-level summaries.
Scene understanding is Zonic's visual timeline model. For each video source, Nandemo separates the file into multiple scenes, gives each scene a start and end time, and attaches visual descriptions or structured observations to that time range. The result is an ordered scene list that represents the whole video as smaller, searchable moments.
Applications usually render scene understanding as cards, timeline blocks, or search results. Dekiru uses the same scene records as evidence when it answers questions or chooses clips for a sequence.
curl "$BASE_URL/projects/$PID/understanding" \-H "x-api-key: $API_KEY"
Transcript And Words
Word-level timestamps are useful when your UI needs precise in/out points. Segment merge, split, and speaker assignment endpoints can support transcript correction workflows.
Transcript outputs describe what was spoken and when it was spoken. Segment-level transcript is easier to read, while word-level timing is better for accurate subtitle alignment, quote selection, jump-to-word interactions, and edit decisions that need exact boundaries.
Videos without speech may have no transcript output. In that case, visual scene understanding and metadata can still be valid Nandemo results.
curl "$BASE_URL/projects/$PID/videos/$VID/words?start=0&end=30" \-H "x-api-key: $API_KEY"
People And Speakers
The people registry returns detected face clusters and their appearances. Speaker cluster rename and merge endpoints are also available for diarized transcript cleanup.
People and speakers are related but not identical. Face detection groups visual appearances across footage. Speaker diarization groups voices in transcript context. A product can expose rename or merge controls when users need clean labels, but read-only integrations can still use the clusters to show where a person appears or speaks.
curl "$BASE_URL/projects/$PID/people" \-H "x-api-key: $API_KEY"
Search
REST search is direct and predictable: ask for visual scenes, spoken script, labels, or all three. Use Dekiru QA when you want an agent to choose tools and combine evidence.
Search uses the indexes built from Nandemo outputs. Visual search is based on scene understanding, script search is based on transcript text, and label search is exact metadata filtering. The response should be treated as candidate moments or videos, not as a final editorial decision.
curl -X POST "$BASE_URL/projects/$PID/videos/search" \"${AUTH[@]}" \-d '{"query":"speaker explaining pricing","type":"all","limit":5}'curl -X POST "$BASE_URL/projects/$PID/videos/search" \"${AUTH[@]}" \-d '{"type":"labels","labels":["interview","keynote"],"match":"all","limit":20}'