Skip to main content
Pureframe AI returns evidence that an agent can use to support an answer. Your integration should preserve that evidence and make it available for verification.

​
What “source-backed” means here

Each search_videos result can include:
  • video_id, start_secs, and end_secs to identify the source and location
  • clip_url to open the relevant part of the video
  • thumbnail_base64 to provide the matched frame to a vision-capable model
  • text_snippet for matching spoken content
  • relevance_score to order results from the same search
A relevance score is not proof that a statement is correct. The agent should base its claim on the returned frame or transcript text and cite the source timestamp.

​
Why this matters for agent reliability

Search narrows a video library to relevant segments that a model can inspect. This reduces the amount of context the model must process, but it does not guarantee a correct answer. Require the agent to call search_videos before making claims about the library. For important claims, require a video_id, timestamp, and short explanation of the supporting evidence.

​
Building this into your own UI

Keep the source fields with any generated summary. Show the video name, timestamp, and playback link next to the answer. Let users inspect the source instead of presenting generated text on its own. See Search results for the REST response fields.