Search for moments
Search reusable indexed evidence in natural language and inspect ranked, timestamped matches before using them downstream.
Learn how to searchTurn video into evidence-backed decisions
DeepFrame is a video intelligence API. Upload video once, then reuse its indexed evidence to search for moments, ask cited questions, run rule-based reviews, or extract structured answers.
Four outcomes, one evidence base
Each path reuses the same indexed video. Start with one outcome and add the others without uploading the video again.
Ask questions
Ask questions across one or more indexed videos, retain follow-up context, and return answers with citations the user can inspect.
Learn how to ask questionsRule-based review
Apply a stored rulebook to indexed evidence. Findings retain result state, rule provenance, limitations, and evidence paths.
Learn how to run a reviewStructured answers
Use a stored analysis profile and report template by ID and version. Named Answers carry typed fields, uncertainty, provenance, and evidence.
Learn how to extract answersWhat A Semantic Timeline Looks Like
This illustrative output shows how an analysis can group continuous video into meaningful events. The model selects the boundaries from your profile instructions and indexed evidence; these are not fixed platform stages or user-supplied processing chunks.
00:00 - 00:18
Prepare and approach
The first meaningful activity begins after the scene is visible.
clip + frame
00:18 - 00:42
Perform the main check
One continuous activity remains one event even when internal inspection uses smaller chunks.
clip + frame
00:42 - 01:05
Record the result
The event ends when the observable action changes or the evidence becomes unclear.
clip + frame
From video to evidence-backed decisions
Prepare a video once, choose the result you need, verify its evidence, then use the result in your application or a report.
- Follow the upload flow
Upload and index once
Create an upload session with the exact file_size, send the bytes to the returned target, and wait until the video is searchable.
- Compare the outcomes
Choose the result you need
Reuse the same indexed evidence to search, ask questions, run a rule-based review, or extract structured answers.
- Read result semantics
Read Findings or Answers
Inspect evidence, uncertainty, limitations, and provenance. Do not turn uncertain or not_found into a clean pass.
- Read Reports And Templates
Format Reports
Use a stored report template to select and format existing answer data. Rerender changes presentation only.