Skip to main content
Pureframe AI is a video intelligence API. Upload your footage and Pureframe AI indexes every frame and every spoken word — then you can find any moment with a natural language query, a reference image, or both.

​
How it works

1

Upload

Send a video file to POST /v1/upload with a collection_id. Pureframe AI returns a job_id immediately — processing runs in the background.
2

Index

Pureframe AI extracts frames, generates visual embeddings, and transcribes speech. The job moves through queued → processing → done. Poll GET /v1/jobs/{job_id} for live progress.
3

Search

Send a text query, an image file, or an image URL to POST /v1/search. Results are ranked, timestamped segments with presigned video URLs, frame thumbnails, and transcribed speech.
4

Use the clips

Each result includes thumbnail_base64 — a base64-encoded JPEG that vision-capable models can consume directly, without fetching a URL.

​
Base URL

All endpoints are versioned under /v1.

​
Response envelope

Every API response shares the same wrapper:
Errors return a non-2xx status with a machine-readable code:
See Errors for the full list.

​
Jump in

Quickstart

Upload a video and run your first search in under 5 minutes.

Key Concepts

The mental model — collections, jobs, modalities, credits.

Search

Text, image, and transcript search, and how they combine.

Agent Vision

Connect Pureframe AI to any AI agent via MCP or function calling.