Video Datalake
Upload videos and live streams and the platform auto-indexes them. Then search a collection (semantic / keyword / image), read any moment (captions, frames, clips, speakers, entities), maintain a face-reference library, and receive safety-detection alerts — all in natural language.
How the video data lake works
A video data lake stores raw video and keeps its derived content queryable. Instead of your agent downloading and watching a video, it asks the lake in natural language and gets back the exact moments and text it needs.Videos · live streams
file · URL · rtmp
↓ index
Video Data Lake
embeddings · captions · events
↓ Datalake API · search · read
Agents · MCP
your LLM apps
↺ agents query the lake in natural language — ingest once, read and act
How agents use it: ingest once → the lake indexes continuously → your agent calls Search to find moments, Get Moment to read captions/frames/clips, and Events to react to what happened — all as plain API calls, or through our MCP Server.
Start here
Quickstart
Create a collection, upload a video, and read results in five minutes.
Core Concepts
Collections, the async model, moments, and derived content.
Authentication
One key, one host, request conventions.
Search
Semantic, keyword, hybrid, and image search over your video.
Drive it with an agent
Video Agent
An agent that operates the Datalake in natural language — with the lake as its long-term video memory.
MCP Server
The official MCP server that connects Claude, Cursor, or any MCP client to the Datalake.
Building with an agent? Every page has a Copy · Open in ChatGPT / Claude button (top-right), the whole site is published as
llms.txt / llms-full.txt for LLM ingestion, and the MCP Server — plus a CLI and a Claude Code / Codex plugin — let an agent operate the Datalake directly.SDKs: the Datalake is REST-first today — every example is copy-paste
curl and Python requests. Official SDKs are on the way.Coming from API 1.0? The previous Visual Intelligence / Visual Search / Visual Agents docs are still online — pick 1.0 · Sunset Aug 13 in the version selector next to the logo at the top of the page. API 1.0 will be sunset on August 13, 2026, so migrate to the Datalake API before then.