TwelveLabs launches Jockey to search and cut video libraries

TwelveLabs launched Jockey on October 15, 2026, a video intelligence agent that sits on top of its Marengo and Pegasus models.

The company called it the official launch in a GlobeNewswire announcement. Jockey first appeared in a research preview that RuntimeWire dates to July 21, 2026, so October 15 is the general-availability date, not a first look. With this release, Jockey becomes the search layer for Marengo 3.5, TwelveLabs' latest embedding model.

Jockey runs inside the ChatGPT and Claude apps, or through an API and an MCP server at mcp.twelvelabs.io/jockey/mcp. In the release, TwelveLabs lists four additions: Jockey keeps search context across the two apps, so a producer can start a query in one and finish it in the other; it pulls video from YouTube, TikTok and Instagram Reels URLs to read trends against a creative brief; it captures transcripts, technical metadata, creator details and engagement metrics with each clip, so results rank by how a clip performed and not only by what is in it; and it can save, share or revoke its output, including an mp4 cut assembled from source footage and backed by playable video citations.

The release says Jockey is available in the ChatGPT and Claude apps, and the Jockey product page carries the same claim, listing a ChatGPT plugin connect step beside its Claude connector. The API docs send a developer down the same path: use Jockey from Claude, ChatGPT or the Agents API.

Jockey is the orchestration layer connecting two models: Marengo 3.5, the latest embedding model, and Pegasus 1.6, released the week before. TwelveLabs sums it up as "Marengo finds it, Pegasus understands it, and Jockey acts on it." Jockey finds and cuts footage. It does not generate video.

Marengo 3.5, in the company's launch post, represents video, audio, images, text and documents in one shared semantic space and removes the four-hour duration cap on asynchronous ingestion. TwelveLabs reports a mean score of 70.57 across five standard text-to-video retrieval datasets, a 6.29-point gain over Marengo 3.0. Pegasus 1.6 adds egocentric video understanding for what TwelveLabs calls Physical AI.

"Jockey is the connective tissue between our models that will take video understanding to the next level," said Jae Lee, CEO and co-founder of TwelveLabs, in the release.

TwelveLabs sells Jockey through its platform plans rather than a Jockey-only price. Its pricing page lists Free at $0 a month with 3 GB of cumulative storage, Plus at $29 a month with 50 GB and Pro at $99 a month with 200 GB, each including storage, media processing and API access. The Jockey product page shows the same three plans on annual billing at $290 and $990 a year.

For a creator with years of footage, the appeal is retrieval rather than rendering. If the index holds, a question that once meant scrubbing hours of rushes becomes a prompt, and the rough cut comes back with the timestamps attached.

Jockey is live: sign up free at jockey.twelvelabs.io, then add it from the Claude connector directory or connect through the API.