AI that truly understands video. Uses multimodal models (Marengo/Pegasus) to search, analyze & generate text from video content at scale.
This is the 8th launch from TwelveLabs. View more
Jockey by TwelveLabs
Launching today
Jockey is the first AI that understands your entire media library across photos and video. It analyzes thousands of hours of footage so you can find and connect people, moments and context across videos. Use it in Claude or ChatGPT, or build with the API/SDK. Ingest YouTube, TikTok and Instagram Reels links; assemble content into MP4s; and create and share video-driven artifacts with playable clips. Every TwelveLabs model upgrade makes Jockey better, without re-integration.
Launch Team
Video is the one area that I feel AI hasn't yet reached its potential in and this tool seems to go in that direction.
Congrats on the launch team!
@sumitsaurabh927 strong +1 and @TwelveLabs is a pioneer here. follow for future launches!
@tessak22 thanks for your continuous support, Tessa! you're the best
@jaelee_ @aiden_lee7 How well does TwelveLabs understand long videos and find specific movement?
oh wow, searchable videos? I have a ton of use cases for these. Congrats on the launch.
handling thousands of hours of video across chat interfaces is impressive, but does the MCP setup ever run into context window bottlenecks when a query spans too many disparate clips in a massive library??
@sumonmatrix great question. The library itself never goes into the chat's context window. Jockey searches your videos and reasons over the matching clips on our side, and ChatGPT or Claude only gets back the result.
What comes back depends on which of our two answer tools the assistant calls:
jockey_query -- returns a written answer with citations to the clips it used. That answer does go into the chat's context, but its length depends on what you asked for, not on how many hours of video were searched.
jockey_render_response -- shows the answer as an interactive visual inside the chat. The assistant only receives a short status line, so the answer itself adds almost nothing to its context.
Follow-up questions continue the same Jockey session, so the assistant doesn't need to pass earlier results back in.
Honestly the main thing I use this for is to find clips of the cat sleeping in the background of family vacation videos. It's a little annoying because the system sometimes sees him as a pillow but it still helps me avoid going through three years of video just to find something strange.
Congrats @aiden_lee7 and team. Eight launches and still shipping from user feedback is the part I respect most. The social ingestion from YouTube and Reels links is a smart addition; no upload step is exactly the kind of friction people quit on.