Jockey by TwelveLabs turns the footage archive your team scrubs through by hand into something you can query like a database, from a free tier to $100 a month.
Somewhere in your company's Drive there is a webinar recording where your best customer explains, unprompted, exactly why they bought. You know it exists. Nobody can find it. So the clip that should be in every sales deck gets recreated from memory, or an editor gets paid to scrub timelines looking for it. At Upwork's median rate of $35 an hour for video editors, just watching a forty-hour archive once costs $1,400. That's the price of watching. Tagging, logging, and cutting are all extra.
Jockey, which TwelveLabs launched on Product Hunt yesterday, is aimed at exactly that job. It's the agent layer on top of the company's two video models: Marengo indexes footage by meaning, Pegasus describes what actually happens inside it, and Jockey is the thing you talk to. Connect your library, then ask in plain language. Find every shot where someone uses the product outdoors. Pull the three most important moments from this recording. Which ad opens with a testimonial hook?
That last one is the tell for who this is really for. Jockey ships with automatic ad tagging: every creative in your library gets watched and labeled by hook, talent, format, and tone the moment it lands. Anyone who has managed a paid social program knows the artifact this replaces. It's the creative tracker spreadsheet, the one with columns for "concept" and "angle" that were filled in consistently for exactly three weeks after the intern built it.
The detail that makes this more than a niche media tool is the integration path. Jockey exposes an MCP server, which means you point Claude at it and your footage archive becomes something you can query from the chat window your team already lives in. No new interface, no new login for the marketing team to forget. Ask Claude for the moment, get the moment. ChatGPT support is listed as coming soon.
TwelveLabs is not a weekend wrapper, either. The company raised a $100M Series B to build video understanding models, and its customers were already using those models for sports analytics and media search before Jockey put a conversational front end on them.
The math is lopsided. Jockey prices by how much media you store: free for 5GB, $20 a month for 100GB, $100 a month for 500GB at the top tier. Set that against the manual version of the workflow. One pass through a modest archive at freelance editing rates costs more than a year of the $20 plan, and that pass produces notes in a document that go stale the day someone uploads new footage. The index doesn't go stale. It's the difference between paying for a search and paying for a search engine.
Now the honest part. Jockey is a research preview with limited sign-ups, so a team that needs production reliability this quarter should treat it as a pilot, not a dependency. If your problem is producing video rather than finding it, this solves nothing; that budget line is a different article. Regulated industries will want legal to read the trust documentation before pointing an indexing model at internal recordings, because Jockey groups people across your footage. TwelveLabs says it does this without building biometric databases or comparing against external records, but "the AI watched every all-hands" is a sentence your compliance team gets to react to. And if your entire archive is a dozen videos on YouTube, you don't need an agent. You need a playlist.
The bigger shift is worth sitting with. Every company quietly became a media company over the last decade, and almost none of them built the retrieval layer that actual media companies run on. The archive kept growing, findability didn't, and the gap got papered over with editor hours and institutional memory that walks out the door with every departure. Footage you can't find might as well never have been shot. As of this week, fixing that starts at twenty dollars a month.