We’ve Learned How to Capture Everything. Here’s How We Remember It.

After my kids were born, my phone’s camera roll slowly became a record of their lives: first steps, first days of school, first games. Like most parents, I recorded more footage of their early years than all of my past videos combined. Some I took because I knew I’d want to relive those moments, others I took without thinking too much about it. Often, I scroll back far enough to see how small they were, and I remember how it felt when they fit in my arms. And on days when life seems to be moving too fast, they’re a wonderful reminder to slow down.

We can capture more moments, preserve more memories, and create more content than any generation before us. Yet the more we accumulate, the harder it becomes to find what matters. That tension applies not only to my family footage but to people whose work depends on video at scale: a creator with hours of footage from a single shoot, a marketing team with years of campaigns, a production company with an archive full of stories no one has had time to revisit. There’s enormous value in that material, but getting to it still takes time, patience, and often a very good memory.

Video is especially hard to make useful. What matters might be said aloud, shown on screen, or only become clear when you see several moments together. A useful answer can’t simply be a string of text, it should take you back to the footage so you can judge it for yourself.

That’s why we built Jockey.

Jockey builds on years of work across TwelveLabs: Marengo, our embedding model, helps find moments by their meaning, while Pegasus, our video language model, helps describe and understand what happens within them. Jockey brings these capabilities together into a video intelligence agent that can reason across an entire library, connect related moments, and turn what it finds into something you can play, share, and build on. You can ask questions of your video and image library the way you would naturally think about it: by person, moment, subject, or context. It has taken researchers, engineers, designers, and teammates across the company to turn that foundation into a product people can actually work with.

For a content creator, that could mean finding the best moments from a shoot and assembling a first cut. For a marketing team, it could mean looking across past campaigns to see how a message or visual style has evolved, then using those insights to make something new. For production companies, sports teams, and robotics companies, the questions differ, but the need is similar: understand what happened across a growing body of video without having to watch every second themselves.

LLMs have changed what we expect from our written knowledge. But so much of our lives and work lives in video. Now Jockey lets us ask questions of our footage, too.

There’s no model or agent that can tell me what it felt like to hold my kids when they were little. Those memories are mine. But the videos help me find my way back to moments I might otherwise have forgotten to revisit.

I hope Jockey helps more people do that with the footage that matters to them and discover what they can make from it next.

Jockey is available now at twelvelabs.io/jockey. Bring in a folder of footage you haven’t looked at in a while and ask it something. That’s usually when it clicks.