Over the year you have racked up the phone videos. And by now you don’t know what you really have in the video library of your phone. You have a few hundred videos sitting in folders and some more within apps but you cannot easily index it or call it at will. The footage you need exists somewhere, but finding it means scrubbing through files one at a time.
Jockey by TwelveLabs is built for exactly that job, and since mid-October it runs inside both Claude and ChatGPT.
What it is
Jockey is a video-understanding agent from TwelveLabs, the company behind the Marengo and Pegasus video models. It indexes your video and photo library, then lets you ask questions about it in plain English from a chat window. On 15 October 2026 TwelveLabs announced: "Today, we're bringing Jockey, TwelveLabs' video intelligence agent, into general availability." The launch then went to the top of Product Hunt, where Jockey was ranked #1 of the day for 16 October 2026 with 424 upvotes.
The TwelveLabs docs describe it as a way to "ask questions across your videos and receive cited moments", and note that "Jockey plans its own steps." You don't pick the operations; it decides whether to search, analyse or segment.
Who it's for
YouTubers and short-form creators with a back catalogue they want to mine for clips, compilations and repurposed shorts.
Editors working through hours of raw footage who want to locate shots by description instead of file name.
Small marketing teams who need to answer "where did we show the product?" or "which hooks worked?" across a library.
If you only make a handful of videos a month and remember what's in each, you don't need it.
What it actually does
The general availability release shipped seven updates, according to the TwelveLabs launch post:
ChatGPT support through an MCP connection, alongside Claude.
Imports from YouTube, Google Drive or your device. For online videos you paste a URL; for Drive or local files Jockey gives you an upload link.
Scene-by-scene breakdowns shown as an interactive answer in the chat, with previews and references.
Sub-second precision: "A clip selected from 12.4 to 16.8 seconds keeps those boundaries."
Timeline editing: trim, reorder and preview the clips Jockey assembles.
MP4 download of the finished reel directly from ChatGPT or Claude.
Shareable links so teammates can reopen an analysis and check the moments behind it.
TwelveLabs' creator guide adds that Jockey "can ingest directly from YouTube, TikTok, and Instagram Reels URLs", indexing the video without you downloading and re-uploading it. One editor quoted there, Nick Gebben, put the benefit simply: "Instead of opening file after file, I was able to search and play in English." Another, Kyle Chalmers, pointed it at his whole channel and had it group videos "by what is actually happening on screen."
The same guide gives a useful prompt template: "Find me every shot of [subject] and give me the timestamps."
Pricing (as of 20 October 2026)
TwelveLabs is moving customers to new pricing. The official pricing page lists, as of 20 October 2026:
Plan | Monthly price | Storage |
|---|---|---|
Free | $0 | 3 GB |
Plus | $29 | 50 GB |
Pro | $99 | 200 GB (optional overage) |
Enterprise | Custom | Custom |
The page says "Customers will begin moving to the new pricing automatically on January 4, 2027", and that if you have an active Jockey Plus or Pro subscription you cannot switch early. It shows a 17% discount for annual billing but no annual dollar amounts. The Jockey product page shows the annual prices instead: $290 a year for Plus and $990 a year for Pro, as of 20 October 2026. Check the pricing page on the day before you pay for anything.
Two details on that page matter for planning. Overage "is off by default", and "new Workload requests are blocked when your Workload allowance is exhausted." On Free and Plus, usage is capped at plan limits.
Getting started
Create a TwelveLabs account from the Jockey page. The Free plan is enough to start.
Add the connector. In Claude: Customize → Connectors → Discover, search "Jockey by TwelveLabs". In ChatGPT: Settings → Plugins → Browse Directory, same search. (If you're new to Claude connectors, our piece on Claude extensions without config covers the basics.)
Connect your TwelveLabs account and choose or create a video collection.
Add footage. Paste YouTube links, or ask for an upload link for Drive and local files. Start small: 3 GB fills quickly with video.
Wait for processing. TwelveLabs says videos must finish processing before you can query them.
Ask a specific question, review the timestamps, then assemble, trim and download the MP4.
Worked example: every sponsor mention across 50 videos
Say you're a tech YouTuber who has run the same VPN sponsor in roughly 50 videos over two years, and the sponsor wants a 60-second supercut for its renewal pitch.
Create a collection called "Sponsor audit" and paste the 50 YouTube links. On the free tier, check how much of your 3 GB allowance that uses; if it won't fit, start with the 15 most recent.
Ask: "Find every segment where I mention [sponsor name] or show their app on screen. Give me video title, start and end timestamps, and one line on what I say."
Spot-check five results against the originals. You're looking for missed reads (usually ones where you said the name quickly) and false positives.
Follow up: "Rank these by energy and clarity, and pick the eight strongest moments under 8 seconds each."
Ask Jockey to assemble those eight into a reel, trim the edges in the timeline, and download the MP4.
Save the analysis as a shareable link and send it to the sponsor so they can see the source moments.
The same flow works for "every time I recommend a product" (for affiliate link audits) or "every outdoor shot of me talking to camera" (for B-roll).
A 20-minute test checklist
Minutes 0–5: Sign up on Free, add the connector in whichever assistant you already use, create one collection.
Minutes 5–10: Add five videos you know well, including one long one. Note how long processing takes.
Minutes 10–14: Ask three questions where you already know the answer (a specific shot, a phrase you said, an on-screen object). Score each: right timestamp, close, or wrong.
Minutes 14–17: Ask one vague question ("what's working?") and one specific one, and compare. TwelveLabs' own advice is "Specificity in, specificity out."
Minutes 17–20: Build a three-clip reel, trim it, download the MP4 and check that the cut points land where you set them.
If at least two of your three known-answer questions come back right, it's worth loading more of your catalogue.
Alternatives
Adobe Premiere Pro's media intelligence search: pick this if you already edit in Premiere and want footage search without uploading anything. Adobe says analysis "happens locally on your computer and Adobe cannot see your footage."
Descript: pick this if your content is talk-driven and you want to edit by editing the transcript. It's built around speech, so it suits podcasts and talking-head videos more than visual search.
The TwelveLabs API: pick this if you're a builder who wants to call the models directly ("choose exactly what runs") and wire video search into your own product or workflow. Our ngram video agent review covers another approach to agent-driven video.
Limitations
Pricing is changing. Existing customers move to the new plans automatically from 4 January 2027, and active Jockey Plus or Pro subscribers cannot switch early.
Free storage is small. 3 GB is a handful of long videos, so you'll test on a sample, not your full catalogue.
Data retention is unclear. On Product Hunt, two users asked how data is kept; neither question had a reply when we checked.
Results depend on your prompts. TwelveLabs says vague questions like "What's working?" produce generic answers.
Hard limits aren't documented in the launch post. It lists no file size, length or processing-time limits.