Jockey
TwelveLabs · Jockey by TwelveLabs: Video Intelligence Agent That Searches, Organizes, and Clips Your Video Library
Jockey is TwelveLabs' video intelligence agent: it indexes a library of videos and photos, answers plain-language questions with timestamped clips, groups footage by person, topic or event, and assembles selected moments into a reel you can trim and download as an MP4. You use it inside Claude or ChatGPT as a connector, or from your own app through the TwelveLabs API. It suits creators, editors and marketing teams with more footage than they can rewatch.
Best for
Creators, video editors and marketing or social teams with large video and photo libraries who already work in Claude or ChatGPT and want to find, compare and clip moments by describing them, plus developers who want library-level video question answering behind an API
Not ideal for
People who need a full timeline editor or generative video tool, teams whose footage lives mainly on TikTok, Instagram or Drive share links they expect to ingest by URL, organizations that need self-hosting, and buyers who need fixed, published response quotas before committing
Who it's for
Creators, video editors, marketing and social teams, and developers who need to search, understand and reuse a large video and photo library
Capabilities
- Search a video and photo library by meaning, using a phrase, an image, a clip, a person, a place or an action, with answers that cite timestamped moments
- Reasons across every item in a knowledge store, so it can compare videos, track a subject across many of them and rebuild a timeline of appearances
- Groups the same person across the library from visual, motion and context cues (TwelveLabs says this is not traditional facial recognition), and sorts footage into sets by person, place, topic, event or project
- Scene-by-scene breakdowns shown as interactive previews with descriptions and video references inside ChatGPT or Claude
- Assembles selected moments into a reel that keeps sub-second start and end points, with timeline editing to trim clips, reorder them and preview the sequence
- Downloads the finished reel as an MP4 and saves analyses as shareable Jockey links that teammates can reopen with the underlying clips
- Adds footage from your device or Google Drive through an upload link, or from a pasted YouTube or direct video URL
- Marketing features listed on the product page: ad tagging by hook, talent, format and tone, hook comparison across a library, talent and logo recognition, social trend intelligence for YouTube, TikTok and Instagram Reels, and 15 to 30 second concept video previews
- Responses API with Python and Node.js SDKs: knowledge stores, multi-turn sessions, streaming, typed JSON output from a JSON Schema, per-request domain instructions, optional web search with cited URLs and intermediate reasoning steps
Limitations
- Each question targets a single knowledge store, and TwelveLabs' docs say you cannot query across stores
- The pricing page describes Workload allowances only as Basic, Medium and High, so it is hard to know in advance how many Jockey responses a plan covers. The API rate limits allow 15 Jockey Response requests a day on Free and 200 a day on Plus and Pro
- Free and Plus have no overage: new Workload requests are blocked once the monthly allowance runs out, and new uploads are blocked at the storage allowance. Overage is optional on Pro and on by default on Enterprise
- Jockey cannot fetch TikTok, Instagram or Google Drive share links by URL, so those files have to go through the upload page
- Uploads through the Claude and ChatGPT connector are capped at 5 GB per video and 32 MB per image (the API's multipart upload goes up to 10 GB per video), and longer videos take longer to index before they can be searched
- It finds, explains and assembles footage, but the documented editing is limited to trimming and reordering clips in a reel, so finishing work still happens in a video editor
- Accounts that existed before October 15, 2026 stay on legacy billing, with different plan terms, until they move: free and Developer accounts on January 4, 2027 unless they switch earlier, and paid Jockey Plus or Pro subscriptions, which cannot switch early, from January 2027 for monthly billing or at the end of an annual term
- No self-hosted deployment is documented. Enterprise adds SSO/SAML and custom limits
Use cases
- Finding every clip where someone mentions a product name across a back catalog of videos
- Pulling a creator's best moments on a topic and assembling a 30-second rough cut to export as an MP4
- Breaking a long video into scenes with descriptions to pick moments for a new edit
- Tagging a library of ads by hook, talent, format and tone and comparing which openings performed
- Writing a creative brief or episode plan from what an existing body of videos already covers
- Adding video and photo search and question answering to an app through the Responses API
Our take
Jockey is most useful as a memory for footage you already own rather than as an editor. Its strengths are library-wide questions with answers that point back to timestamps, grouping the same person across many videos, and getting from a question to a rough MP4 without leaving Claude or ChatGPT. Its reel editing stops at trimming and reordering, so plan on finishing elsewhere. The main thing to check before paying is volume: storage allowances are published, but Workload allowances are not given in responses or hours, and Plus cannot buy overage, so test a realistic week of queries on Free or Plus before relying on it for a team.
Who should use it
Creators and editors who need to find quotes and moments across hours of their own footage, marketing and social teams that want to tag and compare a library of ads, anyone who already works in Claude or ChatGPT and wants video search there, and developers who want library-level video question answering without building their own pipeline.
Who should skip it
Editors looking for a complete editing or generation tool, teams whose videos mostly live on TikTok, Instagram or Drive share links they expect to add by URL, organizations that need self-hosting, and buyers who need a fixed, published number of responses per plan.
Strengths
- Works inside Claude and ChatGPT, with no new editor or chat app to learn
- Answers point back to timestamped footage instead of text alone
- Reasons across a whole library, including the same person across many videos
- Goes from search to a trimmed MP4 reel in the same conversation
- The same agent is available through an API with SDKs, sessions and structured output
Weaknesses
- Workload allowances are not published as a number of responses or hours
- No overage on Free and Plus, so work stops when an allowance runs out
- One knowledge store per question
- TikTok, Instagram and Google Drive links cannot be ingested by URL
- Reel editing is limited to trimming and reordering clips
Jockey pricing
Free
$0
Billed monthly
- 3 GB of stored assets
- Basic Workload allowance and rate limits
- API rate limit of 15 Jockey Response requests a day
- Usage capped at plan limits, no overage
Plus
$29
Billed monthly
- $290 a year on annual billing
- 50 GB of stored assets
- Medium Workload allowance
- High rate limits, including 200 Jockey Response API requests a day
- No overage
Pro
$99
Billed monthly
- $990 a year on annual billing
- 200 GB of stored assets
- High Workload allowance
- High rate limits, including 200 Jockey Response API requests a day
- Optional overage, off by default
Enterprise
Custom
- Custom storage, Workloads and rate limits
- SSO/SAML included
- Overage on by default, by agreement
Free tier limits: 3 GB of stored assets, a Basic monthly Workload allowance and an API rate limit of 15 Jockey Response requests a day. Requests and uploads stop when an allowance is reached.
Note: These plans cover the whole TwelveLabs platform: storage (Managed Intelligence, which includes indexing, Jockey Search and re-indexing for new models) plus a monthly Workload allowance shared by Jockey responses, Analyze, Segment, Embed and Marengo Search. Allowances are listed as Basic, Medium and High rather than in hours or responses. Pro overage is $2 per 1M Jockey response tokens and $0.065 per GB per month of storage. The plans started on October 15, 2026 for new users; existing Free accounts, and Developer accounts that do not choose Plus or Pro, move on January 4, 2027, and legacy Jockey Plus and Pro subscriptions move to the matching plan at a billing date from January 2027 (monthly) or at the end of their annual term. TwelveLabs' docs say monthly subscriptions move on their first billing date in January 2027, while the pricing page says the next billing date on or after January 4, 2027.
Technical specs
Modalities
Video, Image, Audio, Text
API pricing
Jockey responses are measured in tokens and count toward the plan's monthly Workload allowance; Pro overage is $2 per 1M tokens, and storage overage is $0.065 per GB per month
Available models
Where Jockey excels
Mining a creator's back catalog
Jockey can find every time a topic or product comes up across years of videos, return the timestamps and assemble the strongest moments into a short MP4 for a new post.
Ad library review for a marketing team
Tagging ads by hook, talent, format and tone and comparing them across the library helps a team see which openings and formats to reuse before the next shoot.
Video question answering inside a product
The Responses API returns structured, timestamped answers from a knowledge store with sessions and JSON Schema output, so an app can add video search and Q&A without training its own models.
How Jockey fits alongside other tools
Jockey and Eddie AI
Eddie AI logs and transcribes footage for a project and builds rough cuts it hands to Premiere Pro, DaVinci Resolve, Final Cut Pro or Avid. Jockey works across a whole video and photo library to find, explain and pull moments by meaning, with only light reel editing. A team could use Jockey to locate material across an archive and Eddie to cut it into a finished edit.
Jockey and invideo
invideo is a browser editor in which agents build cuts on a multitrack timeline and generate missing shots from many models. Jockey has no multitrack timeline and is built around footage you already have: it searches, understands and clips it, which can feed an edit finished in a tool like invideo.
Jockey and Descript
Descript edits video and podcasts by editing the transcript of a project. Jockey answers questions across many videos and photos at once and returns timestamped moments and MP4 reels, so it suits finding material in a large library before editing it in Descript.
Frequently asked questions
What is Jockey by TwelveLabs?
Jockey is TwelveLabs' video intelligence agent. It indexes your videos and photos into a knowledge store, then answers plain-language questions about them with timestamped clips, organizes footage by person, topic or event, and assembles selected moments into a reel you can trim and download as an MP4.
How do I use Jockey in Claude or ChatGPT?
In Claude, open Customize, then Connectors, search for Jockey by TwelveLabs and add it. In ChatGPT, open Plugins, search for Jockey and add it. You then sign in to or create a free TwelveLabs account, add videos through the upload link Jockey gives you or by pasting a YouTube or direct video URL, and start asking questions once the footage is indexed.
How much does Jockey cost?
Under the pricing introduced on October 15, 2026, Free costs $0 with 3 GB of storage, Plus costs $29 a month ($290 a year) with 50 GB, Pro costs $99 a month ($990 a year) with 200 GB and optional overage, and Enterprise is custom. Each plan also has a monthly Workload allowance, described as Basic, Medium or High, that Jockey responses share with other TwelveLabs API work. Pro overage for Jockey responses is $2 per million tokens.
Is Jockey open source?
No. An earlier Jockey was an open-source project built on LangGraph, but TwelveLabs says the current Jockey was rebuilt and relaunched as a first-party TwelveLabs product. Developers build on it through the TwelveLabs API and SDKs rather than by forking it.
Does Jockey use facial recognition?
TwelveLabs says Jockey groups people across a library from visual, motion and contextual cues rather than traditional facial recognition, does not build biometric databases or compare your data against external records, and does not share identity information between accounts.
Which models power Jockey?
TwelveLabs says Jockey runs on its own models: Marengo, an embedding model that indexes footage by meaning, and Pegasus, a video language model that describes and reasons about what happens in it. Jockey orchestrates them across the whole library.
Jockey vs Eddie AI: what is the difference?
Eddie AI is an assistant editor that logs footage and builds rough cuts, then hands them to Premiere Pro, DaVinci Resolve, Final Cut Pro or Avid. Jockey is aimed at searching and understanding a whole library of videos and photos and pulling moments from it, with only light reel editing, so the two can be used together.
Integrations & fit
About Jockey
Jockey sits on top of TwelveLabs' own video models: Marengo indexes footage by meaning and Pegasus describes what happens in it, and Jockey runs an agentic search loop across the whole library rather than one clip at a time. There is no separate chat app to learn. You add Jockey by TwelveLabs from the connector directory in Claude or ChatGPT, sign in to a TwelveLabs account, then upload files from your device or Google Drive through an upload link, or paste a YouTube or direct video URL. Footage goes into a knowledge store, and once it is indexed you can ask for every moment someone enters a room, every mention of a product, a scene-by-scene breakdown, a comparison of your top videos, or a creative brief built from them. Answers come back with timestamps and video references, scene breakdowns render as interactive previews in the conversation, and selected moments can be turned into a reel with sub-second clip boundaries, trimmed and reordered on a simple timeline, downloaded as an MP4, or shared as a Jockey link. TwelveLabs' Connect web app is where you see stored videos, group them into collections and check usage. Developers reach the same agent through the Responses API and Python and Node.js SDKs, with multi-turn sessions, JSON Schema output, domain instructions and optional web search. Jockey reached general availability on October 15, 2026, alongside new unified TwelveLabs pricing: Free, Plus at $29 a month, Pro at $99 a month and custom Enterprise, each with a storage allowance and a monthly Workload allowance shared with the rest of the TwelveLabs API. The name is older than the product: TwelveLabs says the current Jockey was rebuilt as a first-party service and is not the earlier open-source LangGraph project. The tradeoffs are that each question targets one knowledge store, Workload allowances are described as Basic, Medium and High rather than in hours or responses, only Pro and Enterprise can buy overage, TikTok, Instagram and Google Drive share links cannot be fetched by URL, and Jockey finds and assembles footage rather than replacing a full editor.
Updates from Jockey
TwelveLabs moved Jockey from research preview to general availability, added it to ChatGPT alongside Claude, and shipped import from YouTube and direct links, Google Drive or your device, scene-by-scene breakdowns in the chat, sub-second clipping, timeline editing of reels, MP4 download and shareable Jockey links.
TwelveLabs replaced separate Jockey subscriptions and API plans with one set of plans: Free, Plus at $29 a month, Pro at $99 a month and Enterprise, each with a storage allowance and a monthly Workload allowance shared by Jockey responses and other API work. New users start on these plans, and existing accounts move from January 2027.
Are you the founder? Claim this listing →