TwelveLabs opens research preview of Jockey, an AI agent that searches entire video libraries via Claude

TwelveLabs announced on July 21st, 2026, that its first video-intelligence AI agent, Jockey, is now available in a research-preview program, enabling users to search entire video libraries via Anthropic's Claude with natural-language prompts. The agent leverages TwelveLabs' Pegasus, Marengo, and Search 2.0 models to process thousands of videos and photos simultaneously, with ChatGPT support coming soon.

TwelveLabs opens research preview of Jockey, an AI agent that searches entire video libraries via Claude https://x.com/twelve labs/status/2079685809288908996 https://x.com/twelve labs/status/2079685809288908996 TwelveLabs https://twelvelabs.io announced on X that its first video‑intelligence AI agent, Jockey , is now available in a research‑preview program. In a four‑part thread posted on July 21st, 2026, the company said Jockey moves beyond the usual single‑file analysis model and can “process your entire library, right within Claude.” The preview invites users to sign up at the link provided in the last tweet of the thread. One‑prompt, whole‑library search Typical video‑AI services – such as Google Cloud Video Intelligence https://cloud.google.com/video-intelligence/docs/ or Microsoft Video Indexer https://learn.microsoft.com/en-us/azure/media-services/video-indexer/ – require users to upload a video, then receive a transcript or label set for that single file. Jockey flips that paradigm. According to the announcement, a user can ask Claude‑powered prompts like “ Build me a short from my best moments this month ” or “ What do the hooks for my best‑performing social ads have in common? ” and the agent will scan “thousands of videos and photos together, not one at a time.” The thread lists three core components of the TwelveLabs stack that power Jockey: Pegasus for extraction – the model that pulls visual and audio features from raw media. Marengo for search – a retrieval layer that indexes those features across a library. Search 2.0 for multi‑step retrieval – a higher‑level orchestrator that resolves complex, multi‑turn queries. All three are described as “the latest advancements in video understanding” by the company. While the thread does not disclose performance metrics, the architecture mirrors the increasingly common “foundation‑model‑plus‑search” approach that other AI startups have adopted for text and code. Integration with Claude and upcoming ChatGPT support Jockey is positioned as a plug‑in for Anthropic’s Claude, the conversational LLM that emphasizes safety and instruction‑following. Users can invoke Jockey directly from a Claude chat session, turning a natural‑language request into a cross‑media retrieval operation. The company also notes that “ChatGPT support is coming soon,” indicating plans to broaden access to OpenAI’s conversational interface. The research‑preview model means the feature is not yet generally available. Interested developers can “connect Jockey to your Claude or build via API to start searching your library today,” the tweet reads, with a sign‑up link https://t.co/IGe3rb1uDu https://t.co/IGe3rb1uDu that points to the Jockey landing page http://twelvelabs.io/jockey http://twelvelabs.io/jockey . Why the timing matters The launch arrives amid a surge of enterprise interest in video‑first content. Brands, creators and advertising agencies are amassing petabytes of footage, yet most AI tools still operate at the single‑file level, leaving a gap in search, repurposing and analytics workflows. By offering a library‑wide search agent, TwelveLabs aims to capture a segment that currently relies on manual curation or fragmented toolchains. From a competitive standpoint, the move also differentiates TwelveLabs from pure‑play video‑AI vendors that focus on transcription or object detection. Jockey’s “agent” framing – a conversational, multi‑step interface – aligns with broader industry trends where LLMs are used as front‑ends to domain‑specific knowledge bases e.g., code assistants, product‑catalog bots . If the research preview yields strong user engagement, the company could position Jockey as a core component of future video‑asset‑management platforms. The road ahead The preview is limited in scope, and TwelveLabs has not disclosed pricing, roadmap milestones or a target general‑availability date. No funding round or investor details were mentioned in the announcement, and the company’s prior financing history remains opaque from publicly available sources. What is clear is that TwelveLabs is betting on the convergence of video understanding models and conversational AI to unlock bulk media insights. As more firms look to reuse existing video assets for short‑form social content, an agent that can surface “the best moments” or “common hooks” with a single prompt could become a valuable productivity layer – provided it can scale to the data volumes and latency expectations of enterprise users. Key takeaways - Jockey processes an entire video library rather than a single file, powered by TwelveLabs’ Pegasus, Marengo and Search 2.0 models. - The agent operates inside Anthropic’s Claude and will later support OpenAI’s ChatGPT. - The feature is offered as a research preview; sign‑ups are open via the company’s landing page. - The launch targets a growing need for cross‑media search and repurposing tools in the video‑heavy marketing and advertising ecosystem. The story is based on TwelveLabs’ X thread posted on July 21st, 2026. All links are to the original announcement or publicly available product documentation.