Twelve Labs

Twelve Labs is a question answering tool. A video intelligence platform and API for searching, analyzing and summarizing footage across vision, audio and language, with the Marengo retrieval model and the Pegasus generative model behind it.

Twelve Labs is a video intelligence platform built around multimodal models that read footage the way a person would: watching the frames, listening to the audio and parsing the spoken words at the same time. The product line splits into two engines. Marengo handles the indexing and search layer, turning long recordings into vector representations that can be queried with a phrase, an image or a clip. Pegasus handles the generative side, producing summaries, highlights and answers about what happens inside a video. Around those models sit tools for embedding video search into an existing application and for running analysis jobs at scale. The primary audience is technical. Most of the workflow happens through an API, so teams typically arrive with a library of footage, a product idea and engineers who can call endpoints. Media and entertainment companies use it to make archives navigable, sports and broadcasting groups to find moments in match footage, advertising teams to locate product appearances, and security or government operations to retrieve relevant segments from long camera records. Jockey, a conversational agent built on the same models, gives less technical users a way to ask questions about a video in plain language without writing integration code. In practice the work follows a familiar shape: upload or connect video, let the platform index it against the chosen model, then issue natural language queries and receive the matching moments with timestamps rather than a whole file to scrub through. The analysis endpoints extend this from retrieval into description, so a team can ask for a recap of a meeting or a list of key events without watching the recording. Documentation, SDKs and sample apps are published openly, which shortens the path from first API key to a working prototype. Compared with transcription services, Twelve Labs treats speech as one signal among several, so a query about a visual action still works on silent footage. Compared with general purpose video generators, it reads existing material instead of producing new clips. The tradeoff is that it is infrastructure rather than a finished app: value depends on what a team builds on top of the API.