# Cohere previews video search for company AI agents in Compass

> Video Intelligence returns timestamped clips, transcripts and on-screen descriptions, and Cohere says it beats Gemini and Nova embeddings.

URL: https://claudegpt.si/news/cohere-compass-video-intelligence  
Published: 2026-11-17T10:58:23.025Z  
Author: ClaudeGPT Si Newsroom  
Section: AI Tools  
Topics: Cohere, AI search, Enterprise AI, Embeddings  
Publisher: ClaudeGPT Si

## Key takeaways

- Cohere opened a design-partner preview of Video Intelligence in Compass on November 16, 2026.
- It returns timestamped clips with transcripts and visual descriptions, and links back to the source footage.
- Cohere says it led seven video embedding models by 4.4 points on average in its own tests.

Cohere has opened a preview of Video Intelligence in Compass, its search platform for companies, the company [said on November 16](https://cohere.com/blog/video-intelligence-preview). It lets AI apps and agents search a video library and get back the exact moment that answers a question: a clip, its transcript and a description of what is on screen, with a timestamp and a link to the original footage.

The point is that most company knowledge in video is invisible to search. Meetings, lectures, product demos and safety training hold answers that a text search can't reach. Compass reads both what is said and what is shown, such as slides, charts and demonstrations, and works out where a relevant clip starts and ends when you search, instead of cutting videos into fixed chunks. Results respect the same access permissions as the source files, and video and documents come back in one search.

Cohere says Compass ranked first on average against seven video embedding models across 11 datasets and 11,133 queries, 4.4 points ahead of the strongest alternative on nDCG@10, a standard ranking score. Against named rivals, it reports leads of 2.4 points over Amazon Nova Multimodal Embeddings on public benchmarks, 5.1 over Gemini Embedding 2 on soccer broadcasts, and 2.0 over Qwen3-VL-Embedding-8B on enterprise footage. These are Cohere's own tests; no independent results exist yet.

The preview is open to design partners, who can run it in Cohere's managed cloud or in their own environment. Cohere has not published a price. It fits Cohere's focus on private deployments for businesses, the same pitch it makes against [Mistral](https://claudegpt.si/news/cohere-vs-mistral).

## Sources

- [Video intelligence: Make every moment in your video library searchable](https://cohere.com/blog/video-intelligence-preview), Cohere
