Field noteNov 16, 20265 min read1 source

Introducing Video Intelligence in Compass

In one minute

Cohere announced Video Intelligence in Compass, a preview that lets enterprise applications and agents search video alongside documents and retrieve timestamped moments with transcripts and visual descriptions. The feature could make recorded knowledge easier to verify and use in RAG workflows, but access is currently through a design-partner program.

What changed: Cohere previews Video Intelligence in Compass, adding video understanding alongside its existing enterprise search offering.

What happened

Cohere’s announcement says Video Intelligence brings video collections into Compass’s enterprise search and retrieval system. Users can ask questions such as when a team agreed to a launch date or where a presenter explained a chart, and receive a relevant moment rather than a whole recording. The system is designed to interpret both speech and visual content over time: a transcript alone may miss what a chart shows, and a single frame may not capture how an event unfolded.

Compass determines the relevant start and end times at search time instead of relying on predefined clips. Results include a timestamped clip, transcript and visual description, with a link back to the original footage for verification. The announcement says a single search can retrieve video moments alongside matching pages, slides and spreadsheets from a shared index. Permission-aware retrieval is intended to ensure users see only content they are authorized to access. Cohere also says organizations can configure fields, model choices and deployment options.

Applications and agents can use the retrieval system in enterprise search, retrieval-augmented generation and agentic workflows. Cohere says developers can connect an agent to the Compass MCP server so it can reason over visual material such as slides, charts and demonstrations, as well as spoken content. The company describes potential uses in meeting records, education and training, sports and media archives, and legal or compliance review. These are examples of possible applications, not reported customer deployments.

Cohere reports evaluating Compass against seven video embedding models across 11 datasets: six public benchmarks and five internal enterprise video datasets. The evaluation covered 11,133 queries and used nDCG@10, a measure of how well the top ten results are ranked. Cohere says Compass ranked first on average, 4.4 points ahead of the strongest alternative. It reports leads of 2.4 points over Amazon Nova Multimodal Embeddings on public benchmarks, 5.1 over Gemini Embedding 2 on soccer footage, and 2.0 over Qwen3-VL-Embedding-8B on enterprise knowledge video.

Video Intelligence is not described as generally available. Cohere is recruiting design partners to test it with real workloads. Participants can deploy in Cohere’s managed cloud or their own environment, evaluate performance on their footage and queries, work with Cohere’s engineering team and help shape the roadmap. The announcement gives no public price or general access date.

What the archive adds

New Runtime’s earlier coverage described AWS’s approach to conversational video intelligence as routing questions among specialized video, speech and foundation-model services (September 25, 2026 note). This Cohere announcement is a new product preview: it centers on bringing timestamped video retrieval into Compass’s existing document-search surface.

Trend signal

This supports the pattern of agent-ready software exposing composable capabilities: video search is surfaced through Compass retrieval and an MCP connection, with permissions and source evidence attached. The broader pattern is described in Agent-ready software exposes capabilities.

Why now

Cohere is currently recruiting design partners, who can test the preview against their own footage and queries and work directly with its engineering team. That makes this an evaluation opportunity for organizations with substantial video collections, rather than a generally available release for all Compass users.

Why it matters

Enterprise video is often an expensive-to-search knowledge store: teams may know that a decision, explanation or disclosure exists but still have to scrub through recordings to find and verify it. Timestamped retrieval could reduce that friction and let agents cite a bounded piece of evidence rather than summarize an entire recording. Combining spoken and visual information also matters when the answer depends on a chart, slide or sequence of events, not just a transcript.

For builders, the useful design choice is to keep retrieval verifiable. A timestamp, transcript, visual description and link to the original give reviewers a way to inspect what an agent used. Permission-aware access is equally important when video contains internal meetings or regulated information. These are product claims and evaluation results from Cohere, however; teams should validate retrieval quality and authorization behavior on their own collections before relying on it in consequential workflows.

Limits and uncertainty

The announcement describes a preview and a design-partner program, not broad availability. Cohere’s performance figures come from its own evaluation; the announcement does not establish independent replication or disclose enough detail here to assess all benchmark choices, model configurations or scoring variance. Results on the listed datasets may not predict performance on a particular company’s recordings, accents, camera setups or queries. The announcement also does not give pricing, a general launch date, or quantified latency and operating costs. Permission-aware retrieval is a stated capability, but its effectiveness in specific deployment environments remains to be tested by customers.

Reader takeaway

Cohere’s preview treats video as searchable enterprise evidence, not merely an attachment or a transcript. Its practical promise is precise, permission-aware retrieval that an agent can return with a path back to the source. The immediate next step is evaluation through Cohere’s design-partner program; the performance claims should be tested against the recordings and access rules a team actually has.

Evidence / sources

  1. 1cohere.com/blog/video-intelligence-preview
All Field Notes

Next reads

Read next

Open the graph
  1. 01topicGeminiExplore the gemini topic hub.
  2. 02archiveField NotesOpen the latest editorial analysis.
  3. 03sourcesSourcesSee the original sources this archive cites.
Who read this page?Machine requests, hidden until opened

Loading the privacy-safe route aggregate...

Open the JSON contract