The Pulse
Cohere Moves Compass Retrieval Into Private Cloud Beta
Cohere opened Compass Cloud, a managed version of its enterprise search and retrieval platform, to a limited number of beta partners on September 25, 2026. The service combines document processing, search, permissions and model inference, w
AI.info Team ·
“Now, Compass is entering private beta as a managed offering - Compass Cloud.”
Cohere Team, Cohere
Cohere announced the private beta on September 25, offering the managed service to a limited number of enterprise teams. Compass Cloud gives customers access to the company’s search and retrieval system without requiring them to run the underlying infrastructure themselves. Cohere says it will manage the full retrieval pipeline and model inference.
From North’s retrieval system to a managed service
Compass already supports retrieval for North, Cohere’s enterprise agent workspace, and has also been deployed in self-hosted environments for partners with workloads they cannot move to software-as-a-service platforms. Compass Cloud adds a managed option for teams seeking those retrieval tools without operating the stack. Cohere says self-hosted deployments will remain available for privacy-constrained projects.
Teams can use Compass through its APIs, a Python software development kit, or a dedicated Model Context Protocol server. The MCP server exposes retrieval as reusable tools that compatible clients can call, allowing agent frameworks to access company information without a bespoke integration for each one. Cohere is inviting beta partners to request access and says participants will receive engineering support and a chance to influence the product roadmap.
One service for parsing, search and permissions
The service brings together document processing and retrieval, rather than asking customers to assemble separate systems for ingestion, indexing, ranking and access control. Cohere lists connectors for SharePoint, OneDrive and Google Drive, alongside support for multilingual, multimodal and varied-format content. Its processing pipeline turns complex documents into structured content and generates both dense and sparse representations for search.
Compass can combine semantic, sparse and keyword search, then rerank candidate passages before sending results to a generative model. Cohere says it enforces document-level permissions during retrieval, rather than leaving each application to filter results, and that retention policies can expire content and prevent deleted documents from being synced again. The company presents those controls as part of the retrieval service, not as a separate layer customers must build around it.
Cohere’s benchmark comes with a methodology caveat
Cohere reports that Compass scored 81.1 on nDCG@10 for its High Finance benchmark, compared with 64.8 for Azure Search. The company describes the test as an internal, Cohere-annotated set of questions about investment-banking and hedge-fund presentation materials; it says the result represents a 14-to-16-point improvement. The figures are vendor-reported, not an independent evaluation.
The comparison also used different document preparation: Cohere says Azure Search’s documents were parsed with GPT-4.1 Mini, while its Embed 4 test embedded input PDFs as images without a separate parsing system. That difference matters when interpreting the result because processing choices can affect what each retrieval system has available to search. Cohere has not published pricing or a general-availability date in its beta announcement, leaving prospective customers with an access request rather than a public launch schedule.