Cohere opened a private beta for Compass Cloud on September 25, 2026, offering its enterprise retrieval platform as a managed service. The service is for teams building AI applications with company data; Cohere manages the retrieval pipeline and model inference, while customers use Compass through APIs, an MCP server, or a Python SDK. Self-hosted deployments remain available for projects with strict privacy requirements. Beta access is limited to a number of enterprise teams, which can receive engineering support and give direct input on the product roadmap.

Cohere said Compass has mainly powered retrieval for North, its enterprise agent workspace, and has also been deployed in self-hosted environments for regulated-industry partners. The company said customers want its retrieval capabilities without the operational work of running the system themselves. Compass Cloud packages document processing and retrieval in one configurable service, connecting data sources to applications through parsing, embedding, search, reranking, and access controls.

Cohere framed three pressures on enterprise search: irrelevant results consume model tokens and context space; agents can issue dozens of queries per task, causing latency to accumulate and relevance to drift; and production systems often require separate tools for ingestion, indexing, reranking, access control, and orchestration. Compass is designed to provide governed context, support agentic and conventional search, and consolidate those retrieval functions. Its connectors include SharePoint, OneDrive, and Google Drive. The service parses complex files, creates dense and sparse representations, keeps source files, parsed content, and embeddings as separate records, and applies permissions during retrieval. Retention policies automatically expire content and prevent deleted documents from being synced again.

On Cohere’s High Finance benchmark, an internally annotated set of questions about investment-banking and hedge-fund presentations, Compass scored 81.1 versus 64.8 for Azure Search, a 14–16 point improvement as reported by Cohere. The company describes the result as retrieval accuracy measured by nDCG@10 on a representative financial-industry retrieval-augmented generation workload. In the comparison, Azure Search documents were parsed with GPT-4.1 Mini; Cohere Embed 4 embedded input PDFs directly as images without a separate parsing system.

Teams can use Compass APIs to build custom search and retrieval-augmented generation applications, or connect through Compass’s independently versioned Model Context Protocol server. The server exposes retrieval tools that compatible clients can discover and call across agent frameworks. Cohere says agents can refine searches over multiple steps to narrow the available information and reduce unnecessary context. The company plans a live session with engineering and product leaders on October 8, 2026, about the future of enterprise search.

Cohere opened a private beta for Compass Cloud on September 25, 2026, offering its enterprise retrieval platform as a managed service. The service is for teams building AI applications with company data; Cohere manages the retrieval pipeline and model inference, while customers use Compass through APIs, an MCP server, or a Python SDK. Self-hosted deployments remain available for projects with strict privacy requirements. Beta access is limited to a number of enterprise teams, which can receive engineering support and give direct input on the product roadmap.

Cohere said Compass has mainly powered retrieval for North, its enterprise agent workspace, and has also been deployed in self-hosted environments for regulated-industry partners. The company said customers want its retrieval capabilities without the operational work of running the system themselves. Compass Cloud packages document processing and retrieval in one configurable service, connecting data sources to applications through parsing, embedding, search, reranking, and access controls.

Cohere framed three pressures on enterprise search: irrelevant results consume model tokens and context space; agents can issue dozens of queries per task, causing latency to accumulate and relevance to drift; and production systems often require separate tools for ingestion, indexing, reranking, access control, and orchestration. Compass is designed to provide governed context, support agentic and conventional search, and consolidate those retrieval functions. Its connectors include SharePoint, OneDrive, and Google Drive. The service parses complex files, creates dense and sparse representations, keeps source files, parsed content, and embeddings as separate records, and applies permissions during retrieval. Retention policies automatically expire content and prevent deleted documents from being synced again.

On Cohere’s High Finance benchmark, an internally annotated set of questions about investment-banking and hedge-fund presentations, Compass scored 81.1 versus 64.8 for Azure Search, a 14–16 point improvement as reported by Cohere. The company describes the result as retrieval accuracy measured by nDCG@10 on a representative financial-industry retrieval-augmented generation workload. In the comparison, Azure Search documents were parsed with GPT-4.1 Mini; Cohere Embed 4 embedded input PDFs directly as images without a separate parsing system.

Teams can use Compass APIs to build custom search and retrieval-augmented generation applications, or connect through Compass’s independently versioned Model Context Protocol server. The server exposes retrieval tools that compatible clients can discover and call across agent frameworks. Cohere says agents can refine searches over multiple steps to narrow the available information and reduce unnecessary context. The company plans a live session with engineering and product leaders on October 8, 2026, about the future of enterprise search.