Cohere launches Compass Cloud, a managed retrieval platform, in private beta

From North and self-hosted setups to a managed service

Compass is Cohere's retrieval platform. According to the company, it surfaces the most relevant information from a company's corpus for retrieval-augmented generation (RAG), search and agentic workflows. Until now, Compass has mainly powered retrieval for North, Cohere's enterprise agent workspace, including its document libraries and MCP ecosystem. Cohere has also shipped it into highly secure, self-hosted environments for partners in regulated industries whose workloads cannot be offloaded to SaaS.

Cohere says customer demand for a managed option was "clear and consistent": teams want Compass' retrieval capabilities but many do not want the operational overhead of self-hosting. Compass Cloud lets Cohere manage the full pipeline and model inference. Self-hosted deployments remain available for privacy-constrained projects.

Teams access Compass through its APIs, MCP server or Python SDK. Cohere is working with a limited number of enterprise teams as beta partners, and interested companies can request access.

The problems Compass targets

Cohere argues that enterprise retrieval is no longer judged only on relevance and latency for a single query. It names three developments that change the requirements:

• Token economics: every irrelevant result passed to a model consumes tokens and context space. More precise retrieval means smaller, higher-quality inputs, lower inference costs and faster task completion.
• Agentic access patterns: agents may issue dozens of queries for one task. In these multi-hop loops latency accumulates, relevance can drift and permissions must be enforced at every step.
• A fragmented retrieval stack: production pipelines often combine separate systems for ingestion, indexing, reranking, access control and orchestration, leaving teams to spend effort on integration rather than retrieval quality.

Compass is meant to consolidate that stack into one configurable service with a single interface.

What the pipeline covers

Cohere describes the pipeline in seven steps:

• Connect: out-of-the-box connectors for workspaces such as SharePoint, OneDrive and Google Drive, with multilingual, multimodal and file format-agnostic support.
• Parse: complex documents become structured, searchable data, with vision processing used only where it adds value.
• Embed: dense and sparse representations are generated together, so search can match both semantic intent and domain-specific language.
• Index: source files, parsed content and embeddings are kept as separate records, so a new embedding model can be adopted without re-crawling and re-uploading content.
• Retrieve: semantic, sparse and keyword search can run independently or together in one request, with permissions enforced during retrieval.
• Rerank: Cohere's reranker picks the most relevant passages from a broad candidate set, cutting irrelevant context and the tokens needed for generation.
• Govern: multi-tenant access control and document-level permissions are enforced during retrieval. Retention policies expire content automatically and prevent deleted documents from being resynced.

Benchmark claim and integration options

Cohere reports an accuracy gain on High Finance, an internal, Cohere-annotated benchmark in which models retrieve relevant investment banking and hedge fund presentation material. According to the company, Compass improved on Azure Search by 14-16 points, from 64.8 to 81.1 (nDCG@10). In that test, Azure Search documents were parsed with GPT-4.1 Mini, while Cohere Embed 4 embedded the input PDFs directly as images without a separate parsing solution. The result comes from Cohere's own benchmark and has not been independently verified.

Developers can integrate in two ways. The Compass API suits teams building custom RAG and search applications and gives more control over how content is represented, retrieved, ranked and used. The Model Context Protocol (MCP) route is aimed at agentic retrieval: Compass ships a dedicated, independently versioned MCP server that exposes retrieval as reusable tools through an open standard. MCP-compatible clients can discover and call them without a bespoke integration for each agent framework. Cohere says agents can use it to progressively narrow the retrieval corpus, which reduces unnecessary context and makes complex retrieval tasks faster and more token-efficient.

What the beta offers

The private beta is open to teams building retrieval-heavy and agentic applications. Cohere says participants get hands-on support from its engineering team, early access to the cloud deployment and direct influence on the roadmap. The company also plans a live session on X on October 8th with its engineering and product leadership on where enterprise search goes next.

For enterprises, the move offers a way to use Compass without running the retrieval infrastructure themselves, while regulated organizations can stay on self-hosted deployments.