Cohere Launches Compass Cloud in Private Beta

The provider packages its internal enterprise search engine into a managed cloud service with native MCP support and retrieval-time permissions.

Cohere has opened access requests for the private beta of Compass Cloud, a fully managed cloud version of its enterprise search and retrieval platform. Announced on the company’s official blog and on X, the service requires approval through an application form and targets teams building autonomous agents and retrieval-augmented generation (RAG) pipelines.

Until now, Compass operated primarily inside North, Cohere’s enterprise agent workspace, or through bespoke self-hosted setups for heavily regulated organizations. The cloud release lowers operational overhead by turning that search stack into a turnkey subscription. Instead of standing up servers, managing vector databases, and stitching together parsing scripts, engineering teams connect their data sources directly to a hosted endpoint.

How Compass Cloud handles retrieval architecture

Building production RAG systems inside large organizations usually turns into a patchwork of disconnected software. Teams have to pull data from messy PDF reports, convert text into mathematical vectors, push those embeddings into databases such as Pinecone or Qdrant, and then configure reranking models to weed out irrelevant context.

Compass Cloud packages that pipeline into a unified platform. It includes out-of-the-box connectors for Google Drive, Microsoft SharePoint, and OneDrive, alongside a multimodal parser that processes tables, charts, and embedded graphics. Under the hood, the system pairs traditional keyword search with semantic embeddings, using Cohere’s Embed and Rerank models to reorder matches before passing context to an application or model.

Retrieval-time permissions and the MCP standard

Data governance often stalls enterprise AI rollouts. Cohere claims Compass Cloud addresses this by applying access controls during the query process itself. In practice, the index evaluates user permissions and filters out unauthorized files before feeding context into prompts, preventing confidential data leaks without relying on downstream application logic.

For autonomous agent workloads, Cohere provides REST APIs, an open-source Python SDK on GitHub, and a dedicated server built on Anthropic’s Model Context Protocol (MCP). The open standard acts as a universal bridge between language models and external tool systems. With native MCP support, agent frameworks can query internal company repositories across multi-step research tasks without custom integrations.

Internal benchmarks and missing pricing details

Alongside the rollout, Cohere published figures from “High Finance”, an internal benchmark designed around investment banking queries. The company reported that Compass scored 81.1 on nDCG@10 (a metric evaluating the relevance and ranking of top-ten results), compared to 64.8 for Azure Search. Those figures stem entirely from Cohere’s internal methodology and have not been audited by third-party evaluators.

The company has not disclosed pricing tiers, query rate limits, or an official date for general availability. For organizations operating under strict compliance mandates that forbid cloud data transfers, Cohere confirmed that on-premises and Virtual Private Cloud (VPC) deployments remain supported.

Sources

• Cohere Blog: Introducing Compass Cloud in Private Beta
• Cohere: Compass Product Overview and Access Form
• Official Cohere Announcement on X
• GitHub: Cohere Compass SDK