Compass is coming to the cloud | Cohere
Cohere announced Compass Cloud, a private‑beta managed retrieval service that handles the full pipeline and model inference for enterprise RAG, search, and agentic workflows. Developers can now access the platform via APIs, MCP server or Python SDK without self‑hosting, while self‑hosted options remain for privacy‑sensitive use cases.

Cohere has moved its Compass retrieval platform into a private‑beta managed service called Compass Cloud. The new offering lets developers reach the full pipeline and model inference through APIs, an MCP server or a Python SDK, eliminating the need to self‑host the stack. At the same time, Cohere continues to support self‑hosted deployments for projects with strict privacy requirements. Compass is designed to supply the most relevant pieces of a company’s corpus for retrieval‑augmented generation, search and agentic workflows, which can dramatically cut the number of tokens passed to downstream models. By consolidating ingestion, indexing, reranking and access control into a single configurable service, the platform reduces integration effort and helps keep latency low even when agents issue many queries in a single task. In a benchmark that mimics a financial‑industry RAG workload, Compass delivered a 14‑16‑point accuracy gain over a conventional Azure Search setup. It remains uncertain how the managed service will perform at scale across diverse enterprise data sets, and Cohere has not disclosed any formal metrics beyond the cited benchmark. The long‑term cost advantages of the managed option versus self‑hosting have not been quantified, and the timeline for broader availability beyond the limited beta partners is still unclear.