Cohere shifts to enterprise retrieval

Search and retrieval are the bridge between a chatbot that answers questions and a worker agent that can actually complete a job. For Cohere, that means moving from selling raw model access to selling a system that can look across a company’s documents, pull the right facts, reason across multiple sources, and then take actions like drafting reports or handling workflow tasks inside secure enterprise environments.

• Cohere already had the core building blocks early, with Embed and Rerank products for semantic search, then extended that into North, launched in January 2025, which connects enterprise software and handles longer research tasks. The strategic shift is from model endpoint to full enterprise work layer.
  2 sacra 3 sacra
• In practice, retrieval means getting the actual text a model needs, not just links. Cohere moved from Brave to Tavily because agent systems need grounded passages they can use immediately, and its current priority is strong retrieval over internal company data, where big customers have scattered files across SharePoint, financial systems, and custom knowledge bases.
  1 sacra
• This also changes who Cohere competes with. Better retrieval pushes it closer to Glean and Writer, while search infrastructure vendors like Exa and Tavily sit one layer below as inputs. The value is in making the model more useful on enterprise work, not in becoming a general web search engine.
  1 sacra 2 sacra 4 sacra

The next step is domain specific retrieval and action taking. As Cohere improves access to internal documents, financial filings, legal sources, and other specialized corpora, North can evolve from answering with grounded text to proactively assembling reports, suggesting missing analyses, and eventually handling routine business processes with much less human supervision.