Cohere

Cohere is an LLM platform for enterprises: Command models, Embed and Rerank for retrieval, and the North agent workspace, sold by usage or private deployment.

Cohere puts language models to work on a company's own documents and systems while keeping that data private. Cohere Inc., headquartered in Toronto, builds the models and runs the platform itself. One account reaches generation, retrieval, document and speech models, plus an agent workspace for staff. The same models run on a shared API, on dedicated hosting or inside a private deployment, so security teams decide where data goes. The Command family handles generation. Command A+ is released as open weights under the Apache 2.0 license, and Command A Vision reads images in a prompt. Command R is the midsized model, while Command R7B is made for commodity GPUs and edge devices. Developers get tool use with citations, Structured Outputs that follow a JSON Schema, and a reasoning mode that can be switched on or off. Embed turns text and images into vectors, Rerank reorders search results by relevance, and Parse converts scanned documents, tables and handwriting into structured content. Transcribe is a separate speech model, and the Aya Expanse research models are also on the API. North is the workspace where employees use AI agents, North Mini Code is an open-source agentic coding model, and Model Vault hosts models on dedicated instances with no shared resources. Cohere is built for enterprises that must keep AI inside controlled infrastructure, including regulated industries and the public sector. Developers can try the API on a Trial API key created at signup. Production use is pay as you go, billed by tokens, searches or pages and invoiced monthly. Teams that need guaranteed capacity rent Model Vault instances on monthly or annual commitments. North is sold through a demo with the sales team, and private deployments are priced on request. Typical jobs are retrieval-augmented generation over company documents, enterprise search, support automation and agents that call internal tools. A search team might pair Embed and Rerank with an index it already keeps in Elastic, OpenSearch or Pinecone, then pass the best passages to Command for a cited answer. Companies committed to one cloud can also reach the models through Amazon Bedrock, Microsoft Azure AI Foundry or Oracle Cloud Infrastructure. Document teams run Parse first, so scanned forms and tables become clean text before any model reads them. Cohere fits organizations that want models, retrieval and an agent workspace from one vendor they can deploy privately.