Cohere

Cohere builds enterprise-grade language models including the Command family. AgentXchain connects via api_proxy using Cohere's OpenAI-compatible Chat API.

Which adapter?

api_proxy with provider: "openai" and a custom base_url.

Prerequisites

• A Cohere API key — get one from dashboard.cohere.com
• COHERE_API_KEY set in your environment
• agentxchain CLI installed

Configuration

{
  "runtimes": {
    "cohere-dev": {
      "type": "api_proxy",
      "provider": "openai",
      "model": "command-a-reasoning",
      "auth_env": "COHERE_API_KEY",
      "base_url": "https://api.cohere.com/compatibility/v1/chat/completions"
    }
  },
  "roles": {
    "dev": {
      "runtime": "cohere-dev",
      "mandate": "Implement features and fix bugs",
      "authority": "proposed"
    }
  }
}

Available models

Model	Best for
command-a-reasoning	Complex reasoning and code architecture
command-a	General-purpose coding
command-r-plus	Retrieval-augmented tasks

Verify the connection

export COHERE_API_KEY="..."
agentxchain connector check

Gotchas

• Compatibility endpoint: Cohere provides an OpenAI-compatible endpoint at /compatibility/v1/. Use this with provider: "openai".
• Enterprise features: Cohere offers enterprise deployment options (VPC, on-premise). For these, update base_url to your deployment endpoint.
• Cost rates: Supply operator-specific rates:

{
  "budget": {
    "cost_rates": {
      "command-a-reasoning": { "input_per_million": 2.50, "output_per_million": 10.00 }
    }
  }
}