Cohere Provider Setup: Auth and Model Selection

Cohere offers OpenAI-compatible inference via its Compatibility API. An official external plugin from OpenClaw makes Cohere available.

Built-in catalog

Two reasoning modes are available through the Compatibility API for Cohere models that support reasoning. none is what OpenClaw uses for off, while any enabled thinking level maps to high. Because Command A Vision cannot use tools, OpenClaw leaves agent tools turned off for it.

Get started

1. Add the official plugin:

openclaw plugins install @openclaw/cohere-provider

When a Gateway is running, the install is applied to it automatically; if not, it takes effect at the next startup. Refer to Apply changes and inspect.

2. Generate a Cohere API key.
3. Start onboarding:

openclaw onboard --non-interactive --accept-risk --skip-health \
  --auth-choice cohere-api-key \
  --cohere-api-key "$COHERE_API_KEY"

4. Verify that the catalog is available:

openclaw models list --provider cohere

Cohere becomes the primary model through onboarding only if no primary model has been set already.

Your model entries survive onboarding, and generated catalog rows are left to discovery. In addition, with models.mode: "replace" the built-in catalog gets written, since that mode does not run discovery.

Environment-only setup

Expose COHERE_API_KEY to the Gateway process, then pick the Cohere model:

{
  agents: {
    defaults: {
      model: { primary: "cohere/command-a-plus-05-2026" },
    },
  },
}

Note

For a Gateway that runs as a daemon or inside Docker, define COHERE_API_KEY on that service. Setting it in an interactive shell alone will not reach a Gateway that is already running.

Related

• Model providers
• Models CLI
• Provider directory