Groq Provider Setup: API Key, Models, and Whisper Transcription

Groq delivers extremely fast inference for open-weight models such as Llama, Gemma, Kimi, Qwen, GPT OSS, and others, relying on custom LPU hardware. Through its plugin, Groq exposes both a chat provider that follows the OpenAI API format and an audio media-understanding provider.

Install plugin

After installing the official plugin, restart Gateway:

openclaw plugins install @openclaw/groq-provider
openclaw gateway restart

Getting started

Get an API key

Head to console.groq.com/keys to generate an API key.

Set the API key

export GROQ_API_KEY=gsk_...

Set a default model

{
  agents: {
    defaults: {
      model: { primary: "groq/openai/gpt-oss-120b" },
    },
  },
}

Verify the catalog is reachable

openclaw models list --provider groq

Config file example

{
  env: { vars: { GROQ_API_KEY: "gsk_..." } },
  agents: {
    defaults: {
      model: { primary: "groq/openai/gpt-oss-120b" },
    },
  },
}

Built-in catalog

OpenClaw includes a manifest-driven Groq catalog featuring both reasoning and non-reasoning entries. Execute openclaw models list --provider groq to list the static rows for your current version, or consult console.groq.com/docs/models for Groq's official model list.

The manifest also keeps groq/llama-3.1-8b-instant and groq/llama-3.3-70b-versatile as hidden deprecated compatibility rows, remaining until Groq's shutdown on August 16, 2026. For fresh setups, use groq/openai/gpt-oss-20b and groq/openai/gpt-oss-120b respectively.

Tip

Every OpenClaw release updates the catalog. openclaw models list --provider groq displays the rows your installed version recognizes; verify against console.groq.com/docs/models for any newly added or retired models.

Reasoning models

Groq's reasoning models (marked reasoning: true above) translate OpenClaw's shared /think levels into reasoning_effort values of low, medium, or high. When /think off or /think none is set, reasoning_effort gets omitted from the request instead of being sent as a disabled value.

For the shared /think levels and how OpenClaw maps them for each provider, see Thinking modes.

Audio transcription

Voice messages can be transcribed through the shared tools.media.audio interface because the Groq plugin also sets up an audio media-understanding provider.

To set Groq as the default audio backend:

{
  tools: {
    media: {
      models: [{ provider: "groq", capabilities: ["audio"] }],
    },
  },
}

Environment availability for the daemon

When Gateway operates as a managed service (launchd, systemd, Docker), GROQ_API_KEY needs to be accessible to that process, not merely to your interactive shell.

Warning

A key exported only in an interactive shell won't assist a launchd or systemd daemon unless that environment is imported there as well. Place the key in ~/.openclaw/.env or use env.shellEnv so the gateway process can read it.

Custom Groq model ids

At runtime, OpenClaw accepts any Groq model id. Use the exact id Groq provides and prepend it with groq/. The static catalog handles typical scenarios, while uncatalogued ids fall back to the default OpenAI-compatible template.

{
  agents: {
    defaults: {
      model: { primary: "groq/<your-model-id>" },
    },
  },
}

Related

• Model providers, How to pick providers, reference models, and handle failover.

• Thinking modes, Reasoning intensity settings and how they interact with provider policies.

• Configuration reference, The complete configuration schema, covering provider and audio options.

• Groq Console, Groq's dashboard, API documentation, and pricing details.