DeepSeek Harness

Configuration

Install DeepSeek Harness and Node.js 20 or later. Create an AudioLab workspace key with llm.write, available credits, and access to the selected model. Export it as AUDIOLAB_API_KEY in the terminal that launches the agent.

{
  llm-pi-ai:
    {
      providers:
        {
          audiolab:
            {
              apiKeyEnv: AUDIOLAB_API_KEY,
              api: openai-completions,
              baseURL: https://api.tryaudiolab.ai/v1,
              models:
                [
                  {
                      id: openai/gpt-6-astra,
                      name: AudioLab,
                      contextWindow: 128000,
                      maxTokens: 4096,
                      input: [ text ]
                    }
                ]
            }
        }
    }
}

Automatic Setup

The command previews the files and fields it changes. Confirm to apply; it stores a private backup and prints an undo command. It does not store the API key or change your shell profile. Automatic file setup supports macOS and Linux.

npx https://tryaudiolab.ai/cli/router-0.1.0.tgz setup --agent deepseek

Select AudioLab in Settings → Models, then select its model for a new session.

Launch

npx https://tryaudiolab.ai/cli/router-0.1.0.tgz run deepseek

Verify the Connection

Send a short text prompt, then a read-only tool request in a disposable project. Check the model, request record, and usage in Router. Test streaming and a follow-up turn before using a longer workflow. Client-specific tools and permissions remain managed by DeepSeek Harness.

The setup command can check authentication and model discovery without generating a response. See the protocol contracts for model and feature limits.

Reference

DeepSeek Harness Configuration — Official Documentation

AudioLab Setup Command