Skip to content

config-setup.md

Latest commit

 

History

History
71 lines (53 loc) · 1.74 KB

config-setup.md

File metadata and controls

71 lines (53 loc) · 1.74 KB
Raw

Config Setup

API key and model configuration for the orchestrator. Key resolution order: environment variables first, then ~/.claude/llm-config.json as fallback.

Option 1: Environment Variables (Recommended)

Set in your shell profile (~/.zshrc, ~/.bashrc, etc.):

export ANTHROPIC_API_KEY="sk-ant-..."
export OPENAI_API_KEY="sk-..."
export GOOGLE_AI_API_KEY="AIza..."

The script maps these to the correct provider automatically. No config file needed for default models.

After adding keys, restart your terminal or source ~/.zshrc for changes to take effect.

Option 2: Config File (~/.claude/llm-config.json)

For custom model names, non-default model IDs, or role assignments:

{
  "models": {
    "opus": {
      "provider": "anthropic",
      "model": "claude-opus-4-6"
    },
    "gpt5": {
      "provider": "openai",
      "model": "gpt-5.4"
    },
    "gemini": {
      "provider": "google",
      "model": "gemini-2.0-pro"
    }
  },
  "defaults": {
    "reasoning": "opus",
    "creative": "gpt5",
    "validation": "gemini"
  }
}

If both env var and config file exist for the same provider, the env var wins.

Adding a New Model

Add to models in the config file:

"gpt5-pro": {
  "provider": "openai",
  "model": "gpt-5.4-pro"
}

Optionally assign a default role in defaults. Run python scripts/orchestrate.py check to verify.

Adding a New Provider

See references/api-patterns.md for the implementation steps (adding to ENV_KEY_MAP, call_{provider}(), and PROVIDER_CALLERS). Then document the model in references/model-profiles.md.

Verifying Setup

python scripts/orchestrate.py check

Shows each model's status, key source (env/config/none), and role assignments.