Zum Inhalt springen
← Back to English edition

Installing OpenClaw – Part 3: Configuring Models (OpenAI, Anthropic, OpenRouter)

• openclaw

OpenClaw supports many AI models – how do you sensibly set up OpenAI, Anthropic, and OpenRouter, manage API keys, and define fallbacks? Here's the practical guide.

Tutorial OpenClaw Models OpenAI Anthropic OpenRouter API Keys

📚 Series: Installing & Setting Up OpenClaw — Part 3 of 8
← Part 2: Installation | Part 4: Connecting Telegram & WhatsApp →

In the second part, we installed OpenClaw and started the gateway. Now it’s time to connect AI models to OpenClaw – because without models, your agent can’t generate responses.

OpenClaw supports many providers: OpenAI, Anthropic, OpenRouter, Google Gemini, DeepSeek, and even local LLMs via Ollama. In this part, you’ll learn how to securely manage API keys, configure providers, and set up fallbacks sensibly.

Prerequisites

✅ Running OpenClaw Gateway (see Part 2)
✅ API keys for desired providers (e.g., OpenAI, Anthropic, OpenRouter)
✅ Basic understanding of API keys and environment variables

Understanding the Configuration File

OpenClaw uses a configuration file (typically ~/.openclaw/config.yaml or ~/.openclaw/openclaw.json). There you define:

  • Providers (which models do you use?)
  • Models per provider (e.g., gpt-5.4, claude-3.7-sonnet, deepseek-v3.2)
  • If a provider is unavailable, it switches to a backup provider.

Example: Minimal Config

providers:
  openai:
    enabled: true
    apiKey: "{{secrets.openai.apiKey}}"
    models:
      - gpt-5.4
      - gpt-5-mini
  anthropic:
    enabled: true
    apiKey: "{{secrets.anthropic.apiKey}}"
    models:
      - claude-3.7-sonnet
  openrouter:
    enabled: true
    apiKey: "{{secrets.openrouter.apiKey}}"
    models:
      - deepseek-v3.2
      - gemini-3.1-flash

💡 Tip: Use {{secrets.<name>}} instead of plain text – the secret value lies outside the versioned config.

Storing API Keys Securely

Option A: Environment Variables (Simple)

export OPENAI_API_KEY="sk-..."
export ANTHROPIC_API_KEY="sk-..."
export OPENROUTER_API_KEY="sk-..."

Then in the config:

apiKey: "{{env.OPENAI_API_KEY}}"

Create ~/.openclaw/secrets.json (or similar) and load it via SecretRef in the config:

apiKey: "{{secrets.openai.apiKey}}"

Important: Do not commit this file to Git repos!

Providers in Detail

OpenAI

  • Models: GPT-5.4, GPT-5-Mini, GPT-5-Nano
  • Advantage: High quality, many features (functions, vision)
  • Disadvantage: More expensive than many alternatives

Anthropic

  • Models: Claude 3.7 Sonnet, Claude 4.6
  • Advantage: Very good language quality, good context handling
  • Disadvantage: Fewer features than OpenAI, somewhat more expensive

OpenRouter

  • Models: DeepSeek V3.2, Gemini 3.1 Flash, GLM-Flash, Qwen, etc.
  • Advantage: Affordable, many models on one platform
  • Disadvantage: Quality varies depending on the model

Defining Fallbacks

You can specify which model is used by default and what to switch to if a provider is unavailable:

defaults:
  model: "openai:gpt-5.4"
  fallback:
    - "anthropic:claude-3.7-sonnet"
    - "openrouter:deepseek-v3.2"

This ensures your agent can always respond – even if a provider is temporarily unreachable.

Testing

After updating the config, restart the gateway:

openclaw gateway restart

Then test with:

openclaw status

You should receive a response – if not, check the logs (openclaw gateway logs) or the configuration.

Troubleshooting

  • “API key invalid”: Check if the API key is correct and not expired.
  • “Model not found”: Check if the model is available with the provider.
  • “Rate limit exceeded”: Wait a moment or switch to another model/provider.

Sources & Further Links

This article is part of the series “Installing & Setting Up OpenClaw”. The next part will be published soon. If you have questions or suggestions, feel free to write to me on Mastodon or via email.