LLM Provider Setup

OpenClaw is model-agnostic -- you can plug in any LLM and switch providers without rewriting your agent. This tutorial covers the most popular options.

Supported Providers#

OpenClaw ships with built-in support for 12+ providers:

| Provider | Type | Best For | Cost | |----------|------|----------|------| | Anthropic (Claude) | Cloud | Best reasoning, complex tasks | $3-75/MTok | | OpenAI (GPT) | Cloud | Speed, wide compatibility | $2-60/MTok | | Google (Gemini) | Cloud | Large context windows | $1-20/MTok | | OpenRouter | Cloud | Access to many models via one API | Varies | | Ollama | Local | Privacy, no API costs | Free (hardware costs) | | vLLM | Local | High-throughput inference | Free (hardware costs) | | Amazon Bedrock | Cloud | AWS integration | Varies |

Quick Setup#

The fastest way to start is with environment variables:

# For Anthropic (recommended)
export ANTHROPIC_API_KEY=sk-ant-...

# For OpenAI
export OPENAI_API_KEY=sk-...

# For Ollama (local)
export MODEL_BACKEND_URL=http://localhost:11434

Then run openclaw onboard -- the wizard detects your keys and configures the provider automatically.

Anthropic Claude (Recommended)#

Claude is recommended for OpenClaw because it has the best reasoning capabilities for complex architectural tasks and tool use.

Configuration#

Edit ~/.openclaw/openclaw.json5:

{
  "models": {
    "defaults": {
      "provider": "anthropic",
      "model": "claude-sonnet-4-20250514"
    },
    "providers": {
      "anthropic": {
        "apiKey": "$ANTHROPIC_API_KEY"
      }
    }
  }
}

## OpenAI GPT

Configuration#

openclaw config set llm.provider "openai"
openclaw config set llm.openai.apiKey "sk-xxxxxxxxxxxx"

Or via config file:

{
  "models": {
    "defaults": {
      "provider": "openai",
      "model": "gpt-5.1-codex"
    },
    "providers": {
      "openai": {
        "apiKey": "$OPENAI_API_KEY"
      }
    }
  }
}

Ollama (Local Models)#

Ollama lets you run open-source models locally -- no API costs, full privacy.

Step 1: Install Ollama#

# macOS/Linux
curl -fsSL https://ollama.ai/install.sh | sh

# Pull a model
ollama pull llama3:70b

Step 2: Configure OpenClaw#

{
  "models": {
    "defaults": {
      "provider": "ollama",
      "model": "llama3:70b"
    },
    "providers": {
      "ollama": {
        "baseUrl": "http://localhost:11434",
        "apiKey": "ollama-local"
      }
    }
  }
}

## Multi-Model Fallback

Configure a primary model with automatic fallbacks:

{
  "models": {
    "defaults": {
      "provider": "anthropic",
      "model": "claude-sonnet-4-20250514"
    },
    "fallbacks": [
      { "provider": "openai", "model": "gpt-5" },
      { "provider": "ollama", "model": "llama3:70b" }
    ]
  }
}

If the primary model fails (rate limit, error, timeout), OpenClaw automatically tries the next model in the list.

Per-Channel Model Routing#

Route different channels to different providers:

{
  "channels": {
    "whatsapp": {
      "model": { "provider": "anthropic", "model": "claude-sonnet-4-20250514" }
    },
    "discord": {
      "model": { "provider": "ollama", "model": "llama3:70b" }
    }
  }
}

This lets you use a powerful cloud model for important conversations and a free local model for casual channels.

Cost Management#

Protect your API budget:

openclaw config set llm.limits.dailySpend 10

When the limit is reached, OpenClaw can fall back to a cheaper model or pause until the next day.

Choosing the Right Provider#

Four factors to consider:

1. Budget -- Cloud models cost $0.50-75 per million tokens. Local models are free but need hardware.
2. Hardware -- Local models need a GPU with 8GB+ VRAM for decent performance.
3. Privacy -- If data cannot leave your network, use Ollama or vLLM.
4. Task complexity -- Use the strongest model for tool-enabled agents. Older or smaller models are less robust against prompt injection.

Next Steps#

• Build your first agent -- Put your LLM to work
• LLM providers deep-dive -- Detailed comparison of all providers