ModelsConfiguration

OpenClaw Model Setup: OpenAI, Anthropic, Ollama

Basic guide to configuring models in OpenClaw. Links to official docs for complete provider documentation.

3 min readLast updated Feb 18, 2026

Stuck?Check the troubleshooting index or ask in Discord.

Overview

OpenClaw supports multiple model providers including OpenAI, Anthropic, Google, Ollama, and more. This guide covers how to configure models, set up fallbacks, and optimize for your use case.

Prerequisites

You should have OpenClaw installed and a model backend (Ollama, vLLM) or API keys ready.

Quick Start

The fastest way to get started is setting environment variables:

.env

# Option 1: OpenAI
OPENAI_API_KEY=sk-...

# Option 2: Anthropic
ANTHROPIC_API_KEY=sk-ant-...

# Option 3: Ollama (local)
MODEL_BACKEND_URL=http://localhost:11434

Then set your model in config:

config.yaml

agents:
  defaults:
    model:
      primary: openai/gpt-5

Model Providers

OpenClaw supports many providers. Here's how to configure each:

OpenAI

env

OPENAI_API_KEY=sk-...
MODEL_BACKEND_URL=https://api.openai.com/v1

Anthropic

env

ANTHROPIC_API_KEY=sk-ant-...
MODEL_BACKEND_URL=https://api.anthropic.com

Ollama (Local)

env

MODEL_BACKEND_URL=http://localhost:11434

Make sure Ollama is running: ollama serve

Google (Gemini)

env

GOOGLE_API_KEY=your-google-api-key
MODEL_BACKEND_URL=https://generativelanguage.googleapis.com/v1

Azure OpenAI

env

AZURE_OPENAI_API_KEY=your-key
AZURE_OPENAI_ENDPOINT=https://your-resource.openai.azure.com/
AZURE_OPENAI_DEPLOYMENT=gpt-4

Configuration Options

Full model configuration in your config file:

config.yaml

agents:
  defaults:
    model:
      primary: anthropic/claude-sonnet-4-5
      imageModel:
        primary: openai/gpt-5o
      thinkingDefault: "low"  # off | low | medium | high
      verboseDefault: "off"   # off | on
      elevatedDefault: "on"   # on | off
      timeoutSeconds: 600
      mediaMaxMb: 5
      contextTokens: 200000
      maxConcurrent: 3

Fallback Models

Set fallback models in case your primary fails:

config.yaml

agents:
  defaults:
    model:
      primary: anthropic/claude-opus-4-6
      fallbacks:
        - anthropic/claude-sonnet-4-5
        - openai/gpt-5
        - google/gemini-2-pro

How fallbacks work

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

Model Aliases

Create shortcuts for commonly used models:

config.yaml

agents:
  defaults:
    models:
      "opus":
        provider: anthropic/claude-opus-4-6
      "sonnet":
        provider: anthropic/claude-sonnet-4-5
      "gpt":
        provider: openai/gpt-5
      "mini":
        provider: openai/gpt-5-mini

Now you can switch models with /model opus or /model gpt

Context Windows & Tuning

Adjust context settings for your needs:

Context Tokens

Set the maximum context window. Higher = more memory but more expensive.

yaml

contextTokens: 200000  # Claude 3 Opus max)
contextTokens: 128000  # GPT-4 Turbo max)
contextTokens: 32000   # Smaller models)

Temperature

Control randomness (0 = deterministic, 1 = creative).

yaml

models:
  "my-model":
    params:
      temperature: 0.7

Max Tokens

Limit response length.

yaml

models:
  "my-model":
    params:
      maxTokens: 4096

Image Models

Configure a separate model for image analysis (used when primary model doesn't support images):

config.yaml

agents:
  defaults:
    model:
      primary: anthropic/claude-opus-4-6
      imageModel:
        primary: openrouter/qwen/qwen-2.5-vl-72b-instruct:free
        fallbacks:
          - openrouter/google/gemini-2.0-flash-vision:free

Free image models

OpenRouter offers several free vision models that work well for basic image analysis.