Configuring OpenClaw
Learn how to configure AI models, channels, and system settings.
Beginner10 min
configurationai-modelssetup
Configuring OpenClaw
⚙️ This guide covers the essential configuration options for OpenClaw.
Configuration File Structure
OpenClaw uses a config.yaml file in the project root for all settings.
# config.yaml
gateway:
# Session settings
session:
timeout: 3600 # Session timeout in seconds
max_history: 50 # Max messages to remember
# Channel configurations
channels:
- name: wechat
enabled: true
webhook_port: 8080
- name: telegram
enabled: false
bot_token: ${TELEGRAM_BOT_TOKEN}
# AI Model settings
models:
default: claude
providers:
- name: claude
model: claude-3-sonnet-20240229
api_key: ${ANTHROPIC_API_KEY}
- name: openai
model: gpt-4-turbo-preview
api_key: ${OPENAI_API_KEY}
Configuring AI Models
Using Claude (Anthropic)
- Get your API key from Anthropic Console
- Add to .env:
ANTHROPIC_API_KEY=sk-ant-xxxxx
- Set as default in config:
models:
default: claude
Using ChatGPT (OpenAI)
- Get your API key from OpenAI Platform
- Add to .env:
OPENAI_API_KEY=sk-xxxxx
- Set as default in config:
models:
default: openai
Using Other Models
OpenClaw supports various AI providers:
| Provider | Model | Notes | |----------|-------|-------| | Anthropic | Claude 3 | Recommended for best results | | OpenAI | GPT-4 | Reliable alternative | | DeepSeek | DeepSeek Chat | Cost-effective | | Replicate | Various | For image models |
Configuring Channels
channels:
- name: wechat
enabled: true
webhook_port: 8080
See Connect WeChat for detailed setup.
Telegram
channels:
- name: telegram
enabled: true
bot_token: ${TELEGRAM_BOT_TOKEN}
See Connect Telegram for detailed setup.
System Prompts
Customize how OpenClaw responds:
prompts:
system: |
You are a helpful AI assistant.
Be concise and friendly.
Always offer to help with follow-up questions.
# Per-channel overrides
channels:
wechat: |
You are assisting via WeChat.
Keep responses short and mobile-friendly.
Testing Configuration
After making changes:
npm run dev
Check the console for any configuration errors.