Advanced Configuration
Deep dive into OpenClaw configuration
💡 Chapter Goal: Master OpenClaw's advanced configuration techniques, including Antigravity Manager setup, multi-model switching, cost optimization, and performance tuning.
⚙️ Chapter Contents
- 11.1 Antigravity Manager Complete Configuration Guide
- 11.2 Multi-Model Switching Strategy
- 11.3 Memory Search Configuration
- 11.4 Cost Optimization Solutions
- 11.5 Performance Tuning Tips
- 11.6 Detailed Model Provider Configuration
- 11.7 Detailed Tool System
- 11.8 CLI Command Full Reference
11.1 Antigravity Manager Complete Configuration Guide
11.1.1 What is Antigravity Manager?
Definition:
Antigravity Manager is an AI API proxy tool that allows you to access multiple AI models (Claude, Gemini, GPT, etc.) through a local service, unifying API key and request management.
Project Address: https://github.com/lbjlaq/Antigravity-Manager
Why use Antigravity Manager?
By combining OpenClaw with Antigravity Manager, you can:
- ✅ Local Deployment: All data is processed locally, protecting privacy
- ✅ Unified Management: One tool to manage all AI models
- ✅ Cost Control: Use your own API keys, avoiding intermediary markups
- ✅ Flexible Switching: Switch between different models at any time without modifying code
- ✅ Skill Expansion: Install various practical Skills via ClawHub
11.1.2 System Requirements and Prerequisites
System Requirements:
- macOS 10.15+, Windows 10+, or Linux
- At least 4GB RAM
- Stable network connection
What you need to prepare:
- Antigravity Manager installation package
- API Key for AI models (or an exclusive account)
- Basic command-line operation skills
11.1.3 Install Antigravity Manager
macOS Users
- Visit Antigravity Manager Releases
- Download the latest
.dmgfile - Double-click the
.dmgfile and drag the application to theApplicationsfolder - Open the application (for the first time, you might need to allow it in "System Preferences → Security & Privacy")
Windows Users
- Visit Antigravity Manager Releases
- Download the latest
.exeinstaller package - Run the installer and follow the prompts to complete the installation
- Launch Antigravity Manager
Linux Users
- Visit Antigravity Manager Releases
- Download the latest
.AppImageor.debfile - Grant execution permissions and run:
chmod +x Antigravity-Manager-*.AppImage
./Antigravity-Manager-*.AppImageVerify Installation
After launching, the application will run an API service locally, default address: http://127.0.0.1:8045
Visit this address in your browser. If you can see the management interface, the installation is successful.
11.1.4 Configure AI Model Accounts
Antigravity Manager requires you to provide AI model API keys to function.
Solution 1: Use Official API
Claude API
- Visit Anthropic Console
- Register an account and link a credit card
- Create an API Key
- Copy and save it
Gemini API
- Visit Google AI Studio
- Log in with your Google account
- Create an API Key
- Copy and save it
OpenAI API
- Visit OpenAI Platform
- Register an account and link a credit card
- Create an API Key
- Copy and save it
Solution 2: Purchase Exclusive Account (Recommended)
If you don't want to apply for an API yourself, you can purchase an exclusive account:
Recommended: Student account Gemini 3 Pro exclusive account for 12 months (supports Antigravity)
Advantages:
- ✅ Exclusive account, no need to worry about rate limits
- ✅ Supports Antigravity Manager
- ✅ 12-month validity
- ✅ High cost-effectiveness
- ✅ Ready to use immediately
Configure API Key in Antigravity Manager
- Open the Antigravity Manager management interface
- Click "API Keys"
- Select the corresponding AI service provider (Claude, Gemini, OpenAI)
- Enter the API Key
- Click "Save"
11.1.5 Generate User Token
The User Token is the credential for OpenClaw to access Antigravity Manager.
- In the Antigravity Manager interface, click "User Tokens" in the top right corner
- Click "Create New Token"
- Copy the generated Token (e.g.,
sk-82bc103b51f24af888af525a7835e87c) - ⚠️ Important: Save this Token securely, it will only be displayed once!
11.1.6 Configure OpenClaw
Configure Claude Sonnet 4.5 (Default Model)
This is the most commonly used model, suitable for daily conversations and code generation.
# Add local-anthropic provider
cat ~/.openclaw/openclaw.json | jq '.models.providers["local-anthropic"] = {
"baseUrl": "http://127.0.0.1:8045",
"apiKey": "你的User_Token",
"auth": "api-key",
"api": "anthropic-messages",
"models": [
{
"id": "claude-sonnet-4-5-20250929",
"name": "Local Claude Sonnet 4.5",
"reasoning": false,
"input": ["text"],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 200000,
"maxTokens": 8192
}
]
}' > /tmp/openclaw-temp.json && mv /tmp/openclaw-temp.json ~/.openclaw/openclaw.json
# Set as default model
openclaw config set agents.defaults.model.primary "local-anthropic/claude-sonnet-4-5-20250929"Note: Replace 你的User_Token with the Token generated in step three.
Configure Claude Opus 4.5 Thinking (Reasoning Model)
This is Claude's reasoning model, suitable for complex problems and deep thinking.
cat ~/.openclaw/openclaw.json | jq '.models.providers["local-anthropic-opus"] = {
"baseUrl": "http://127.0.0.1:8045",
"apiKey": "你的User_Token",
"auth": "api-key",
"api": "anthropic-messages",
"models": [
{
"id": "claude-opus-4-5-thinking",
"name": "Local Claude Opus 4.5 Thinking",
"reasoning": true,
"input": ["text"],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 200000,
"maxTokens": 8192
}
]
}' > /tmp/openclaw-temp.json && mv /tmp/openclaw-temp.json ~/.openclaw/openclaw.jsonConfigure Gemini 3 Pro Image (Multimodal Model)
This is Google's multimodal model, supporting image recognition and analysis.
cat ~/.openclaw/openclaw.json | jq '.models.providers["local-google"] = {
"baseUrl": "http://127.0.0.1:8045/v1beta",
"apiKey": "你的User_Token",
"auth": "api-key",
"api": "google-generative-ai",
"models": [
{
"id": "gemini-3-pro-image",
"name": "Local Gemini 3 Pro Image",
"reasoning": false,
"input": ["text", "image"],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 2000000,
"maxTokens": 8192
}
]
}' > /tmp/openclaw-temp.json && mv /tmp/openclaw-temp.json ~/.openclaw/openclaw.json11.1.7 Verify Configuration
Check Model List
openclaw models listYou should see:
Model Input Ctx Local Auth Tags
local-anthropic/claude-sonnet-4-5-20250929 text 195k yes yes default
local-anthropic-opus/claude-opus-4-5-thinking text 195k yes yes configured
local-google/gemini-3-pro-image text,image 1953k yes yes configuredRestart Gateway
openclaw gateway restartTest Connection
openclaw message send "你好,介绍一下你自己"If a reply is returned normally, the configuration is successful.
11.1.8 Usage
Use Default Model (Claude Sonnet 4.5)
Just send a message:
openclaw message send "写1个Python脚本,打印Hello World"Switch to Opus Thinking Model
Suitable for complex problems requiring deep thought:
openclaw config set agents.defaults.model.primary "local-anthropic-opus/claude-opus-4-5-thinking"
openclaw gateway restartSwitch to Gemini Image Model
Suitable for scenarios requiring image recognition:
openclaw config set agents.defaults.model.primary "local-google/gemini-3-pro-image"
openclaw gateway restartTemporarily Use a Specific Model
Use a specific model temporarily without modifying the default configuration:
# Use Opus Thinking
openclaw agent --model "local-anthropic-opus/claude-opus-4-5-thinking" --message "解释量子计算的原理"
# Use Gemini Image
openclaw agent --model "local-google/gemini-3-pro-image" --message "分析这张图片" --image ./photo.jpg11.1.9 Model Selection Guide
Claude Sonnet 4.5
Applicable Scenarios:
- Daily conversations
- Code generation
- Document writing
- Quick Q&A
Features:
- Fast speed
- Low cost
- High quality
- Context window: 200k tokens
Claude Opus 4.5 Thinking
Applicable Scenarios:
- Complex reasoning
- Mathematical problems
- Algorithm optimization
- Deep analysis
Features:
- Strong reasoning ability
- Visible thought process
- Suitable for complex problems
- Context window: 200k tokens
Gemini 3 Pro Image
Applicable Scenarios:
- Image recognition
- Multimodal tasks
- Document analysis
- Design review
Features:
- Supports image input
- Ultra-large context window
- Accurate recognition
- Context window: 2000k tokens
11.1.10 Advanced Configuration
Configure Model Alias
Give the model an easy-to-remember name:
openclaw config set agents.defaults.models."local-anthropic/claude-sonnet-4-5-20250929".alias "我的Claude"Add Multiple API Keys
If you have multiple Antigravity accounts, you can configure multiple providers:
cat ~/.openclaw/openclaw.json | jq '.models.providers["local-anthropic-2"] = {
"baseUrl": "http://127.0.0.1:8045",
"apiKey": "另1个User_Token",
"auth": "api-key",
"api": "anthropic-messages",
"models": [...]
}' > /tmp/openclaw-temp.json && mv /tmp/openclaw-temp.json ~/.openclaw/openclaw.jsonConfigure Cost Tracking
Although local API cost is 0, you can set virtual costs to track usage:
{
"cost": {
"input": 0.003,
"output": 0.015,
"cacheRead": 0.0003,
"cacheWrite": 0.00375
}
}Backup Configuration
cp ~/.openclaw/openclaw.json ~/.openclaw/openclaw.json.backupRestore Configuration
cp ~/.openclaw/openclaw.json.backup ~/.openclaw/openclaw.json
openclaw gateway restart11.1.11 Quick Command Reference
# View model list
openclaw models list
# View current default model
openclaw config get agents.defaults.model.primary
# Switch default model
openclaw config set agents.defaults.model.primary "模型ID"
# Restart Gateway
openclaw gateway restart
# View configuration file
cat ~/.openclaw/openclaw.json | jq '.models.providers'
# Send message
openclaw message send "你的消息"
# Temporarily use a specific model
openclaw agent --model "模型ID" --message "你的消息"11.1.12 Model ID Quick Reference
local-anthropic/claude-sonnet-4-5-20250929
local-anthropic-opus/claude-opus-4-5-thinking
local-google/gemini-3-pro-image11.1.13 Troubleshooting
Problem 1: Model list is empty
Cause: Configuration file format error or incorrect path
Solution:
# Check configuration file
cat ~/.openclaw/openclaw.json | jq '.models.providers'
# If an error is returned, restore backup
cp ~/.openclaw/openclaw.json.backup ~/.openclaw/openclaw.jsonProblem 2: API connection failed
Cause: Antigravity Manager not started or port occupied
Solution:
# Check if API is normal
curl http://127.0.0.1:8045/v1/models
# Check port occupation (macOS/Linux)
lsof -i :8045
# Restart Antigravity ManagerProblem 3: Model not effective after configuration
Cause: Forgot to restart Gateway
Solution:
openclaw gateway restartProblem 4: User Token invalid
Cause: Token expired or entered incorrectly
Solution:
- Regenerate Token in Antigravity Manager
- Update apiKey in the configuration file
- Restart Gateway
Test connection
openclaw test api
### 10.1.5 Practical Cases
**Case 1: Configure Claude Sonnet**Steps:
- Get Claude API Key
- Add in Antigravity Manager
- Configure OpenClaw
- Test usage
Result: You: Hello OpenClaw (Claude Sonnet): Hello! I am Claude...
**Case 2: Multi-Account Management**Scenario: Manage multiple Claude accounts
Configuration:
- Claude Account 1: Daily use
- Claude Account 2: Backup
- Claude Account 3: Peak usage
Advantages:
- Distribute load
- Avoid rate limits
- Improve availability
---
## 11.2 Multi-Model Switching Strategy
### 11.2.1 Model Feature Comparison
| Model | Advantages | Disadvantages | Applicable Scenarios |
|------|------|------|----------|
| Claude Sonnet | Good balance | Medium price | Daily conversation |
| Claude Opus | Strongest capability | Most expensive | Complex tasks |
| GPT-5.2 | Rich features | Slower response | Creative work |
| Gemini 3 Pro | Large free tier | Average capability | Simple tasks |
| DeepSeek-V3 | High cost-effectiveness | Chinese optimized | Programming tasks |
### 11.2.2 Scenario-Based Selection Strategy
**Daily Conversation**:Recommendation: Claude Sonnet 4.5 Reasons:
- Fast response speed
- Stable quality
- Moderate price
**Complex Reasoning**:Recommendation: Claude Opus 4.6 Reasons:
- Strongest reasoning ability
- Highest accuracy
- Suitable for difficult problems
**Image Recognition**:Recommendation: Gemini 3 Pro Reasons:
- Strong multimodal capability
- Large free tier
- Accurate recognition
**Programming Tasks**:Recommendation: DeepSeek-V3 Reasons:
- Strong coding ability
- Cheap price
- Chinese friendly
### 11.2.3 Model Disaster Recovery Mechanism (Fallback)
> 🛡️ **High Availability Guarantee**: Ensure uninterrupted service by configuring primary and fallback models.
#### What is Model Disaster Recovery?
When the primary model encounters the following situations, the system will automatically switch to fallback models:
- API call failure
- Request timeout
- Rate Limit
- Service unavailability
<img src="https://upload.maynor1024.live/file/1771085328347_service-disaster-recovery.png" alt="Service Disaster Recovery Configuration Example" />
#### Basic Disaster Recovery Configuration
**Configuration file path**: `~/.openclaw/openclaw.json`
```json
{
"agents": {
"defaults": {
"model": {
"primary": "anthropic/claude-opus-4-6",
"fallbacks": [
"openai-codex/gpt-5.3-codex",
"google-antigravity/claude-opus-4-6-thinking"
]
}
},
"list": [
{
"id": "main",
"default": true,
"model": {
"primary": "anthropic/claude-opus-4-6",
"fallbacks": [
"openai-codex/gpt-5.3-codex",
"google-antigravity/claude-opus-4-6-thinking"
]
}
}
]
}
}Workflow:
1. Attempt to use primary model: anthropic/claude-opus-4-6
↓ Failure
2. Switch to fallback model 1: openai-codex/gpt-5.3-codex
↓ Failure
3. Switch to fallback model 2: google-antigravity/claude-opus-4-6-thinking
↓ Failure
4. Return error messagePractical Case 1: Cost-Optimized Disaster Recovery
Scenario: Prioritize cheaper models, use high-quality models if they fail
{
"agents": {
"defaults": {
"model": {
"primary": "deepseek/deepseek-chat",
"fallbacks": [
"anthropic/claude-sonnet-4-5",
"anthropic/claude-opus-4-6"
]
}
}
}
}Advantages:
- ✅ Daily use of DeepSeek (extremely low cost)
- ✅ Automatically switches to Claude Sonnet when DeepSeek is rate-limited
- ✅ Uses Claude Opus as a fallback for critical task failures
- ✅ Cost savings of 80%+
Practical Case 2: Performance-First Disaster Recovery
Scenario: Prioritize the strongest model, degrade if it fails
{
"agents": {
"defaults": {
"model": {
"primary": "anthropic/claude-opus-4-6",
"fallbacks": [
"anthropic/claude-sonnet-4-5",
"deepseek/deepseek-chat"
]
}
}
}
}Advantages:
- ✅ Ensures best quality
- ✅ Automatic degradation during peak hours
- ✅ Guarantees uninterrupted service
Practical Case 3: Multi-Provider Disaster Recovery
Scenario: Cross-provider disaster recovery to avoid single points of failure
{
"agents": {
"defaults": {
"model": {
"primary": "anthropic/claude-sonnet-4-5",
"fallbacks": [
"openai/gpt-4o",
"google/gemini-2.0-flash-exp",
"deepseek/deepseek-chat"
]
}
}
}
}Advantages:
- ✅ Switches to OpenAI if Anthropic fails
- ✅ Switches to Google if OpenAI fails
- ✅ Finally uses DeepSeek as a fallback
- ✅ Maximizes service availability
Command Line Configuration
# Set primary model
openclaw config set agents.defaults.model.primary "anthropic/claude-opus-4-6"
# Set fallback models (requires manual JSON editing)
# Or use jq command
cat ~/.openclaw/openclaw.json | jq '.agents.defaults.model.fallbacks = [
"openai-codex/gpt-5.3-codex",
"google-antigravity/claude-opus-4-6-thinking"
]' > /tmp/openclaw-temp.json && mv /tmp/openclaw-temp.json ~/.openclaw/openclaw.json
# Restart Gateway for configuration to take effect
openclaw gateway restartVerify Disaster Recovery Configuration
# View current configuration
openclaw config get agents.defaults.model
# Output example:
{
"primary": "anthropic/claude-opus-4-6",
"fallbacks": [
"openai-codex/gpt-5.3-codex",
"google-antigravity/claude-opus-4-6-thinking"
]
}Disaster Recovery Best Practices
1. Choose different providers:
✅ Recommended: Anthropic → OpenAI → Google
❌ Not recommended: Claude Opus → Claude Sonnet (same provider)2. Configure by capability gradient:
✅ Recommended: High capability → Medium capability → Low capability
❌ Not recommended: Low capability → High capability (wastes resources)3. Consider cost factors:
✅ Recommended: Cheap → Medium → Expensive (cost optimization)
✅ Recommended: Expensive → Medium → Cheap (quality priority)4. Limit fallback quantity:
✅ Recommended: 2-3 fallback models
❌ Not recommended: 5+ fallback models (overly complex)11.2.4 Multiple Authentication Profiles + Token Rotation
🔐 Account Management: Configure multiple authentication profiles to achieve account rotation and load balancing.
What is an Authentication Profile?
An authentication profile allows you to configure multiple accounts for the same provider. The system will rotate through them in a specified order to avoid single-account rate limits.
Basic Configuration
Configuration file path: ~/.openclaw/openclaw.json
{
"auth": {
"profiles": {
"openai-codex:default": {
"provider": "openai-codex",
"mode": "oauth"
},
"anthropic:default": {
"provider": "anthropic",
"mode": "token"
},
"anthropic:manual": {
"provider": "anthropic",
"mode": "token"
},
"google-antigravity:mail1@gmail.com": {
"provider": "google-antigravity",
"mode": "oauth",
"email": "mail1@gmail.com"
},
"google-antigravity:mail2@gmail.com": {
"provider": "google-antigravity",
"mode": "oauth"
}
},
"order": {
"anthropic": [
"anthropic:default",
"anthropic:manual"
],
"google-antigravity": [
"google-antigravity:mail1@gmail.com",
"google-antigravity:mail2@gmail.com"
]
}
}
}Configuration Description
profiles field:
- Defines all available authentication configurations
- Format:
"Provider:Identifier" mode: Authentication method (oauthortoken)email: OAuth account email (optional)
order field:
- Defines the usage order of accounts for each provider
- The system will rotate through them in order
- Automatically switches to the next account if the current one hits a rate limit
Practical Case 1: Anthropic Dual Account Rotation
Scenario: Configure 2 Claude API Keys to avoid rate limits
{
"auth": {
"profiles": {
"anthropic:account1": {
"provider": "anthropic",
"mode": "token"
},
"anthropic:account2": {
"provider": "anthropic",
"mode": "token"
}
},
"order": {
"anthropic": [
"anthropic:account1",
"anthropic:account2"
]
}
}
}Configure API Key:
# Configure two API Keys separately in Antigravity Manager
# Or add in OpenClaw configuration:
{
"models": {
"providers": {
"anthropic": {
"apiKey": "sk-ant-api-key-1",
...
},
"anthropic-2": {
"apiKey": "sk-ant-api-key-2",
...
}
}
}
}Workflow:
1. Use account1 to send request
2. account1 reaches rate limit → automatically switches to account2
3. account2 reaches rate limit → waits for account1 to recover
4. RepeatsPractical Case 2: Google Multi-Email Rotation
Scenario: Use multiple Google accounts to access Gemini
{
"auth": {
"profiles": {
"google-antigravity:work@gmail.com": {
"provider": "google-antigravity",
"mode": "oauth",
"email": "work@gmail.com"
},
"google-antigravity:personal@gmail.com": {
"provider": "google-antigravity",
"mode": "oauth",
"email": "personal@gmail.com"
},
"google-antigravity:backup@gmail.com": {
"provider": "google-antigravity",
"mode": "oauth",
"email": "backup@gmail.com"
}
},
"order": {
"google-antigravity": [
"google-antigravity:work@gmail.com",
"google-antigravity:personal@gmail.com",
"google-antigravity:backup@gmail.com"
]
}
}
}Advantages:
- ✅ 3 accounts rotate, reducing rate limit probability by 66%
- ✅ Free tiers stack (3x free tier)
- ✅ Automatic load balancing during peak hours
Practical Case 3: Mixed Authentication Mode
Scenario: Use both OAuth and API Token simultaneously
{
"auth": {
"profiles": {
"anthropic:oauth-account": {
"provider": "anthropic",
"mode": "oauth"
},
"anthropic:token-account": {
"provider": "anthropic",
"mode": "token"
}
},
"order": {
"anthropic": [
"anthropic:oauth-account",
"anthropic:token-account"
]
}
}
}Use Cases:
- OAuth account: Daily use (more secure)
- Token account: Backup (more stable)
Configuration Best Practices
1. Recommended account quantity:
✅ Recommended: 2-3 accounts
❌ Not recommended: 5+ accounts (complex to manage)2. Choice of authentication method:
OAuth: More secure, suitable for personal accounts
Token: More stable, suitable for API keys3. Rotation strategy:
✅ Sort by usage frequency (high frequency → low frequency)
✅ Sort by account level (paid → free)4. Monitoring and maintenance:
# View currently used authentication configuration
openclaw config get auth.profiles
# Test if authentication is valid
openclaw test api11.2.5 Automatic Switching Configuration
Switching based on task type:
{
"rules": [
{
"condition": "task.type === 'code'",
"model": "deepseek-v3"
},
{
"condition": "task.type === 'image'",
"model": "gemini-3-pro"
},
{
"condition": "task.complexity === 'high'",
"model": "claude-opus-4.6"
},
{
"condition": "default",
"model": "claude-sonnet-4.5"
}
]
}Switching based on cost:
{
"rules": [
{
"condition": "cost.daily < 10",
"model": "claude-opus-4.6"
},
{
"condition": "cost.daily >= 10",
"model": "claude-sonnet-4.5"
}
]
}11.3 Memory Search Configuration
🧠 Intelligent Memory: Configure Memory Search to allow OpenClaw to remember historical conversations, providing more intelligent context awareness.
11.3.1 What is Memory Search?
Memory Search is OpenClaw's memory system, which can:
- Remember historical conversation content
- Search relevant session records
- Provide context awareness
- Support hybrid retrieval (vector + text)
11.3.2 Basic Configuration
Configuration file path: ~/.openclaw/openclaw.json
{
"agents": {
"defaults": {
"memorySearch": {
"sources": ["memory", "sessions"],
"experimental": {
"sessionMemory": true
},
"provider": "gemini",
"remote": {
"apiKey": "AIzaSy**************************"
},
"fallback": "gemini",
"model": "gemini-embedding-001",
"query": {
"hybrid": {
"enabled": true,
"vectorWeight": 0.7,
"textWeight": 0.3
}
}
}
}
}
}11.3.3 Configuration Item Details
sources (Data Sources)
{
"sources": ["memory", "sessions"]
}Optional values:
memory: Long-term memory (cross-session)sessions: Session records (current session)
Recommended configuration:
// Use only long-term memory
"sources": ["memory"]
// Use both long-term memory and session records
"sources": ["memory", "sessions"]experimental (Experimental Features)
{
"experimental": {
"sessionMemory": true
}
}sessionMemory:
true: Enable session memory (recommended)false: Disable session memory
provider (Embedding Model Provider)
{
"provider": "gemini"
}Supported providers:
gemini: Google Gemini (recommended, free)openai: OpenAI Embeddingslocal: Local embedding models
Recommendation: Use Gemini (free and effective)
remote (Remote API Configuration)
{
"remote": {
"apiKey": "AIzaSy**************************"
}
}Get Gemini API Key:
- Visit Google AI Studio
- Log in with your Google account
- Create an API Key
- Copy and paste it into the configuration
fallback (Fallback Provider)
{
"fallback": "gemini"
}When the primary provider fails, use the fallback provider.
model (Embedding Model)
{
"model": "gemini-embedding-001"
}Gemini Embedding Models:
gemini-embedding-001: Standard model (recommended)text-embedding-004: Advanced model
OpenAI Embedding Models:
text-embedding-3-small: Small model (cheaper)text-embedding-3-large: Large model (better performance)
query (Query Configuration)
{
"query": {
"hybrid": {
"enabled": true,
"vectorWeight": 0.7,
"textWeight": 0.3
}
}
}hybrid (Hybrid Retrieval):
enabled: Whether to enable hybrid retrievalvectorWeight: Vector search weight (0-1)textWeight: Text search weight (0-1)
Weighting suggestions:
Semantic search priority: vectorWeight: 0.7, textWeight: 0.3
Keyword search priority: vectorWeight: 0.3, textWeight: 0.7
Balanced mode: vectorWeight: 0.5, textWeight: 0.511.3.4 Practical Case 1: Basic Configuration (Gemini)
Scenario: Use the free Gemini embedding model
{
"agents": {
"defaults": {
"memorySearch": {
"sources": ["memory", "sessions"],
"experimental": {
"sessionMemory": true
},
"provider": "gemini",
"remote": {
"apiKey": "你的Gemini_API_Key"
},
"model": "gemini-embedding-001",
"query": {
"hybrid": {
"enabled": true,
"vectorWeight": 0.7,
"textWeight": 0.3
}
}
}
}
}
}Advantages:
- ✅ Completely free
- ✅ Excellent performance
- ✅ Simple configuration
11.3.5 Practical Case 2: Advanced Configuration (OpenAI)
Scenario: Use OpenAI embedding model (higher precision)
{
"agents": {
"defaults": {
"memorySearch": {
"sources": ["memory", "sessions"],
"experimental": {
"sessionMemory": true
},
"provider": "openai",
"remote": {
"apiKey": "sk-your-openai-api-key"
},
"fallback": "gemini",
"model": "text-embedding-3-large",
"query": {
"hybrid": {
"enabled": true,
"vectorWeight": 0.8,
"textWeight": 0.2
}
}
}
}
}
}Advantages:
- ✅ Higher precision
- ✅ Supports more languages
- ✅ Has a fallback option
Cost:
- text-embedding-3-small: $0.02/million tokens
- text-embedding-3-large: $0.13/million tokens
11.3.6 Practical Case 3: Local Deployment (Privacy First)
Scenario: Use a local embedding model to protect privacy
{
"agents": {
"defaults": {
"memorySearch": {
"sources": ["memory", "sessions"],
"experimental": {
"sessionMemory": true
},
"provider": "local",
"model": "all-MiniLM-L6-v2",
"query": {
"hybrid": {
"enabled": true,
"vectorWeight": 0.6,
"textWeight": 0.4
}
}
}
}
}
}Advantages:
- ✅ Completely local, protects privacy
- ✅ No API Key required
- ✅ No usage restrictions
Disadvantages:
- ❌ Requires local computing resources
- ❌ Slightly lower precision than cloud models
11.3.7 Command Line Configuration
# Enable Memory Search
openclaw config set agents.defaults.memorySearch.experimental.sessionMemory true
# Set provider
openclaw config set agents.defaults.memorySearch.provider "gemini"
# Set API Key (requires manual JSON editing)
# Or use jq command
cat ~/.openclaw/openclaw.json | jq '.agents.defaults.memorySearch.remote.apiKey = "你的API_Key"' > /tmp/openclaw-temp.json && mv /tmp/openclaw-temp.json ~/.openclaw/openclaw.json
# Restart Gateway
openclaw gateway restart11.3.8 Verify Configuration
# View current configuration
openclaw config get agents.defaults.memorySearch
# Test memory search
openclaw message send "记住:我喜欢喝咖啡"
openclaw message send "我喜欢喝什么?"
# Should return: According to my memory, you like to drink coffee.11.3.9 Use Cases
Scenario 1: Personal Assistant
You: Remember my birthday is January 1, 1990
OpenClaw: Okay, remembered.
(A few days later)
You: When is my birthday?
OpenClaw: According to my memory, your birthday is January 1, 1990.Scenario 2: Project Management
You: Project A's deadline is March 1, 2026
OpenClaw: Noted.
(A week later)
You: When is Project A due?
OpenClaw: Project A's deadline is March 1, 2026.Scenario 3: Knowledge Accumulation
You: DeepSeek API price is $0.001/thousand tokens
OpenClaw: Remembered.
(Next conversation)
You: Which model is the cheapest?
OpenClaw: According to my memory, DeepSeek is the cheapest, priced at $0.001/thousand tokens.
### 11.3.10 Best Practices
**1. Choose the Right Provider**:免费用户:Gemini(免费且效果好) 付费用户:OpenAI(精度更高) 隐私优先:Local(完全本地)
Free Users: Gemini (free and effective) Paid Users: OpenAI (higher accuracy) Privacy First: Local (fully local)
**2. Adjust Hybrid Retrieval Weight**:语义理解为主:vectorWeight: 0.7-0.8 关键词匹配为主:textWeight: 0.6-0.7 平衡模式:各 0.5
Semantic understanding dominant: vectorWeight: 0.7-0.8 Keyword matching dominant: textWeight: 0.6-0.7 Balanced mode: 0.5 each
**3. Regularly Clean Memory**:
```bash
# 清理过期记忆
openclaw memory clean --older-than 30d
# 查看记忆使用情况
openclaw memory stats# Clean expired memory
openclaw memory clean --older-than 30d
# View memory usage
openclaw memory stats4. Back up Important Memories:
# 导出记忆
openclaw memory export --output memory-backup.json
# 导入记忆
openclaw memory import memory-backup.json# Export memory
openclaw memory export --output memory-backup.json
# Import memory
openclaw memory import memory-backup.json11.3.11 Troubleshooting
Problem 1: Memory search not working
Cause: API Key invalid or not configured
Solution:
# 检查配置
openclaw config get agents.defaults.memorySearch
# 测试 API Key
curl -H "Content-Type: application/json" \
-d '{"contents":[{"parts":[{"text":"test"}]}]}' \
"https://generativelanguage.googleapis.com/v1beta/models/gemini-embedding-001:embedContent?key=你的API_Key"# Check configuration
openclaw config get agents.defaults.memorySearch
# Test API Key
curl -H "Content-Type: application/json" \
-d '{"contents":[{"parts":[{"text":"test"}]}]}' \
"https://generativelanguage.googleapis.com/v1beta/models/gemini-embedding-001:embedContent?key=你的API_Key"Problem 2: Search results inaccurate
Cause: Inappropriate hybrid retrieval weights
Solution:
// 调整权重
{
"query": {
"hybrid": {
"vectorWeight": 0.8, // 提高语义搜索权重
"textWeight": 0.2
}
}
}// Adjust weights
{
"query": {
"hybrid": {
"vectorWeight": 0.8, // Increase semantic search weight
"textWeight": 0.2
}
}
}Problem 3: Memory occupying too much space
Cause: Long-term accumulation without cleanup
Solution:
# 查看记忆大小
openclaw memory stats
# 清理旧记忆
openclaw memory clean --older-than 60d
# 压缩记忆数据库
openclaw memory compact# View memory size
openclaw memory stats
# Clean old memory
openclaw memory clean --older-than 60d
# Compact memory database
openclaw memory compact11.4 Cost Optimization Solutions
11.4.1 Token Consumption Analysis
View Consumption Statistics:
# 查看今日消耗
openclaw stats today
# 输出示例:
今日Token消耗:
- Claude Sonnet:150K tokens ($0.75)
- Gemini Pro:50K tokens ($0.00)
- 总计:200K tokens ($0.75)
任务分布:
- 文件搜索:30%
- 日程管理:20%
- 知识管理:25%
- 其他:25%# View today's consumption
openclaw stats today
# Output example:
Today's Token Consumption:
- Claude Sonnet:150K tokens ($0.75)
- Gemini Pro:50K tokens ($0.00)
- Total:200K tokens ($0.75)
Task Distribution:
- File Search:30%
- Schedule Management:20%
- Knowledge Management:25%
- Other:25%Consumption Optimization Suggestions:
⚠️ 高消耗任务:
- 文件搜索:每次10K tokens
- 建议:优化搜索范围
✅ 优化方案:
- 使用缓存
- 减少上下文
- 优化提示词⚠️ High Consumption Tasks:
- File Search: 10K tokens per search
- Suggestion: Optimize search scope
✅ Optimization Solutions:
- Use caching
- Reduce context
- Optimize prompts11.4.2 Caching Strategy
Enable Caching:
# 启用响应缓存
openclaw config set cache.enabled true
# 设置缓存时间(小时)
openclaw config set cache.ttl 24
# 设置缓存大小(MB)
openclaw config set cache.maxSize 1000# Enable response caching
openclaw config set cache.enabled true
# Set cache time (hours)
openclaw config set cache.ttl 24
# Set cache size (MB)
openclaw config set cache.maxSize 1000Caching Effect:
未启用缓存:
- 相同问题每次都调用API
- Token消耗:10K/次
- 成本:$0.05/次
启用缓存后:
- 相同问题直接返回缓存
- Token消耗:0
- 成本:$0
- 节省:100%Without caching:
- Same question calls API every time
- Token consumption: 10K/call
- Cost: $0.05/call
With caching enabled:
- Same question returns cached response directly
- Token consumption: 0
- Cost: $0
- Savings: 100%11.4.3 Model Downgrade Solution
Downgrade Strategy:
1. 简单任务用便宜模型
2. 复杂任务用贵模型
3. 失败后降级重试1. Use cheaper models for simple tasks
2. Use more expensive models for complex tasks
3. Downgrade and retry after failureConfiguration Example:
{
"fallback": [
"claude-opus-4.6", // 首选
"claude-sonnet-4.5", // 降级1
"gemini-3-pro" // 降级2
]
}{
"fallback": [
"claude-opus-4.6", // Primary
"claude-sonnet-4.5", // Fallback 1
"gemini-3-pro" // Fallback 2
]
}11.4.4 Cost Control in Practice
Case 1: Reduce Cost by 50%
原方案:
- 全部使用Claude Opus
- 日均消耗:$20
优化方案:
- 简单任务用Sonnet
- 复杂任务用Opus
- 启用缓存
优化后:
- 日均消耗:$10
- 节省:50%Original plan:
- All tasks use Claude Opus
- Daily consumption: $20
Optimized plan:
- Simple tasks use Sonnet
- Complex tasks use Opus
- Enable caching
After optimization:
- Daily consumption: $10
- Savings: 50%Case 2: Maximize Free Quota
策略:
1. 优先使用Gemini(免费额度大)
2. 超额后切换到DeepSeek(便宜)
3. 重要任务用Claude
效果:
- 月成本:$5
- 节省:90%Strategy:
1. Prioritize Gemini (large free quota)
2. Switch to DeepSeek after exceeding quota (cheaper)
3. Use Claude for important tasks
Effect:
- Monthly cost: $5
- Savings: 90%11.5 Performance Tuning Tips
11.5.1 Response Speed Optimization
Before Optimization:
平均响应时间:5秒
用户体验:一般Average response time: 5 seconds
User experience: AverageOptimization Solutions:
1. 启用缓存
2. 减少上下文
3. 使用流式输出
4. 并发处理1. Enable caching
2. Reduce context
3. Use streaming output
4. Concurrent processingAfter Optimization:
平均响应时间:2秒
用户体验:优秀
提升:60%Average response time: 2 seconds
User experience: Excellent
Improvement: 60%11.5.2 Concurrency Processing Optimization
Configure Concurrency:
# 设置最大并发数
openclaw config set concurrency.max 5
# 设置队列大小
openclaw config set concurrency.queueSize 100# Set maximum concurrency
openclaw config set concurrency.max 5
# Set queue size
openclaw config set concurrency.queueSize 10011.5.3 Memory Management
Monitor Memory Usage:
# 查看内存使用
openclaw stats memory
# 输出示例:
内存使用情况:
- 当前:512MB
- 峰值:800MB
- 平均:600MB# View memory usage
openclaw stats memory
# Output example:
Memory Usage:
- Current: 512MB
- Peak: 800MB
- Average: 600MBOptimization Suggestions:
⚠️ 内存占用高:
- 清理缓存
- 减少并发
- 重启服务⚠️ High memory usage:
- Clear cache
- Reduce concurrency
- Restart service📝 Chapter Summary
Learned about OpenClaw's advanced configurations:
- Antigravity Manager configuration
- Multi-model switching strategy
- Cost optimization solutions
- Performance tuning tips
Mastering these techniques can:
- Reduce costs by over 50%
- Improve response speed by 60%
- Enhance system stability
11.6 Model Provider Configuration Details
🤖 Multi-model Support: OpenClaw supports 20+ mainstream AI model providers, offering flexible configuration to meet different needs.
11.6.0 Quick Configuration: Using the Command Line Wizard (Recommended for Newbies)
💡 Simplest Way: Use the openclaw onboard command to start the configuration wizard and interactively configure models.
Launch Configuration Wizard
openclaw onboardExecuting this will launch the command-line interactive configuration wizard.
Configuration Process
Step 1: Select Initialization Mode
◇ 初始化模式
│ 快速开始◇ Initialization Mode
│ Quick StartStep 2: Select Model Provider
◆ 模型/认证提供商
│ ○ OpenAI (Codex OAuth + API key)
│ ○ Anthropic
│ ○ MiniMax
│ ○ Moonshot AI
│ ○ Google
│ ○ OpenRouter
│ ○ Qwen
│ ○ Z.AI (GLM 4.7)
│ ○ Copilot
│ ○ Vercel AI Gateway
│ ○ OpenCode Zen
│ ○ Xiaomi
│ ○ Synthetic
│ ○ Venice AI
│ ○ Skip for nowUse arrow keys to select, spacebar to confirm.
Step 3: Enter API Key
Enter the API Key for the corresponding provider as prompted.
Step 4: Select Default Model
Select the default model from the list of available models.
Step 5: Complete Configuration
Configuration is automatically saved and Gateway restarts.
Advantages of the Command Line Wizard
✅ Interactive Operation: Step-by-step guidance, less prone to errors ✅ Real-time Verification: API Key validity is checked immediately after input ✅ Automatic Configuration: Configuration files are generated automatically ✅ One-click Save: Automatically saves and restarts the service ✅ Error Prompts: Clear error messages are provided for configuration errors
Verify Configuration
After configuration is complete, verify if the model is available:
# 查看已配置的模型
openclaw models list
# 测试模型连接
openclaw message send "你好,测试一下"# View configured models
openclaw models list
# Test model connection
openclaw message send "Hello, test it out"Modify Configuration
If you need to modify the configuration, run again:
openclaw onboardYou can add, delete, or modify model providers.
11.6.1 Supported Model Providers
International Models
| Provider | Model | Features | Price |
|---|---|---|---|
| OpenAI | GPT-4o, GPT-4o-mini | Comprehensive features, mature ecosystem | High |
| Anthropic | Claude 3.5 Sonnet, Claude 3 Opus | Strong reasoning capabilities, high security | Medium-High |
| Gemini 2.0 Flash, Gemini 1.5 Pro | Strong multimodal capabilities, large free quota | Medium | |
| xAI | Grok 2 | Real-time information, humorous style | Medium |
| Mistral | Mistral Large, Mistral Small | Open-source friendly, high cost-performance | Medium |
| Cohere | Command R+, Command R | Enterprise-grade, RAG optimized | Medium |
Domestic Models
| Provider | Model | Features | Price |
|---|---|---|---|
| DeepSeek | DeepSeek-V3, DeepSeek-Chat | King of cost-performance, strong programming capabilities | Very Low |
| Moonshot AI | Kimi k2.5 | Ultra-long context (2 million characters) | Low |
| ZhipuAI | GLM-4, GLM-4V | Multimodal, Chinese optimized | Medium |
| Baichuan Intelligent | Baichuan-4 | Good Chinese understanding | Medium |
| MiniMax | abab6.5 | Speech synthesis, role-playing | Medium |
| Alibaba Cloud | Qwen-Max, Qwen-Plus | Alibaba ecosystem, enterprise-grade | Medium |
| Baidu | ERNIE 4.0 | Baidu ecosystem, knowledge enhanced | Medium |
Local Models
| Provider | Model | Features | Price |
|---|---|---|---|
| Ollama | Llama 3.1, Qwen2.5 | Fully local, privacy protection | Free |
| LM Studio | Various open-source models | Graphical interface, easy to use | Free |
11.6.2 Configure OpenAI
{
"models": {
"mode": "merge",
"providers": {
"openai": {
"baseUrl": "https://api.openai.com/v1",
"apiKey": "sk-your-api-key",
"auth": "api-key",
"api": "openai-chat",
"models": [
{
"id": "gpt-4o",
"name": "GPT-4o",
"contextWindow": 128000,
"maxTokens": 16384
},
{
"id": "gpt-4o-mini",
"name": "GPT-4o Mini",
"contextWindow": 128000,
"maxTokens": 16384
}
]
}
}
}
}11.6.3 Configure Anthropic (Claude)
{
"models": {
"mode": "merge",
"providers": {
"anthropic": {
"baseUrl": "https://api.anthropic.com",
"apiKey": "sk-ant-your-api-key",
"auth": "api-key",
"api": "anthropic",
"models": [
{
"id": "claude-3-5-sonnet-20241022",
"name": "Claude 3.5 Sonnet",
"contextWindow": 200000,
"maxTokens": 8192
},
{
"id": "claude-3-opus-20240229",
"name": "Claude 3 Opus",
"contextWindow": 200000,
"maxTokens": 4096
}
]
}
}
}
}11.6.4 Configure Google Gemini
{
"models": {
"mode": "merge",
"providers": {
"google": {
"baseUrl": "https://generativelanguage.googleapis.com/v1beta",
"apiKey": "your-google-api-key",
"auth": "api-key",
"api": "google-ai",
"models": [
{
"id": "gemini-2.0-flash-exp",
"name": "Gemini 2.0 Flash",
"contextWindow": 1000000,
"maxTokens": 8192
},
{
"id": "gemini-1.5-pro",
"name": "Gemini 1.5 Pro",
"contextWindow": 2000000,
"maxTokens": 8192
}
]
}
}
}
}11.6.5 Configure DeepSeek (Recommended)
{
"models": {
"mode": "merge",
"providers": {
"deepseek": {
"baseUrl": "https://api.deepseek.com",
"apiKey": "sk-your-api-key",
"auth": "api-key",
"api": "openai-chat",
"models": [
{
"id": "deepseek-chat",
"name": "DeepSeek Chat",
"contextWindow": 64000,
"maxTokens": 4096
},
{
"id": "deepseek-coder",
"name": "DeepSeek Coder",
"contextWindow": 64000,
"maxTokens": 4096
}
]
}
}
}
}11.6.6 Configure Kimi (Moonshot AI)
{
"models": {
"mode": "merge",
"providers": {
"moonshot": {
"baseUrl": "https://api.moonshot.cn/v1",
"apiKey": "sk-your-api-key",
"auth": "api-key",
"api": "openai-chat",
"models": [
{
"id": "moonshot-v1-8k",
"name": "Kimi k2.5 8K",
"contextWindow": 8000,
"maxTokens": 4096
},
{
"id": "moonshot-v1-32k",
"name": "Kimi k2.5 32K",
"contextWindow": 32000,
"maxTokens": 4096
},
{
"id": "moonshot-v1-128k",
"name": "Kimi k2.5 128K",
"contextWindow": 128000,
"maxTokens": 4096
}
]
}
}
}
}11.6.7 Configure Ollama (Local Model)
{
"models": {
"mode": "merge",
"providers": {
"ollama": {
"baseUrl": "http://localhost:11434",
"auth": "none",
"api": "ollama",
"models": [
{
"id": "llama3.1:8b",
"name": "Llama 3.1 8B",
"contextWindow": 128000,
"maxTokens": 4096
},
{
"id": "qwen2.5:7b",
"name": "Qwen 2.5 7B",
"contextWindow": 32000,
"maxTokens": 4096
}
]
}
}
}
}11.6.8 Multi-Provider Configuration Example
{
"models": {
"mode": "merge",
"providers": {
"deepseek": {
"baseUrl": "https://api.deepseek.com",
"apiKey": "sk-deepseek-key",
"auth": "api-key",
"api": "openai-chat",
"models": [
{
"id": "deepseek-chat",
"name": "DeepSeek Chat",
"contextWindow": 64000,
"maxTokens": 4096
}
]
},
"anthropic": {
"baseUrl": "https://api.anthropic.com",
"apiKey": "sk-ant-key",
"auth": "api-key",
"api": "anthropic",
"models": [
{
"id": "claude-3-5-sonnet-20241022",
"name": "Claude 3.5 Sonnet",
"contextWindow": 200000,
"maxTokens": 8192
}
]
},
"ollama": {
"baseUrl": "http://localhost:11434",
"auth": "none",
"api": "ollama",
"models": [
{
"id": "llama3.1:8b",
"name": "Llama 3.1 8B",
"contextWindow": 128000,
"maxTokens": 4096
}
]
}
}
},
"agents": {
"defaults": {
"model": {
"primary": "deepseek/deepseek-chat",
"fallback": [
"anthropic/claude-3-5-sonnet-20241022",
"ollama/llama3.1:8b"
]
}
}
}
}11.6.9 Model Selection Strategy
Select by Task Type:
// 编程任务
"deepseek/deepseek-coder"
// 长文档处理
"moonshot/moonshot-v1-128k"
// 复杂推理
"anthropic/claude-3-opus-20240229"
// 日常对话
"deepseek/deepseek-chat"
// 多模态(图片)
"google/gemini-2.0-flash-exp"
// 本地隐私
"ollama/llama3.1:8b"// Programming tasks
"deepseek/deepseek-coder"
// Long document processing
"moonshot/moonshot-v1-128k"
// Complex reasoning
"anthropic/claude-3-opus-20240229"
// Daily conversation
"deepseek/deepseek-chat"
// Multimodal (images)
"google/gemini-2.0-flash-exp"
// Local privacy
"ollama/llama3.1:8b"Select by Cost:
极低成本:DeepSeek ($0.001/千tokens)
低成本:Kimi, GLM-4 ($0.01/千tokens)
中等成本:Gemini, Mistral ($0.05/千tokens)
高成本:Claude, GPT-4 ($0.15/千tokens)
免费:Ollama(本地)Very Low Cost: DeepSeek ($0.001/thousand tokens)
Low Cost: Kimi, GLM-4 ($0.01/thousand tokens)
Medium Cost: Gemini, Mistral ($0.05/thousand tokens)
High Cost: Claude, GPT-4 ($0.15/thousand tokens)
Free: Ollama (local)11.7 Tool System Details
🔧 Extended Capabilities: OpenClaw's tool system enables AI to perform various operations, from file management to API calls.
11.7.1 Built-in Tool List
File System Tools
| Tool | Function | Example |
|---|---|---|
read_file | Read file content | Read configuration file |
write_file | Write file | Save notes |
list_directory | List directory | View file list |
search_files | Search files | Find all PDFs |
move_file | Move file | Organize files |
delete_file | Delete file | Clean up temporary files |
Shell Tools
| Tool | Function | Example |
|---|---|---|
execute_command | Execute command | Run script |
run_script | Run script | Batch processing tasks |
Network Tools
| Tool | Function | Example |
|---|---|---|
web_search | Web search | Search for latest information |
fetch_url | Fetch URL | Download content |
api_call | API call | Call third-party services |
Data Processing Tools
| Tool | Function | Example |
|---|---|---|
parse_json | Parse JSON | Process API response |
parse_csv | Parse CSV | Process tabular data |
extract_text | Extract text | Extract from PDF |
11.7.2 Enable and Disable Tools
View Available Tools:
openclaw tools listEnable Tools:
openclaw tools enable read_file write_fileDisable Tools:
openclaw tools disable execute_commandConfiguration File Method:
{
"tools": {
"enabled": [
"read_file",
"write_file",
"list_directory",
"web_search"
],
"disabled": [
"execute_command",
"delete_file"
]
}
}11.7.3 Tool Permission Control
Set Tool Permissions:
{
"tools": {
"permissions": {
"read_file": {
"allowedPaths": [
"~/Documents",
"~/Downloads"
],
"deniedPaths": [
"~/.ssh",
"~/.openclaw"
]
},
"execute_command": {
"allowedCommands": [
"ls",
"cat",
"grep"
],
"deniedCommands": [
"rm",
"sudo"
]
}
}
}
}11.7.4 Custom Tool Development
Create Custom Tool:
// ~/.openclaw/tools/my-tool.js
export default {
name: "my_custom_tool",
description: "我的自定义工具",
parameters: {
type: "object",
properties: {
input: {
type: "string",
description: "输入参数"
}
},
required: ["input"]
},
async execute({ input }) {
// 工具逻辑
return {
success: true,
result: `处理结果: ${input}`
};
}
};// ~/.openclaw/tools/my-tool.js
export default {
name: "my_custom_tool",
description: "My custom tool",
parameters: {
type: "object",
properties: {
input: {
type: "string",
description: "Input parameter"
}
},
required: ["input"]
},
async execute({ input }) {
// Tool logic
return {
success: true,
result: `Processing result: ${input}`
};
}
};Register Tool:
openclaw tools register ~/.openclaw/tools/my-tool.js11.7.5 Tool Usage Examples
File Search:
你:帮我找到所有包含"发票"的 PDF 文件
OpenClaw 使用工具:
1. search_files(pattern="*.pdf", content="发票")
2. 返回结果:找到 3 个文件
- 发票_2024_01.pdf
- 报销发票.pdf
- 采购发票_Q1.pdfYou: Help me find all PDF files containing "invoice"
OpenClaw uses tool:
1. search_files(pattern="*.pdf", content="invoice")
2. Returns result: Found 3 files
- Invoice_2024_01.pdf
- Expense_Invoice.pdf
- Procurement_Invoice_Q1.pdfWeb Search:
你:Claude 3.5 Sonnet 最新价格是多少?
OpenClaw 使用工具:
1. web_search(query="Claude 3.5 Sonnet pricing")
2. fetch_url(url="https://www.anthropic.com/pricing")
3. 返回结果:
- 输入:$3/百万 tokens
- 输出:$15/百万 tokensYou: What is the latest price for Claude 3.5 Sonnet?
OpenClaw uses tool:
1. web_search(query="Claude 3.5 Sonnet pricing")
2. fetch_url(url="https://www.anthropic.com/pricing")
3. Returns result:
- Input: $3/million tokens
- Output: $15/million tokensData Processing:
你:分析这个 CSV 文件的销售数据
OpenClaw 使用工具:
1. read_file(path="sales.csv")
2. parse_csv(content=...)
3. 分析数据并生成报告You: Analyze the sales data in this CSV file
OpenClaw uses tool:
1. read_file(path="sales.csv")
2. parse_csv(content=...)
3. Analyze data and generate report11.7.6 Tool Chaining
OpenClaw can automatically combine multiple tools to complete complex tasks:
任务:下载网页并保存为 Markdown
工具链:
1. fetch_url(url) → 获取网页内容
2. extract_text(html) → 提取文本
3. convert_to_markdown(text) → 转换格式
4. write_file(path, content) → 保存文件Task: Download webpage and save as Markdown
Toolchain:
1. fetch_url(url) → Get webpage content
2. extract_text(html) → Extract text
3. convert_to_markdown(text) → Convert format
4. write_file(path, content) → Save file11.7.7 Tool Security Best Practices
1. Principle of Least Privilege:
{
"tools": {
"enabled": [
"read_file", // 只启用必要的工具
"web_search"
]
}
}{
"tools": {
"enabled": [
"read_file", // Only enable necessary tools
"web_search"
]
}
}2. Path Restrictions:
{
"tools": {
"permissions": {
"read_file": {
"allowedPaths": ["~/Documents"] // 限制访问范围
}
}
}
}{
"tools": {
"permissions": {
"read_file": {
"allowedPaths": ["~/Documents"] // Restrict access scope
}
}
}
}3. Command Whitelist:
{
"tools": {
"permissions": {
"execute_command": {
"allowedCommands": ["ls", "cat"] // 只允许安全命令
}
}
}
}{
"tools": {
"permissions": {
"execute_command": {
"allowedCommands": ["ls", "cat"] // Only allow safe commands
}
}
}
}11.8 CLI Command Full Reference
📟 Command Line Tools: OpenClaw provides powerful CLI tools for easy management and operation.
11.8.1 Core Commands
Version and Help
# 查看版本
openclaw --version
openclaw -v
# 查看帮助
openclaw --help
openclaw -h
# 查看子命令帮助
openclaw gateway --help# View version
openclaw --version
openclaw -v
# View help
openclaw --help
openclaw -h
# View subcommand help
openclaw gateway --helpInitialization and Configuration
# 运行配置向导
openclaw onboard
# 快速开始向导
openclaw setup
# 查看配置
openclaw config list
# 获取配置项
openclaw config get models.providers
# 设置配置项
openclaw config set gateway.port 18790
# 删除配置项
openclaw config delete models.providers.test# Run configuration wizard
openclaw onboard
# Quick start wizard
openclaw setup
# View configuration
openclaw config list
# Get configuration item
openclaw config get models.providers
# Set configuration item
openclaw config set gateway.port 18790
# Delete configuration item
openclaw config delete models.providers.test11.8.2 Gateway Management
# 安装/启动 Gateway
openclaw gateway install
# 查看状态
openclaw gateway status
# 停止 Gateway
openclaw gateway stop
# 重启 Gateway
openclaw gateway restart
# 查看日志
openclaw logs
openclaw logs --follow
openclaw logs --tail 100
# 清理日志
openclaw logs clear# Install/Start Gateway
openclaw gateway install
# View status
openclaw gateway status
# Stop Gateway
openclaw gateway stop
# Restart Gateway
openclaw gateway restart
# View logs
openclaw logs
openclaw logs --follow
openclaw logs --tail 100
# Clear logs
openclaw logs clear11.8.3 Channel Management
# 列出所有渠道
openclaw channels list
# 查看渠道状态
openclaw channels status
# 添加渠道
openclaw channels add
# 删除渠道
openclaw channels remove feishu
# 测试渠道
openclaw channels test feishu# List all channels
openclaw channels list
# View channel status
openclaw channels status
# Add channel
openclaw channels add
# Delete channel
openclaw channels remove feishu
# Test channel
openclaw channels test feishu11.8.4 Pairing Management
# 列出配对请求
openclaw pairing list
openclaw pairing list feishu
# 批准配对
openclaw pairing approve feishu <CODE>
# 拒绝配对
openclaw pairing reject feishu <CODE>
# 清理过期配对
openclaw pairing cleanup# List pairing requests
openclaw pairing list
openclaw pairing list feishu
# Approve pairing
openclaw pairing approve feishu <CODE>
# Reject pairing
openclaw pairing reject feishu <CODE>
# Clean up expired pairings
openclaw pairing cleanup11.8.5 Plugin Management
# 列出已安装插件
openclaw plugins list
# 搜索插件
openclaw plugins search feishu
# 安装插件
openclaw plugins install @openclaw/feishu
# 卸载插件
openclaw plugins uninstall @openclaw/feishu
# 更新插件
openclaw plugins update @openclaw/feishu
# 更新所有插件
openclaw plugins update --all# List installed plugins
openclaw plugins list
# Search plugins
openclaw plugins search feishu
# Install plugin
openclaw plugins install @openclaw/feishu
# Uninstall plugin
openclaw plugins uninstall @openclaw/feishu
# Update plugin
openclaw plugins update @openclaw/feishu
# Update all plugins
openclaw plugins update --all11.8.6 Tool Management
# 列出所有工具
openclaw tools list
# 启用工具
openclaw tools enable read_file write_file
# 禁用工具
openclaw tools disable execute_command
# 注册自定义工具
openclaw tools register ~/my-tool.js
# 测试工具
openclaw tools test read_file# List all tools
openclaw tools list
# Enable tools
openclaw tools enable read_file write_file
# Disable tools
openclaw tools disable execute_command
# Register custom tool
openclaw tools register ~/my-tool.js
# Test tool
openclaw tools test read_file11.8.7 Agent Management
# 列出 Agents
openclaw agents list
# 创建 Agent
openclaw agents create my-agent
# 删除 Agent
openclaw agents delete my-agent
# 切换 Agent
openclaw agents switch my-agent
# 查看 Agent 配置
openclaw agents config my-agent# List Agents
openclaw agents list
# Create Agent
openclaw agents create my-agent
# Delete Agent
openclaw agents delete my-agent
# Switch Agent
openclaw agents switch my-agent
# View Agent configuration
openclaw agents config my-agent11.8.8 Session Management
# 列出会话
openclaw sessions list
# 查看会话详情
openclaw sessions show `<session-id>`
# 删除会话
openclaw sessions delete `<session-id>`
# 清理所有会话
openclaw sessions clear
# 导出会话
openclaw sessions export `<session-id>` --output session.json
# 导入会话
openclaw sessions import session.json# List sessions
openclaw sessions list
# View session details
openclaw sessions show `<session-id>`
# Delete session
openclaw sessions delete `<session-id>`
# Clear all sessions
openclaw sessions clear
# Export session
openclaw sessions export `<session-id>` --output session.json
# Import session
openclaw sessions import session.json11.8.9 Statistics and Monitoring
# 查看统计信息
openclaw stats
# 查看今日统计
openclaw stats today
# 查看本周统计
openclaw stats week
# 查看 API 消耗
openclaw stats api
# 查看内存使用
openclaw stats memory
# 查看性能指标
openclaw stats performance# View statistics
openclaw stats
# View today's statistics
openclaw stats today
# View this week's statistics
openclaw stats week
# View API consumption
openclaw stats api
# View memory usage
openclaw stats memory
# View performance metrics
openclaw stats performance11.8.10 Testing and Diagnostics
# 测试 API 连接
openclaw test api
# 测试渠道
openclaw test channel feishu
# 测试工具
openclaw test tool read_file
# 运行诊断
openclaw diagnose
# 检查配置
openclaw validate config
# 检查健康状态
openclaw health check# Test API connection
openclaw test api
# Test channel
openclaw test channel feishu
# Test tool
openclaw test tool read_file
# Run diagnostics
openclaw diagnose
# Check configuration
openclaw validate config
# Check health status
openclaw health check11.8.11 Data Management
# 备份数据
openclaw backup create
# 列出备份
openclaw backup list
# 恢复备份
openclaw backup restore `<backup-id>`
# 清理缓存
openclaw cache clear
# 清理临时文件
openclaw cleanup temp
# 导出数据
openclaw export --output data.json
# 导入数据
openclaw import data.json# Create backup
openclaw backup create
# List backups
openclaw backup list
# Restore backup
openclaw backup restore `<backup-id>`
# Clear cache
openclaw cache clear
# Clean up temporary files
openclaw cleanup temp
# Export data
openclaw export --output data.json
# Import data
openclaw import data.json11.8.12 Updates and Maintenance
# 检查更新
openclaw update check
# 更新到最新版本
openclaw update
# 更新到指定版本
openclaw update --version 2026.3.2
# 回滚版本
openclaw rollback
# 卸载
openclaw uninstall# Check for updates
openclaw update check
# Update to latest version
openclaw update
# Update to specific version
openclaw update --version 2026.3.2
# Rollback version
openclaw rollback
# Uninstall
openclaw uninstall11.8.13 Development and Debugging
# 开发模式启动
openclaw dev
# 调试模式
openclaw --debug
# 详细日志
openclaw --verbose
# 运行测试
openclaw test
# 构建项目
openclaw build
# 清理构建
openclaw clean# Start in development mode
openclaw dev
# Debug mode
openclaw --debug
# Verbose logs
openclaw --verbose
# Run tests
openclaw test
# Build project
openclaw build
# Clean build
openclaw clean11.8.14 Common Command Combinations
Quick Restart:
openclaw gateway stop && openclaw gateway installView Real-time Logs:
openclaw logs --follow | grep ERRORBackup and Update:
openclaw backup create && openclaw updateClean and Restart:
openclaw cache clear && openclaw gateway restartFull Diagnostics:
openclaw diagnose && openclaw health check && openclaw test api11.8.15 Environment Variables
# 设置日志级别
export OPENCLAW_LOG_LEVEL=debug
# 设置配置目录
export OPENCLAW_HOME=~/.openclaw
# 设置 Gateway 端口
export OPENCLAW_PORT=18789
# 设置 API Key
export DEEPSEEK_API_KEY=sk-xxx
export MOONSHOT_API_KEY=sk-xxx# Set log level
export OPENCLAW_LOG_LEVEL=debug
# Set configuration directory
export OPENCLAW_HOME=~/.openclaw
# Set Gateway port
export OPENCLAW_PORT=18789
# Set API Key
export DEEPSEEK_API_KEY=sk-xxx
export MOONSHOT_API_KEY=sk-xxx11.8.16 Configuration File Locations
# 主配置文件
~/.openclaw/openclaw.json
# 日志文件
~/.openclaw/logs/gateway.log
# 缓存目录
~/.openclaw/cache/
# 数据目录
~/.openclaw/data/
# 插件目录
~/.openclaw/plugins/
# 工具目录
~/.openclaw/tools/# Main configuration file
~/.openclaw/openclaw.json
# Log file
~/.openclaw/logs/gateway.log
# Cache directory
~/.openclaw/cache/
# Data directory
~/.openclaw/data/
# Plugin directory
~/.openclaw/plugins/
# Tool directory
~/.openclaw/tools/📝 Chapter Summary
Learned about OpenClaw's advanced configurations:
Core Content
- Antigravity Manager Configuration - Unified API management
- Multi-model Switching Strategy - Scenario-based selection + model disaster recovery mechanism
- Memory Search Configuration - Intelligent context awareness
- Cost Optimization Solutions - Reduce costs by 50%+
- Performance Tuning Tips - Improve response speed by 60%
- Model Provider Configuration - Support for 20+ mainstream models
- Tool System Details - Extend AI capabilities
- CLI Command Full Reference - 100+ commands explained
Practical Skills
- ✅ Configure multiple AI model providers
- ✅ Configure model disaster recovery mechanisms (primary + fallbacks)
- ✅ Configure multiple authentication profiles for account rotation
- ✅ Configure memory search system
- ✅ Select the optimal model based on tasks
- ✅ Use the tool system to extend functionality
- ✅ Master CLI commands for efficient management
- ✅ Optimize costs and performance
Recommended Configuration
- Daily Use: DeepSeek (best cost-performance)
- Long Documents: Kimi (2 million character context)
- Complex Tasks: Claude 3.5 Sonnet (strong reasoning capabilities)
- Local Privacy: Ollama (fully local)
- Disaster Recovery Plan: DeepSeek → Claude Sonnet → Claude Opus
- Memory Search: Gemini Embedding (free and effective)
Next Chapter Preview: Chapter 12 will delve into practical case studies, learning complete workflows for personal productivity enhancement.