OpenClaw Zero Token: Use DeepSeek/Claude/ChatGPT APIs for Free via Browser

Difficulty: ⭐⭐⭐☆☆ (Medium) | Duration: 15-20 minutes | Outcome: Learn how to call mainstream AI models like DeepSeek/Claude/ChatGPT at zero cost

Target Audience

This article is aimed at developers with 1-3 years of experience. You do not need to be a large model expert, but you should:

• Be familiar with basic command line operations
• Have a basic understanding of Node.js
• Want to know how to call AI models without spending a dime

Core Dependencies and Environment

Before starting, let's prepare the development environment. You need to install the following software:

nvm install 22 && nvm use 22  

Operating System Support: macOS, Linux, Windows (WSL2). If you are using Windows, installing WSL2 is the easiest option; a single command wsl --install will do the trick.

Project Structure

Once cloned, you'll find the project structure to be very clear:

openclaw-zero-token/
├── src/
│   ├── providers/           # Authentication modules and API clients for various platforms  
│   │   ├── deepseek-web-auth.ts  
│   │   ├── deepseek-web-client.ts  
│   │   ├── claude-web-auth.ts  
│   │   └── doubao-web-client.ts  
│   ├── agents/             # Streaming response handling  
│   ├── commands/           # Authentication process commands  
│   └── browser/            # Core for Chrome automation  
├── ui/                     # Web UI (Lit 3.x)  
├── .openclaw-state/       # Local state (where credentials are stored)  
│   ├── openclaw.json      # Configuration file  
│   └── agents/main/agent/  
│       └── auth.json      # Sensitive credentials  
├── scripts/                # Utility scripts  
├── onboard.sh             # Configuration wizard  
├── server.sh              # Service management script  
└── start-chrome-debug.sh # Chrome debug startup  

Step-by-Step Tutorial

Next, we will complete the entire configuration process step by step.

Step 1: Clone the Project and Install Dependencies

First, pull the project to your local machine, then install the dependencies:

# Clone the repository  
git clone https://github.com/linuxhsj/openclaw-zero-token.git  
cd openclaw-zero-token  

# Install dependencies (pnpm will automatically handle all sub-dependencies)  
pnpm install  

Step 2: Compile the Project

The project is written in TypeScript, so you need to compile it to JavaScript before it can run:

# Compile TypeScript  
pnpm build  

Once compilation is complete, you will see a dist directory created. This is where the compiled code resides.

Next, build the Web UI (optional but recommended, as you will need it when accessing http://127.0.0.1:3001):

pnpm ui:build  

Step 3: Start Chrome Debug Mode

This is the crucial step—we need a Chrome browser that can be remotely debugged. The project's script will help you with this automatically:

./start-chrome-debug.sh  

After running, you will see a Chrome window pop up. This window is used specifically for logging into various AI platforms, and its debug port 9222 is open.

Step 4: Log into AI Platforms

In the Chrome browser that just appeared, open the websites of the AI platforms you want to use:

• DeepSeek: https://chat.deepseek.com
• Claude Web: https://claude.ai
• ChatGPT: https://chat.openai.com
• Kimi: https://kimi.moonshot.cn
• Qwen: https://chat.qwen.ai
• Doubao: https://www.doubao.com

Log in using your username and password or via QR code. You need to log in separately for each platform.

Step 5: Run the Configuration Wizard (Onboard)

Now return to the terminal and run the configuration wizard to capture the login credentials from the browser:

./onboard.sh  

Or if you prefer to run the compiled version manually:

node openclaw.mjs onboard  

The configuration wizard will prompt you to select an authentication provider. Here we will use DeepSeek as an example:

? Auth provider: DeepSeek (Browser Login)  

? DeepSeek Auth Mode:  
  > Automated Login (Recommended)  # Automatically capture credentials  
    Manual Paste                   # Manually paste credentials  

Select “Automated Login” and wait a few seconds. The wizard will automatically:

1. Connect to Chrome’s debug port
2. Visit the DeepSeek page
3. Capture your login credentials (Cookie, Bearer Token, etc.)
4. Save them to the auth.json file

If you want to configure multiple platforms, you can run ./onboard.sh multiple times and choose a different provider each time.

Step 6: Start the Gateway Service

After authentication is configured, start the core Gateway service:

./server.sh start  

Or run directly:

node openclaw.mjs gateway  

Once started, the Gateway will listen on port 3001. You can open your browser and visit http://127.0.0.1:3001 to see the Web UI interface.

./server.sh start   # Start  
./server.sh stop    # Stop  
./server.sh restart # Restart  
./server.sh status  # Check status  

Step 7: API Call in Practice

Now comes the exciting part—calling free AI models via API!

# Call the DeepSeek model  
curl http://127.0.0.1:3001/v1/chat/completions \
  -H "Authorization: Bearer YOUR_GATEWAY_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-web/deepseek-chat",
    "messages": [{"role": "user", "content": "Hello, please introduce yourself in one sentence"}]
  }'

Step 8: Switch Between Models

In the Web UI, you can use the /model command to switch between different AI models:

# Switch to Claude Web  
/model claude-web  

# Switch to Doubao  
/model doubao-web  

# Or specify a particular model version  
/model claude-web/claude-3-5-sonnet-20241022  
/model doubao-web/doubao-seed-2.0  

View all available models:

/models  

Example output:

Model                                      Input      Ctx      Local Auth  Tags
doubao-web/doubao-seed-2.0                 text       63k      no    no    default,configured,alias:Doubao Browser
claude-web/claude-3-5-sonnet-20241022      text+image 195k     no    no    configured,alias:Claude Web
deepseek-web/deepseek-chat                 text       64k      no    no    configured

Supported Models Overview

The project has been tested with a wide range of platforms and models:

Common Troubleshooting

Issue 1: Cannot Connect to Chrome Debug Mode

Symptom: After running ./start-chrome-debug.sh, onboard reports it cannot connect to the browser.

Solution:

1. Check whether the Chrome process is running: ps aux | grep Chrome
2. Confirm that port 9222 is listening: lsof -i :9222
3. If the port is occupied, kill the process: kill -9 <PID>

Issue 2: Credential Capture Fails

Symptom: The onboard process waits for a long time and eventually times out.

Solution:

1. Ensure you have completed login in Chrome
2. Refresh the target AI platform page
3. Re-run ./onboard.sh and choose “Manual Paste” mode to copy cookies manually

Issue 3: API Call Returns 401

Symptom: The curl request returns unauthorized.

Solution:

1. Check whether the token is correct: find gateway.auth.token in .openclaw-state/openclaw.json
2. Confirm that the Gateway service is running: ./server.sh status
3. Try restarting the service: ./server.sh restart

Issue 4: Session Expired

Symptom: API calls suddenly fail with login-related errors.

Solution:

1. Open the previously started Chrome debug window
2. Log in again to the corresponding AI platform
3. Run ./onboard.sh again to recapture credentials

Issue 5: Model Switch Not Taking Effect

Symptom: After switching models, the API still returns results from the previous model.

Solution:

1. Restart the Gateway service: ./server.sh restart
2. Check that the model name is correct (case-sensitive, including slashes)
3. Use /models to confirm the model appears in the list

Issue 6: Port 3001 Already in Use

Symptom: Starting the Gateway shows “Port 3001 already in use.”

Solution:

1. Find the process occupying the port: lsof -i :3001
2. Kill it: kill -9 <PID>
3. Or modify the configuration file to use a different port

Advanced Directions

Once you’ve mastered the basics, you can explore more advanced features:

1. Tool Calling

All supported tool-calling models can execute local commands, read/write files, and automate browsers. This means AI can not only answer questions, but also actually help you perform tasks—such as writing code, manipulating files, or even controlling a browser.

2. Configure Multiple Platforms Simultaneously

You can configure multiple platforms in separate onboard runs, then switch anytime using /model. For example, use Claude for coding in the morning, DeepSeek for reasoning in the afternoon, and Kimi for reading long documents at night.

3. Custom Providers

The project is open source. You can add support for new platforms in the src/providers/ directory. As long as you follow the existing template to implement authentication and an API client, you can integrate any AI model with a web version.

4. Security Hardening

How It Works (Brief Overview)

You might wonder: how does this achieve “token-free” API calls? The core principle is actually straightforward:

1. Browser Automation: Use Playwright to control Chrome and connect via the CDP (Chrome DevTools Protocol)
2. Credential Capture: Listen to network requests, extract Authorization headers, and obtain session IDs from cookies
3. API Forwarding: Use captured credentials to send requests to the platform’s Web API, then convert responses into an OpenAI-compatible format

That’s why you must keep the Chrome debug window running—all requests are sent from that browser context.