Getting Started with AutoClaw: Gemini/Claude/OpenAI/OpenRouter API Integration Tutorial

AutoClaw is a stable, highly engineered, and easily scalable AI Agent framework designed specifically for headless systems. Compared to agents that rely on visual parsing, AutoClaw is purely instruction-driven, possessing stronger engineering attributes and extreme stability, making it ideal for running in servers, CI/CD pipelines, or container environments.

This article will detail how to configure LLM APIs for AutoClaw, with a primary recommendation of using the Defapi platform to achieve half-price calls.

Introduction

AutoClaw is developed based on Node.js and TypeScript, using the OpenAI SDK to interact with various LLMs. By default, it uses the official OpenAI API, but through simple configuration, you can switch to any LLM provider compatible with the OpenAI ChatCompletions interface.

GitHub Repository: https://github.com/tsingliuwin/autoclaw
Official Documentation: Supports custom API endpoint configuration via baseUrl
Tech Stack: Node.js, TypeScript, OpenAI SDK, Commander.js

Method 1: Defapi (Recommended - Half-Price Discount)

Defapi is an AI API aggregation platform providing OpenAI-compatible interfaces. Its biggest advantage is that the price is only half of the official rate. For automation scenarios requiring large-scale AI calls, this can significantly reduce costs.

Why Choose Defapi?

Price Advantage: 50% of the official price, drastically lowering usage costs.
Fully Compatible: Provides a standard v1/chat/completions interface.
Multi-Model Support: Supports GPT-4o, GPT-4o-mini, and many other models.
Stable and Reliable: Features high availability and low latency.

Configuration Steps

1. Get Defapi API Key

Visit https://defapi.org to register an account and obtain your API Key.

2. Configuration File Method

Create a .autoclaw/setting.json file in the project root directory:

{
  "apiKey": "your-defapi-key",
  "baseUrl": "https://api.defapi.org/v1",
  "model": "openai/gpt-4o"
}

Or create a global configuration file at ~/.autoclaw/setting.json:

mkdir -p ~/.autoclaw

{
  "apiKey": "your-defapi-key",
  "baseUrl": "https://api.defapi.org/v1",
  "model": "openai/gpt-4o"
}

3. Environment Variable Method

You can also configure it via environment variables:

export OPENAI_API_KEY="your-defapi-key"
export OPENAI_BASE_URL="https://api.defapi.org/v1"
export OPENAI_MODEL="openai/gpt-4o"

4. Interactive Configuration

Run the setup wizard for interactive configuration:

autoclaw setup

Follow the prompts to enter:

API Key: Your Defapi API Key
Base URL: https://api.defapi.org/v1
Model: openai/gpt-4o

Method 2: Official OpenAI API

If you already have an official OpenAI API Key, you can use it directly.

Configuration Method

Configuration File .autoclaw/setting.json:

{
  "apiKey": "sk-...",
  "baseUrl": "https://api.openai.com/v1",
  "model": "gpt-4o"
}

Environment Variables:

export OPENAI_API_KEY="sk-..."
export OPENAI_BASE_URL="https://api.openai.com/v1"
export OPENAI_MODEL="gpt-4o"

Method 3: OpenRouter

OpenRouter is an aggregation platform supporting multiple LLMs, allowing access to hundreds of models through a unified interface.

Configuration Method

Configuration File .autoclaw/setting.json:

{
  "apiKey": "sk-or-v1-...",
  "baseUrl": "https://openrouter.ai/api/v1",
  "model": "openai/gpt-4o"
}

Supported Model Examples

{
  "model": "openai/gpt-4o"
}

{
  "model": "anthropic/claude-3.5-sonnet"
}

{
  "model": "google/gemini-pro"
}

Method 4: DeepSeek

DeepSeek is a leading LLM provider from China with high cost-performance ratios.

Configuration Method

Configuration File .autoclaw/setting.json:

{
  "apiKey": "sk-...",
  "baseUrl": "https://api.deepseek.com/v1",
  "model": "deepseek-chat"
}

Method 5: Ollama (Local Models)

If you prefer to run models locally, you can use Ollama.

Configuration Method

Configuration File .autoclaw/setting.json:

{
  "apiKey": "ollama",
  "baseUrl": "http://localhost:11434/v1",
  "model": "llama3"
}

Supported Models

Ollama supports various local models, including:

llama3
mistral
codellama
phi3

Verifying Everything Works

After configuration, run the following command to verify:

Interactive Mode Test

autoclaw

Input a test message:

Hello, can you hear me?

If everything is set up correctly, AutoClaw will reply.

Headless Mode Test

autoclaw "List all files in current directory" --no-interactive

Internal Mechanism Explanation

How AutoClaw Works

At its core, AutoClaw is an agent built on the OpenAI SDK. Here are the key mechanisms:

1. Client Initialization

In src/agent.ts, AutoClaw initializes the client using the OpenAI SDK:

this.client = new OpenAI({
  apiKey: apiKey,
  baseURL: baseURL
});

The baseURL here is flexible, which is why AutoClaw can support various LLM providers seamlessly. Any service provider offering an OpenAI-compatible interface can be integrated.

2. Chat Completions Call

Every time a user provides input, AutoClaw calls the v1/chat/completions interface:

const response = await this.client.chat.completions.create({
  model: this.model,
  messages: this.messages,
  tools: getToolDefinitions() as any,
  tool_choice: "auto"
});

3. Configuration Priority

AutoClaw uses a hierarchical configuration system with priority from highest to lowest:

CLI Arguments (e.g., -m gpt-4o)
Environment Variables (OPENAI_API_KEY, etc.)
Project Configuration (./.autoclaw/setting.json)
Global Configuration (~/.autoclaw/setting.json)

This design is ideal for using different API configurations across development, testing, and production environments.

4. Tool Calling Mechanism

AutoClaw supports Function Calling to invoke various built-in tools:

Shell command execution
File reading/writing
Web search
Screenshot capture
Email sending
Notification pushing

Common Use Cases

1. CI/CD Automation

Use AutoClaw in CI/CD pipelines for code review and automated testing:

autoclaw "Review the changes in this PR" -y --no-interactive

2. Batch File Processing

Automate the processing of large numbers of files:

autoclaw "Convert all .md files to .txt files" --no-interactive

3. Data Collection and Reporting

Periodically collect data and generate reports:

autoclaw "Generate a summary report of server logs" --no-interactive

4. Containerized Batch Tasks

Run multiple parallel AutoClaw instances in Kubernetes:

# k8s deployment example
apiVersion: apps/v1
kind: Deployment
metadata:
  name: autoclaw-worker
spec:
  replicas: 10
  template:
    spec:
      containers:
      - name: autoclaw
        image: autoclaw:latest
        env:
        - name: OPENAI_API_KEY
          valueFrom:
            secretKeyRef:
              name: autoclaw-secret
              key: api-key

5. Scheduled Maintenance Tasks

Combine with cron for scheduled operations:

0 9 * * * autoclaw "Check disk usage and notify if >80%" -y

Introduction

Related Resources

Method 1: Defapi (Recommended - Half-Price Discount)

Why Choose Defapi?

Configuration Steps

1. Get Defapi API Key

2. Configuration File Method

3. Environment Variable Method

4. Interactive Configuration

Method 2: Official OpenAI API

Configuration Method

Method 3: OpenRouter

Configuration Method

Supported Model Examples

Method 4: DeepSeek

Configuration Method

Method 5: Ollama (Local Models)

Configuration Method

Supported Models

Verifying Everything Works

Interactive Mode Test

Headless Mode Test

Internal Mechanism Explanation

How AutoClaw Works

1. Client Initialization

2. Chat Completions Call

3. Configuration Priority

4. Tool Calling Mechanism

Common Use Cases

1. CI/CD Automation

2. Batch File Processing

3. Data Collection and Reporting

4. Containerized Batch Tasks

5. Scheduled Maintenance Tasks