openrouter-agent 0.1.0

This package provides integration between Pydantic AI and OpenRouter, allowing you to use any model available on OpenRouter with the Pydantic AI framework. The implementation leverages the OpenAI-compatible API provided by OpenRouter.

OpenRouter Agent for Pydantic AI

A Python library that extends the Pydantic AI framework to support OpenRouter models.

Table of Contents

• Overview
• Installation
• Configuring Your API Key via Environment Variable
  • Method 1: Using a .env File (Recommended for Projects)
  • Method 2: Setting Temporarily in Your Terminal Session
• Usage
  • Command Line Interface
  • Python API Example
  • Custom Configuration
• Class Reference
  • OpenRouterModel
  • OpenRouterAgent
• Environment Variables
• Available Models
• License

Overview

This package provides integration between Pydantic AI and OpenRouter, allowing you to use any model available on OpenRouter with the Pydantic AI framework. The implementation leverages the OpenAI-compatible API provided by OpenRouter.

Installation

# Install from the repository in editable mode to make changes
pip install -e .

# Or you can install through PyPi
pip install openrouter-agent

You'll also need to set up your OpenRouter API key. You can either:

1. Set the OPENROUTER_API_KEY environment variable
2. Pass the API key directly when initializing the agent

Configuring Your API Key via Environment Variable

Using an environment variable (OPENROUTER_API_KEY) is the recommended way to provide your API key securely. Here’s how you can set it:

Method 1: Using a .env File (Recommended for Projects)

This method keeps your key associated with your project without hardcoding it.

1. Create or Copy: In your project's main directory, create a file named .env. If your project includes a .env-sample or .env.example file, you can copy it to .env.

2. Add Your Key: Open the .env file and add the following line, replacing your_actual_api_key with the key you generated on OpenRouter Keys:

   OPENROUTER_API_KEY=your_actual_api_key
   

3. Load the Variable: For your Python code to access this variable, you usually need a library like python-dotenv. Install it (pip install python-dotenv) and load it early in your script:

   from dotenv import load_dotenv
   load_dotenv()
   # Now os.getenv("OPENROUTER_API_KEY") or libraries that check env vars will find it.
   

   (Note: Some frameworks or tools might load .env files automatically.)

Method 2: Setting Temporarily in Your Terminal Session

This is useful for quick tests but needs to be done every time you open a new terminal session.

1. Get Your Key: Copy your API key from OpenRouter Keys.

2. Run the Command: Execute the command specific to your terminal, replacing your_actual_api_key with your key:

   • Linux, macOS, Git Bash (or other sh-like shells):

     export OPENROUTER_API_KEY='your_actual_api_key'
     

   • Windows Command Prompt (cmd.exe):

     set OPENROUTER_API_KEY=your_actual_api_key
     

   • Windows PowerShell:

     $env:OPENROUTER_API_KEY = 'your_actual_api_key'
     

3. Run Your Code: Now you can run your Python script or the openrouter-agent CLI command from the same terminal window.

Important: Avoid committing your .env file (if it contains secrets) to version control like Git. Add .env to your .gitignore file.

Usage

Command Line Interface

The package includes a command-line interface (CLI) that allows you to generate and execute shell commands:

# Install the package with CLI support
pip install .

# Generate a command without executing it (copies to clipboard)
openrouter-agent query "show me all .txt files in the current directory"

# Generate and execute a command
openrouter-agent execute "show me all .txt files in the current directory"

Or run directly with Python:

# Generate a command without executing it (copies to clipboard)
python src/cli.py query "show me all .txt files in the current directory"

# Generate and execute a command
python src/cli.py execute "show me all .txt files in the current directory"

Python API Example

from openrouter_agent import Agent

# Initialize with default settings (uses openrouter/quasar-alpha model)
agent = Agent()

# Run a simple completion
result = agent.run("What is the capital of France?")
print(result)

You can also import the specific classes if needed:

from openrouter_agent import OpenRouterAgent, OpenRouterModel

Custom Configuration

from openrouter_agent import Agent

# Initialize with custom settings
agent = Agent(
    agent_name="my-custom-agent",
    openrouter_model_name="anthropic/claude-3-opus-20240229",
    temp=0.7,
    max_result_retries=5,
    openrouter_api_key="your-api-key-here"  # Or use environment variable
)

# Use the agent
result = agent.run("Explain quantum computing in simple terms.")
print(result)

Class Reference

OpenRouterModel

The OpenRouterModel class inherits from OpenAIModel and configures it to work with OpenRouter:

OpenRouterModel(
    openrouter_model_name: str = "openrouter/quasar-alpha",
    openrouter_url: str = "https://openrouter.ai/api/v1",
    openrouter_api_key: Optional[str] = os.getenv("OPENROUTER_API_KEY")
)

OpenRouterAgent

The OpenRouterAgent class inherits from Pydantic AI's Agent class and uses OpenRouterModel for completions:

OpenRouterAgent(
    agent_name: str = uuid.uuid4().hex,  # Auto-generates a UUID if not provided
    openrouter_model_name: str = "openrouter/quasar-alpha",
    temp: float = 0.1,
    max_result_retries: int = 10,
    openrouter_url: str = "https://openrouter.ai/api/v1",
    openrouter_api_key: Optional[str] = os.getenv("OPENROUTER_API_KEY"),
    instrument: bool = True,
    **kwargs: Any  # Additional arguments passed to the Agent constructor
)

Environment Variables

• OPENROUTER_API_KEY: Your OpenRouter API key

Available Models

You can use any model available on OpenRouter by specifying its identifier in the openrouter_model_name parameter. Some popular options include:

• openrouter/quasar-alpha (default)
• anthropic/claude-3-opus-20240229
• anthropic/claude-3-sonnet-20240229
• meta-llama/llama-3-70b-instruct

Check the OpenRouter documentation for a complete list of available models.

License

MIT