Pydantic AI + FastAPI Starter

A FastAPI microservice for AI agent interactions using Pydantic AI. This project provides a simple API wrapper around multiple LLM providers (OpenAI, Google Gemini, and Ollama) with a consistent interface.

Features

• 🚀 FastAPI-based microservice with clean architecture
• 🤖 Multiple AI agent implementations (OpenAI, Google Gemini, Ollama)
• 📝 Structured requests and responses using Pydantic models
• 🔄 Support for conversation history
• 🛠️ Custom tools for enhanced agent capabilities
• 👤 User context integration through dependencies

Installation

1. Clone this repository:

git clone https://github.com/yourusername/pydantic-ai-fastapi-starter.git
cd pydantic-ai-fastapi-starter

2. Create a virtual environment and activate it:

python -m venv .venv
# On Windows
.venv\Scripts\activate
# On macOS/Linux
source .venv/bin/activate

3. Install the dependencies:

pip install -r requirements.txt

4. Create a .env file in the root directory with your API keys:

OPENAI_API_KEY=your_openai_api_key
GEMINI_API_KEY=your_gemini_api_key

Running the Application

Start the FastAPI server with:

python -m app.main

The API will be available at http://localhost:8000.

API Endpoints

Root endpoint

• GET /: Basic test route that returns a welcome message

Agent Query Endpoint

• POST /api/v1/agent/query: Process a query using an AI agent

Request body:

{
  "query": "What is artificial intelligence?",
  "history": [], 
  "llm_provider": "openai",
  "deps": "John"
}

Supported LLM providers:

• openai: Uses OpenAI models
• gemini: Uses Google Gemini models
• ollama: Uses local Ollama models

Tools Integration

The agents in this project are enhanced with custom tools that extend their capabilities:

• get_today_date: A tool that provides the agents with the current date
• get_user_name: A context-aware tool that retrieves the user's name from dependencies

These tools can be accessed by the agents through:

1. Direct tool registration:

tools=[
    Tool(get_today_date, takes_ctx=False),
    Tool(get_user_name, takes_ctx=True),
]

2. Decorator-based registration (alternative approach):

@agent.tool()
def get_user_name(ctx: RunContext[str]) -> str:
    """Get the user's name."""
    return ctx.deps

Project Structure

pydantic-ai-microservice/
├── app/
│   ├── agents/
│   │   ├── base_agent.py
│   │   ├── gemini_agent.py
│   │   ├── ollama_agent.py
│   │   └── openai_agent.py
│   ├── api/
│   │   └── routes.py
│   ├── core/
│   │   └── config.py
│   ├── models/
│   │   └── schemas.py
│   ├── tools/
│   │   └── common.py
│   └── main.py
├── .env
├── .gitignore
├── pyproject.toml
├── README.md
└── requirements.txt

Development

This project uses Ruff for linting and formatting. Configuration can be found in pyproject.toml.

License

MIT License