Symfony AI Agent Best Practices Demo

This project is a demonstration of the best practices for using the symfony/ai-agent component in a Symfony 7.4 application, as inspired by the official Symfony AI initiative. It showcases how to build robust, testable, and maintainable AI features with clean architecture.

Features

• Dependency Injection: Agents are configured in YAML and injected into services.
• Type-Safe Tools: Demonstrates the #[AsTool] attribute with strictly typed parameters using PHP Enums.
• Dynamic Context Injection: Shows how to use an InputProcessor to dynamically add context to an AI prompt.
• Testability: Includes a unit test using the MockAgent to test AI-integrated services without making real API calls.

Installation

1. Clone the repository:

   git clone https://github.com/mattleads/AIAgentSample.git
   cd AIAgentSample

2. Install dependencies using Composer:

   composer install

Configuration

1. Create a local environment file: Copy the distributed environment file .env to .env.local. This file is ignored by Git and will store your secret API keys.

   cp .env .env.local

2. Set your OpenAI API Key: Open the .env.local file and replace the placeholder value with your actual OpenAI API key.

   # .env.local
   OPENAI_API_KEY=sk-proj-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Usage Examples

This project includes several console commands to demonstrate the different features of the symfony/ai-agent component.

1. Basic Agent Interaction

This command tests the default AI agent configured in config/packages/ai.yaml.

Command:

php bin/console app:test-agent

Expected Output: You will see a direct response from the LLM based on the prompt "Hello Symfony!".

Hello there! I'm a helpful Symfony assistant, ready to assist you with your technical questions. How can I help you today?

2. Using AI Tools (OrderStatusTool)

This command demonstrates how the AI can use a custom tool to perform an action. The agent will interpret the natural language prompt, identify the correct tool (get_order_status), and call it with the provided parameters.

Command:

php bin/console app:test-tool 12345 eu

Expected Output: The agent will use the OrderStatusTool to retrieve the fictional order status.

Prompting agent: What is the status of order 12345 in the eu region?
Agent response:
Order 12345 in region eu is currently: SHIPPED

3. Dynamic Context (UserContextProcessor)

This command demonstrates how the UserContextProcessor can dynamically inject information into the prompt before it is sent to the AI. The command simulates a logged-in user to provide context.

Command:

php bin/console app:test-context

Expected Output: The AI will respond with knowledge of the user's identity, which was provided by the context processor.

Simulated logging in user: cli.user@example.com

Prompting agent: Who am I?
Agent response:
You are cli.user@example.com (ID: 1).

Note: The UserContextProcessor added the user info to the AI prompt behind the scenes.

Running Tests

This project includes a unit test for the SupportAssistant service that uses a MockAgent to avoid making real API calls.

To run the test suite, execute the following command:

vendor/bin/phpunit