Optimizing Custom Agents¶

This guide explains how to use AIAI CLI to optimize your own custom AI agents. While the demo agent is useful for learning, the real power of AIAI CLI comes from optimizing your own agent code.

Prerequisites¶

Before optimizing your custom agent, ensure:

Your agent has a main entry point file
The entry point file has a main(inputs=None) function that runs your agent with the provided inputs (the inputs parameter is optional)
AIAI CLI is installed and configured

Creating an Entrypoint File¶

AIAI CLI requires a specific entrypoint file structure to interact with your agent. This file must:

Have a main(inputs=None) function that accepts an optional input example
Run your agent with the provided inputs
Return the agent's output

Here's what a minimal entrypoint file should look like:

# entrypoint.py
def main(inputs=None):
    # Initialize your agent
    agent = get_your_agent()

    # Use the provided inputs or a default one
    inputs = inputs or "Default inputs input here"

    # Run your agent with the inputs
    result = agent.run(inputs)  # Or however your agent accepts inputs

    # Return the result
    return result

AIAI CLI will call this main() function with different synthetic examples during optimization.

Preparing Your Agent¶

To prepare your agent for optimization:

Document Dependencies: Ensure all dependencies are properly installed
Test Your Agent: Make sure your agent runs correctly before optimization

Running the Optimization¶

To optimize your custom agent:

# Run AIAI CLI
aiai

# When prompted, select option 2 to optimize your own agent

When prompted, provide the path to your agent's entry point file. AIAI CLI will:

Validate your entrypoint file
Analyze your agent's code
Generate synthetic test data
Run your agent with multiple examples
Extract optimization rules
Generate a comprehensive report