Skip to main content

​
Overview

The basic OpenAI demo is the smallest possible ICRL example. It uses a trivial environment that succeeds after one action, runs a single training episode, and prints the result. Use it to confirm your API key and environment are working.

​
Prerequisites

  • OPENAI_API_KEY set in your environment
  • Python 3.12+
  • icrl-py installed

​
Run

uv run python examples/basic_openai_demo.py
Optional model override (default is gpt-4o-mini): 
MODEL=gpt-4o-mini uv run python examples/basic_openai_demo.py

​
What It Does

  1. Creates a MinimalEnvironment that accepts any action and returns success
  2. Creates an Agent with LiteLLMProvider and basic prompts
  3. Runs one agent.train() call with goal "Demonstrate one minimal training run"
  4. Prints model, success status, and step count

​
Expected Output

Model: gpt-4o-mini
Success: True
Steps: 1

​
Source

examples/basic_openai_demo.py — about 60 lines, including the minimal environment definition.