OpenAI in Python

A Python project for interacting with LLMs via the OpenAI SDK and OpenRouter's free models. Includes an interactive REPL chat and a temperature comparison tool.

Project Reference: OpenAI API in Python - roadmap.sh

Project Structure

OpenAI in Py/
├── .env                      # API key (not tracked)
├── .gitignore
├── requirements.txt
├── main.py                   # Interactive chat REPL
├── compare_temps.py          # Temperature comparison tool
└── openai_in_py/
    ├── __init__.py
    └── client.py             # Shared client, retry logic, defaults

Prerequisites

• Python 3.9+
• An OpenRouter account with an API key

Setup

# Clone or download the project
cd "OpenAI in Py"

# Create and activate virtual environment
python -m venv .venv
.\.venv\Scripts\Activate.ps1

# Install dependencies
pip install -r requirements.txt

Configuration

Create a .env file in the project root:

OPENAI_API_KEY=sk-or-v1-your-key-here

Get your key from OpenRouter Settings.

Usage

Interactive Chat (REPL)

python main.py

Model: openai/gpt-oss-120b:free
Temperature: 0.7
Max tokens: 1024
System: You are a helpful assistant.
----------------------------------------
Type /quit to exit.

You > Hello
Assistant > Hi there! How can I help you today?

You > /quit
Bye!

CLI Flags

Flag	Default	Description
--model	openai/gpt-oss-120b:free	OpenRouter model ID
--temperature	0.7	Randomness (0.0 = deterministic, 2.0 = max creative)
--max-tokens	1024	Max response length
--system	"You are a helpful assistant."	System message

Examples

# Use a different model
python main.py --model meta-llama/llama-3.3-70b-instruct:free

# Set custom temperature and max tokens
python main.py --temperature 0.3 --max-tokens 512

# Roleplay as a pirate
python main.py --system "You are a pirate. Respond only in pirate speak."

---

Temperature Comparison

python compare_temps.py "Tell me a joke in one sentence"

Note: The default model (gpt-oss-20b) is a standard chat model that respects temperature. Reasoning models like gpt-oss-120b ignore temperature and will produce near-identical outputs across values. Free-tier models can be slow — use --model to switch if needed.

Model: openai/gpt-oss-20b:free
Prompt: Tell me a joke in one sentence
Temperatures: [0.0, 0.7, 1.5, 2.0]
============================================================

--- Temperature: 0.0 ---

Why did the scarecrow win an award? Because he was outstanding in his field!

--- Temperature: 0.7 ---

Why did the scarecrow become a motivational speaker? Because he was outstanding in his field!

--- Temperature: 1.5 ---

What do you call a fake noodle? An impasta!

--- Temperature: 2.0 ---

Why did the scarecrow win an award? Because he was outstanding in his field.

CLI Flags

Flag	Default	Description
prompt (required)	—	The prompt to send
--model	openai/gpt-oss-20b:free	OpenRouter model ID
--max-tokens	1024	Max response length
--system	"You are a helpful assistant."	System message
--temps	0.0 0.7 1.5 2.0	Space-separated temperature values

Examples

# Custom temperatures
python compare_temps.py "Write a haiku about coding" --temps 0.0 0.5 1.0 1.5 2.0

# Different model
python compare_temps.py "Explain quantum physics" --model meta-llama/llama-3.3-70b-instruct:free

---

Available Free Models

All models below cost $0 on OpenRouter:

Model ID	Size	Best For
openai/gpt-oss-120b:free	117B MoE	General purpose (default)
nvidia/nemotron-3-ultra-550b-a55b:free	550B MoE	Complex reasoning
qwen/qwen3-coder:free	480B MoE	Code generation
qwen/qwen3-next-80b-a3b-instruct:free	80B MoE	Chat
meta-llama/llama-3.3-70b-instruct:free	70B	General purpose
openai/gpt-oss-20b:free	21B MoE	Fast responses
openrouter/free	—	Auto-routes to any free model

Full list: openrouter.ai/models

---

How It Works

1. OpenAI SDK — The project uses the official openai Python package
2. OpenRouter routing — Requests are sent to OpenRouter's API (https://openrouter.ai/api/v1) which routes to the chosen model provider
3. Shared client — openai_in_py/client.py provides build_client() and chat_with_retry() used by both scripts
4. Environment variables — The API key is loaded from .env via python-dotenv
5. Retry logic — Free models are rate-limited. Both scripts automatically retry up to 3 times with the provider's suggested wait time
6. History capping — The REPL limits conversation history to 20 messages to avoid token limits
7. Request timeout — 60-second timeout prevents hanging on slow free-tier responses

Package Architecture

openai_in_py/client.py
├── build_client()          → OpenAI client configured for OpenRouter
├── chat_with_retry()       → API call with automatic retry on 429
├── DEFAULT_MODEL           → "openai/gpt-oss-120b:free"
├── DEFAULT_TEMPERATURE     → 0.7
├── DEFAULT_MAX_TOKENS      → 1024
├── DEFAULT_SYSTEM          → "You are a helpful assistant."
├── MAX_RETRIES             → 3
└── REQUEST_TIMEOUT         → 60

API Key Safety

The .env file is included in .gitignore and will not be committed. Never share your API key publicly.

License

This project is for educational purposes.