GPT-5 Codex CLI

A lightweight terminal client that mirrors the experience of products like OpenAI's Codex CLI, Claude Code, and Gemini Code. The tool focuses on high-level task execution (not just code generation) by defaulting to the gpt-5-codex model with both streaming output and the web search tool enabled.

Features

• Autodetects OPENAI_API_KEY and guides you to create one if it is missing.
• Streaming responses that closely resemble established AI coding assistants.
• Web search tool enabled by default so the model can pull in fresh information as needed.
• Clean interactive loop for multi-turn task planning and execution, plus one-shot mode via CLI arguments.

Installation

pip install -e .

This project depends on the official OpenAI Python SDK. Install it automatically through the editable install above or explicitly with pip install openai.

Usage

Interactive session:

export OPENAI_API_KEY="sk-..."
gpt5-codex

Single task run:

gpt5-codex "Plan a two-day itinerary for Tokyo that balances food, art, and tech"

Optional flags:

• --model MODEL_NAME – override the default gpt-5-codex model.
• --temperature FLOAT – tune the generation temperature (default 0.2).
• --no-search – disable the built-in web search tool.
• --no-stream – disable streaming output.

Testing

pip install -e .[dev]
pytest

Publishing to PyPI

The repository includes a standard pyproject.toml. To publish:

python -m build
python -m twine upload dist/*

Note: Uploading requires a PyPI account and network access, which may not be available in every environment.