What it does

chatGPT-shell-cli is a single bash script that hits OpenAI’s chat and image endpoints. You chat, generate images, list models, or even request shell commands — all from the prompt you already have open. It pipes, it parameters, it runs. Dependencies are just curl and jq.

The interesting bit

The project leans into shell minimalism rather than fighting it. Chat context for non-chat models is hand-rolled in bash: the script prepends previous Q&A pairs and a system prompt so older endpoints feel stateful. There’s even a command: prefix that generates shell commands and asks before executing — with a warning if filesystem changes are involved.

Key highlights

• Supports gpt-3.5-turbo, gpt-4, and legacy completion models via flags
• Pipe mode: echo "question" | chatgpt for scripting workflows
• Image generation via image: prefix; iTerm users get inline previews
• Custom initial prompts from file or flag for “stay in character” sessions
• Arch Linux AUR package available; otherwise one curl | sudo bash install

Caveats

• The default install method is curl ... | sudo -E bash, which the README presents without checksum verification
• Chat context for non-turbo models is client-side and unbounded; long sessions will grow tokens and cost

Verdict

Worth a look if you live in tmux and resent installing a Python env just to ask GPT-4 a quick question. Skip it if you need streaming responses, plugins, or anything resembling a real TUI.