openai-cli

A fast, minimal command-line client for the OpenAI Responses API, written in Rust.

---

Features

• Direct prompts -- inline text, files, stdin, or $EDITOR
• Interactive REPL -- multi-turn conversations with history
• Streaming -- tokens print as they arrive
• JSON mode -- machine-readable output with --json / -j
• Clipboard -- copy responses with --copy / -c
• Retry logic -- automatic exponential backoff on rate limits and server errors
• Reasoning effort -- --reasoning / -r for reasoning models (o3, o4-mini)
• Web search -- enabled by default; the model searches when helpful, disable with --no-web
• System prompt -- --system / -s to override default instructions per request
• Structured logging -- --verbose / -v or RUST_LOG=debug for tracing output
• Single binary -- no runtime dependencies

Installation

Homebrew (Recommended for macOS/Linux)

brew install Skyline-9/tap/openai-cli

Pre-built binary

Platform	Architecture	Download
macOS	Apple Silicon (arm64)	openai-darwin-arm64
Linux	x86-64	openai-linux-amd64
Windows	x86-64	openai-windows-amd64.exe

# macOS
curl -Lo openai https://github.com/Skyline-9/openai-cli/releases/download/v0.3.1/openai-darwin-arm64
chmod +x openai && mv openai ~/.local/bin/

# Linux
curl -Lo openai https://github.com/Skyline-9/openai-cli/releases/download/v0.3.1/openai-linux-amd64
chmod +x openai && mv openai ~/.local/bin/

From source

git clone https://github.com/Skyline-9/openai-cli.git
cd openai-cli
cargo build --release
ln -sf "$(pwd)/target/release/openai" ~/.local/bin/openai

Requires Rust 1.80+ and an OpenAI API key.

Quick start

export OPENAI_API_KEY="sk-..."

Inline prompt

openai ask -p "Explain quicksort in one sentence"

From a file

openai complete --file prompt.txt
# or
openai complete -p @prompt.txt

Heredoc

openai ask -p @- <<'PROMPT'
Explain quicksort in one sentence,
and then give a one-line pseudocode summary.
PROMPT

Open your editor

openai ask --editor

Interactive REPL

openai chat

Usage

Usage: openai [OPTIONS] <COMMAND>

Commands:
  complete  Send a prompt and get a completion [aliases: ask]
  repl      Start interactive REPL with conversation history [aliases: chat]

Options:
  -m, --model <MODEL>              OpenAI model to use [default: gpt-5.2]
  -k, --max-tokens <MAX_TOKENS>    Maximum number of tokens in the response [default: 16384]
      --temperature <TEMPERATURE>  Temperature for response generation [default: 0.23]
  -t, --token <TOKEN>              OpenAI API token (overrides OPENAI_API_KEY)
  -j, --json                       Output raw JSON instead of streaming text
  -r, --reasoning <EFFORT>         Reasoning effort: low, medium, high (for o3/o4-mini)
  -s, --system <PROMPT>            System prompt / instructions (overrides default)
      --no-web                     Disable web search (enabled by default)
  -v, --verbose                    Enable verbose/debug logging
  -h, --help                       Print help
  -V, --version                    Print version

Prompt input precedence

-p > positional arg > --file > --editor > stdin (pipe)

Examples

# Different model
openai -m gpt-4o ask -p "Hello"

# High creativity
openai --temperature 0.9 ask -p "Write a haiku about Rust"

# JSON output piped to jq
openai -j ask -p "What is 2+2?" | jq .text

# Copy response to clipboard
openai ask -c -p "Draft a commit message for this diff"

# Custom system prompt
openai -s "You are a Python expert. Reply with code only." ask -p "fizzbuzz"

# Reasoning model with high effort
openai -m o4-mini -r high ask -p "Prove that sqrt(2) is irrational"

# Debug logging
openai -v ask -p "test"

Configuration

Environment Variable	Description	Default
OPENAI_API_KEY	API key (required)	--
OPENAI_API_URL	Override the API endpoint	https://api.openai.com/v1/responses
RUST_LOG	Logging filter (e.g. debug)	warn
EDITOR / VISUAL	Editor for --editor flag	vi

CI

The project uses a GitHub Actions workflow that runs cargo fmt --check, cargo clippy, and cargo test on every push.

License

MIT