kardolus/chatgpt-cli

ChatGPT CLI is a powerful, multi-provider command-line interface for working with modern LLMs. It supports OpenAI, Azure, Perplexity, LLaMA, and more, with features like streaming, interactive chat, prompt files, image/audio I/O, MCP tool calls, and an experimental agent mode for safe, multi-step automation.

ChatGPT CLI – a terminal front‑end for modern LLMs

What it is – A single‑binary command‑line tool that lets you talk to a variety of large‑language‑model providers (OpenAI, Azure, Perplexity, LLaMA, etc.) directly from the shell. It supports plain queries, streaming responses, interactive REPL sessions, and a set of “agent” capabilities that let the model call tools (shell, file system, web search, MCP services) to solve multi‑step tasks.

Key capabilities

  • Streaming & query modes – real‑time token‑by‑token output or one‑shot ask‑and‑receive.
  • Thread‑based context – each logical conversation (thread) keeps its own history with a sliding‑window token limit; history is stored under ~/.chatgpt-cli.
  • Prompt files – --prompt <file> injects a reusable instruction file before your query.
  • Agent mode – experimental ReAct or Plan/Execute loops that let the model think, act, and observe using built‑in tools (shell commands, file ops, web search). Budgets (steps, tokens, time) and policy rules (allowed tools, work‑dir sandbox) keep execution safe.
  • MCP (Model Context Protocol) integration – call external HTTP or stdio tools, have their JSON‑RPC result inserted into the conversation as [MCP: <tool>] context.
  • Multimedia I/O – upload images (--image), generate or edit images (--draw), upload audio (--audio), transcribe (--transcribe), and synthesize speech (--speak).
  • Model discovery – -l/--list-models shows what models the current provider offers.
  • Layered configuration – flags > env vars > config.yaml > defaults; per‑provider sections let you store keys for OpenAI, Azure, Perplexity, 302 AI, Atlas Cloud, etc.

Installation

  • Homebrew (macOS): brew tap kardolus/chatgpt-cli && brew trust kardolus/chatgpt-cli && brew install chatgpt-cli
  • Pre‑built binaries – curl the appropriate chatgpt-<os>-<arch> from the latest GitHub release, make it executable, and place it on your $PATH (examples for Apple Silicon, Intel macOS, Linux amd64/arm64/386, FreeBSD, Windows are in the README).

Getting started

  1. Export your OpenAI key: export OPENAI_API_KEY=sk‑…
  2. Create the history folder: mkdir -p ~/.chatgpt-cli
  3. Try a simple query: chatgpt "What is the capital of the Netherlands?"
  4. Launch an interactive REPL: chatgpt --interactive
  5. Use a prompt file: git diff | chatgpt --prompt ./prompts/write_pull-request.md
  6. Run an agent: chatgpt "Why is my test failing?" --agent

Configuration highlights

  • config.yaml lives in ~/.chatgpt-cli (or a custom directory via --config-dir).
  • Provider‑specific sections let you set API keys, base URLs, and model defaults.
  • Agent‑specific settings (budget, policy, work-dir) are configurable in the same file or via --set-… flags.
  • Autocompletion can be enabled for Bash/Zsh/Fish with the --completion helper.

Why you might care

  • Keeps you in the terminal, avoiding context switches to a web UI.
  • The agent mode gives you a lightweight “auto‑GPT” that can run scripts, edit files, or fetch live web data while respecting safety limits.
  • MCP support means you can plug in any custom tool that follows the Model Context Protocol without writing code.

Links & resources

Related

  • Project
    Niek/chatgpt-web

    ChatGPT‑web is an open‑source, single‑page web UI that lets you chat with any OpenAI‑compatible LLM (OpenAI, Anthropic, Gemini, etc.). Conversations are stored locally, the app supports voice I/O, image generation, streaming responses, and can be run in the browser, via Docker, or packaged as a native desktop app.

  • Project
    appleboy/CodeGPT

    CodeGPT is a Go‑based CLI that uses LLM APIs (OpenAI, Azure, Gemini, Anthropic, Ollama, Groq, OpenRouter, LiteLLM) to auto‑generate Conventional‑Commit messages and short code‑review summaries. It installs a Git `prepare‑commit‑msg` hook, supports many providers, proxy settings, language translation, streaming output, and a flexible prompt/template system. Install via Homebrew, Chocolatey, an install script, pre‑compiled binary, or `go install`. Configure API keys (or dynamic helpers), choose models, and customize prompts. Ideal for developers who want AI‑assisted commit messages and quick reviews directly from the terminal.

  • Dispatch
    Getting Started with ChatGPT: Guide to Core Features and Workflows

    OpenAI provides a foundational guide to ChatGPT, detailing the distinction between Chat and Work modes, prompt engineering basics, and the integration of voice capabilities for enhanced productivity.

  • Project
    miuuyy/codex-chatgpt-web

    A bridge that integrates ChatGPT Web models into the Codex workflow, allowing users to use their ChatGPT account plans and tools via browser automation and MCP.

  • Project
    labring/FastGPT

    FastGPT is an open‑source platform for building AI agents and RAG‑enabled applications. It offers a visual workflow editor, built‑in knowledge‑base handling, plugin hot‑reload, debugging tools, and can be launched instantly with Docker or via a hosted cloud service.