Typing Commands in the Terminal for AI to Write Code? 3363 People Say: I'd Rather Use a Mouse

Claude Code, Codex CLI, OpenCode — these AI programming tools share one thing: you have to stare at a black terminal and type commands line by line.

Not that they're bad. They're just too primitive.

You ask AI to modify three files, run two commands, and spend $0.47 — this information is scattered in the scrolling output of the terminal, and you have to dig through it yourself. Say something wrong? No undo, no rollback, you can only start a new conversation. Want to check the project structure? Switch out to open Finder. Want to commit code? Switch out to open the terminal.

3363 developers couldn't take it anymore. They gave a project called ccgui a star.

One Client, Three AI Engines Under One Roof

What ccgui does is simple: it puts Claude Code, Codex CLI, OpenCode — these command-line AI programming tools — into a graphical interface.

It's not just a wrapper. It's a redesign of the interaction.

Within the same project, you can mix and match different engines — Claude Code for logic, Codex CLI for testing, OpenCode for refactoring. Switch at any time without closing and reopening. API providers are flexible too: official, relay, aggregated platforms — each engine can store multiple provider configurations.

Close the app and reopen it, conversations are still there. Interrupted? You can continue right where you left off. How much context each session has used is clearly visible.

Chat Box Designed for Writing Code

Normal chat boxes send text. ccgui's chat box sends code.

The input box supports @ to reference project files — @src/utils/auth.ts lets the AI know which file you want to modify. Slash commands, pasting images, uploading attachments — all supported.

The AI's workflow is transparent: which files were modified, what commands were run, what content was read — all displayed in real-time cards. Not a messy terminal output, but a structured information stream.

Said something wrong? Rewind to a previous state. Want to try a different path? Fork from a certain node and start over. This is something the terminal can't do.

Too lazy to type? Voice input. Can't write good prompts? Built-in prompt enhancement helps polish them. While the AI is working, you can queue up the next question.

Not Just Chat, a Full Development Panel

What surprised me most about ccgui is not the chat box, but the panel beside it.

File Tree: browse, preview, copy, paste, rename, drag files directly into the conversation. No need to switch windows.

Built-in Terminal: full terminal experience. When AI runs commands, you see real-time output; when you want to run a command yourself, no need to switch out.

Git Panel: stash, commit, branches, worktree, view diffs, browse commit history. The AI helps write commit messages. Yes, it can really generate commit messages based on your diff.

Global Search: files, sessions, historical messages, skills, commands — all in one search box.

Stay Organized When Tasks Pile Up

AI programming has a problem: when tasks pile up, things get out of control. ccgui uses four things to fix this.

Plan Panel: the AI's execution plan is listed step by step — you can see at a glance what's done and what's next. Not a black box.

Kanban Board: drag and drop task cards to manage the entire iteration. Yes, a Kanban board built right into an AI programming tool.

Task Center: every AI execution is recorded. Retry on failure, browse execution logs anytime.

Intent Canvas: drag nodes on a canvas to plan — think before you act. This feature is very new, but the approach is right: think first, then do, which is better than do first, then fix.

Project Intelligence: What Makes ccgui Unique

The features above are also being done by other tools. The following ones are where ccgui truly differs.

Project Knowledge Map: AI scans your project and generates an interactive structural graph. File relationships, API interfaces, module dependencies — all clear at a glance. Supports incremental updates: when you change code, the graph updates accordingly.

Project Memory: store key project conventions and lessons learned as long-term memory. The AI remembers next time you open it. This solves a real pain point: every new conversation starts with the AI knowing nothing about your project.

Context Ledger: exactly what context the AI used in its response, and how much each part occupies — all transparent. You know whether it's guessing.

Usage Statistics: token consumption, costs, cache hit rates — all reported. You can even set a monthly budget cap. What's the biggest fear of using AI programming? Checking the bill at the end of the month. This feature cured my anxiety.

Extensions and Personalization

MCP Marketplace, Skills Marketplace, Plugin Marketplace: one click to install, giving the AI new capabilities. MCP protocol lets the AI connect to external tools, Skills let the AI learn specific workflows, plugins extend interface functions.

Browser Agent: lets the AI read web content. No need to copy-paste documentation; just let the AI read it directly.

Themes: 15+ built-in themes, VS Code style. Customizable colors, window transparency, font size.

Bilingual interface (Chinese/English), all shortcuts customizable.

Full support for macOS / Windows / Linux, with in-app auto-update.

Tech Stack

Tauri 2 + React 19 + TypeScript + Rust.

All data stored on your own computer. No uploads, no tracking.

Open source under the MIT license.

How to Install

Go to the Releases page and download the installer for your platform:

After installation, configure the AI engine's API key or local CLI in settings, add your project folder, and you're ready to go.

3363 stars, 280+ forks, MIT license. This is not a demo; it's a tool that people actually use.

The terminal won't disappear. But the interaction method of AI programming is due for an evolution.

Project URL: https://github.com/zhukunpenglinyutong/desktop-cc-gui