I Used Claude Code for 3 Months and Only Then Realized It Was Missing a Leg

You might not believe it, but I used Claude Code for three months before learning it had a panel mode.

Not the mild surprise of "oh, so this feature exists," but the facepalm of "what was I even doing before?" Like using an iPhone for a year and suddenly discovering you can customize Control Center — it's not that you couldn't use it, it's that you never knew it existed.

Claude Code in the terminal walks on one leg. It can write code, fix bugs, and run commands, but you can't see the file tree, task progress, cost breakdown, or conversation history timeline. You just stare at the blinking cursor and pray the AI doesn't go off track.

Until I installed ccgui.

Not a Skin, a Different Track

Let me be clear: ccgui is not just a skin for Claude Code.

It's a standalone desktop client that packs four engines — Claude Code, Codex CLI, OpenCode, and Gemini CLI — into one interface. No need to open four terminal windows, memorize four sets of shortcuts, or manage four configurations. Open ccgui, pick an engine, and start chatting.

It's like before you ordered takeout from four different restaurants, and now you have a food court — one seat, four kitchens to choose from.

The project has 3363 stars on GitHub, MIT license, 100% open source. Built with Tauri 2 + React 19 + TypeScript + Rust, fully cross-platform for macOS, Windows, and Linux. All data stays local, nothing is uploaded.

Download: https://www.mossx.ai/download

Four Engines, Four Personalities

Claude Code

The veteran. Opus for architecture analysis, Sonnet for daily coding, Haiku for quick queries. Context understanding is still top-tier, long-running tasks rarely drift off.

Codex CLI

From OpenAI. Fast response, efficient for code completion and simple refactoring. Great for lightweight tasks like "rename this interface" or "add a field" — no need to use an Opus-level brain.

OpenCode

Open source solution, supports Gemini and other models. If you don't want to be tied to any single vendor, OpenCode has your back.

Gemini CLI

Google's Gemini model command-line tool. Long context, multimodal understanding, suitable for scenarios needing "look at an image and write code."

These four engines aren't just for show — you can actually mix them in the same project. Use Claude Code for complex refactoring, Codex for quick completions, Gemini for image-based generation — switch as needed without reconfiguring your environment.

Panel Mode: Finally Seeing What the AI Is Doing

This is ccgui's core differentiating feature.

When using Claude Code in the terminal, the AI changes files, runs commands, and costs money — you can only dig through logs afterward. ccgui's panel mode visualizes all of this:

File Tree Panel: See which files the AI changed and what was changed at a glance. No need to go through git diff line by line.

Task Panel: Real-time display of the AI's execution plan, current progress, and to-do items. No longer a black box.

Sub-agent Panel: Full traceability of which tools the AI called and which subtasks were executed.

Cost Panel: Real-time statistics on tokens consumed and costs per conversation. No more heart-stopping monthly bills.

These four panels are something the terminal can never give you. Not that the terminal is bad, but its information density has a ceiling — it can only display a stream of text, not multi-dimensional state simultaneously.

Conversation Canvas: Chat Box Built for Writing Code

ccgui's chat box is not a stripped-down version of a general chat; it's deeply customized for coding scenarios:

@ File Reference: Type @ to directly reference project files, and the AI automatically reads the file content as context. No need to manually copy-paste code.

Rewind/Fork: Conversation went off track? Rewind to any node and start over. Want to try a different approach? Fork a branch to explore in parallel. In the terminal, you can only Ctrl+C and start over.

Voice Input: Too lazy to type? Just speak. Describing complex requirements by voice is three times faster than typing.

Prompt Enhancement: Built-in prompt optimization that automatically completes your vague description into a more precise instruction.

Queued Questions: While the AI is executing a task, you can type the next question in advance and add it to a queue. No need to wait for the AI to finish before thinking of the next step.

Project Intelligence: The AI Finally Knows Your Project

This is ccgui's hidden trump card.

With ordinary AI programming tools, each new conversation starts with amnesia. ccgui's project intelligence feature lets the AI truly understand your project:

Knowledge Map: Automatically scans the project structure and generates a code relationship graph. The AI knows which files depend on each other, so changing one won't cause a cascade of issues.

Project Memory: Remembers project conventions, architecture decisions, and tech stack preferences across sessions. No need to repeat "We're using React 19 + TypeScript" every time.

Context Ledger: Tracks context consumption per conversation to help you optimize token usage. Which files consume the most context, which conversations cost the most tokens — all clear.

Usage Statistics: Token consumption and costs by day, week, or project. In team collaboration, who used how much is clear at a glance.

MCP + Skills: The AI's Toolbox and Skill Pack

MCP Server: ccgui has built-in support for the Model Context Protocol (MCP). Configure MCP servers in the settings page, and the AI can call external tools — database queries, API tests, file system operations — all within the conversation.

Skills System: Includes a curated set of skill packs that can be activated with one click. Code review, test generation, documentation writing — no need to craft your own prompts, just select a skill.

Plugin Marketplace: Community-contributed plugins are continuously updated to extend the AI's capabilities.

Settings Page: Everything You Need

ccgui's settings are not a half-hearted "it can be configured" — they truly consider developer needs:

• Multi-engine configuration, each with independent API key and model preferences
• MCP server management, clear CRUD operations
• Theme switching, 15+ themes available, both dark and light
• Custom font, font size, and line height
• Keyboard shortcut binding
• Proxy settings (HTTP/SOCKS5 support)
• Data export and backup

The MCP settings page alone is richer than the entire settings page of many "AI programming tools."

Update Cadence: A New Version Every 4 Days

Looking at the CHANGELOG, ccgui releases a new version roughly every 4 days. v0.5.14 was released on June 26, v0.5.12 on June 22, v0.5.11 on June 21.

Each version isn't just a version number bump — they actually fix bugs and add features:

• v0.5.14: Fixed tool call stuttering, curated skills, fixed cold start empty state loop
• v0.5.12: Support for full-screen images and Mermaid charts, fixed React maximum update depth
• v0.5.11: Claude vendor sorting, model fetching, streaming rendering performance optimization

This iteration speed shows the team is seriously building the product, not just doing a KPI-driven "monthly update."

Honest Limitations

ccgui is currently at v0.5.x, not 1.0. That means:

• Occasional bugs. 196 open issues aren't just decoration, though most are feature requests
• Some advanced features are still being polished, like full sub-agent chain tracing
• The Windows experience is slightly worse than macOS, some features are still being adapted
• You need to install the corresponding CLI tools locally (Claude Code, Codex, etc.) — ccgui is a client, not a replacement

But given the 4-day update cadence, these issues are likely half-fixed by the time you read this article.

Walking on Two Legs

The terminal won't disappear. The flexibility and programmability of CLI can never be replaced by GUI.

But GUI can give you what the terminal can't: visualization, multi-dimensionality, traceability. You don't have to choose between the "terminal camp" and the "GUI camp" — ccgui doesn't replace the terminal, it adds a leg to it.

A leg that can see the file tree, track task progress, tally cost details, and manage multiple engines.

Open source: https://github.com/zhukunpenglinyutong/desktop-cc-gui Download: https://www.mossx.ai/download

After installing it, you might spend three seconds regretting why you didn't install it sooner.