🤹 Juggler: A Visual, Open-Source GUI for AI Coding Agents

Created by the mind behind JUCE, Juggler isn't just another AI coding assistant—it's a specialized environment for developers who prefer a "hands-on" approach to how Large Language Models (LLMs) interact with their source code.

The Core Philosophy: Instead of treating the AI as a black box or a simple chat interface, Juggler provides a visual workbench where you can inspect tool calls, manage branching threads, and manually edit the context provided to the model.

---

🛠️ The Visual Experience: Beyond the Chatbox

Most AI agents force you into a linear transcript. Juggler throws away the linear chat replaces it with a conversation tree.

The Miller Column Interface

Juggler utilizes a Finder-style Miller column view. This allows for deep graphical navigation:

• Root Level: The main conversation starts on the left.
• Expansion: Selecting an item expands its properties and children to the right.
• Capabilities: You can drill down into specifics, backtrack through decisions, compare different paths, and edit raw context on the fly.

Visualizing the Workflow

The structure of a Juggler session can be represented as a directed tree $\mathcal{T}$, where each node $n \in \mathcal{T}$ represents a state of the conversation or a tool execution.

---

🔌 An Extensible Platform

Juggler is designed as a platform where almost every functional element is a JavaScript extension. This means you can inspect, fork, or entirely replace the following:

If a specific AI workflow requires a unique UI or a specialized visualization, Juggler provides the framework to build it rather than relying on a headless Python script.

---

🏗️ Architecture & Connectivity

Juggler employs a multi-client architecture. While it looks like a standard desktop app, it is actually a local webserver hosting a collaborative session.

Supported LLMs

Juggler is compatible with a wide array of providers:

• Claude Code (API or CLI)
• OpenAI (API or Codex plan)
• Gemini
• Deepseek
• Ollama (Local)
• OpenRouter & Z.AI

Client Flexibility

Because the backend is a server, you can connect via:

1. The Native Desktop App.
2. Web Browser tabs (on the same or different machines).

---

🚀 Getting Started

Installation Checklist

• Download the latest build from juggler.studio or the GitHub Releases page.
• Install the bundled server and app.
• Configure your LLM API keys.

OS-Specific Setup

🍎 macOS

1. Open the .dmg and drag Juggler to your Applications folder.
2. Note on Security: If macOS Gatekeeper blocks the app, Right-Click $\rightarrow$ Open $\rightarrow$ Open, or navigate to System Settings $\rightarrow$ Privacy & Security $\rightarrow$ Open Anyway.
3. The server is bundled and starts automatically upon launch.

🐧 Linux

1. Download the juggler server binary.
2. Execute it via the terminal.
3. Connect using the desktop app or a web browser.

🪟 Windows

The installer provides both the GUI app and the juggler.exe command-line server in a single directory, ensuring version parity. It can also add the binary to your system PATH.

---

🖥️ Server Management & Networking

To start a headless session, simply run:

juggler

This will output a connection URL and a QR code.

Network Access Levels

• Localhost (Default): Only the machine running the server can access it.
• LAN Access: Press p in the terminal or use the --public flag.
  • Warning: LAN access is unpassworded; only enable this on trusted networks.
• WAN Access: Not included in the open-source repository. Official binaries from juggler.studio include proprietary WAN modes for internet-wide access (see LICENSING.md).

---

🖼️ Gallery

The primary workbench view.

Managing and inspecting JS extensions.

A single session synced across multiple browser clients.

Optimized for wide-screen productivity.

Accessing your agent via a mobile browser.