Show HN: Juggler – an open-source GUI coding agent, by the creator of JUCE
Summary
Juggler is an open-source GUI coding agent that provides a visual workbench for interacting with LLMs to edit code, featuring inspectable tool calls, branching threads, and editable context. It runs locally or remotely and supports multiple models.
View Cached Full Text
Cached at: 07/14/26, 04:17 PM
juggler-ai/juggler
Source: https://github.com/juggler-ai/juggler
Juggler
Yes, it’s another AI coding agent. The industry definitely needed one more.
If Juggler has an angle, it’s that it’s for people who want to be more hands-on over what the LLM is doing to their codebase. It gives you a visual workbench: inspectable tool calls, branching threads, editable context.
More blurb on the website: https://juggler.studio
And here’s the TL;DR:
- It is a proper GUI. This is not a console app. It’s all about graphical visual navigation, inspection, and control.
- The session is a tree, not a doom-scroll. It’s a Yjs document, not a transcript. Create sub-threads, drill down, backtrack, compare, and edit.
- Everything important is visible. Tool calls, approvals, thread structure, item properties, raw context — laid out in Finder-style Miller columns instead of buried in collapsible chat.
- It’s plugins all the way down. Context items, slash commands, LLM loop strategies, and their UIs are JavaScript extensions you can inspect, fork, or replace.
- It runs locally, remotely, or both at once. Use the same session with the same UI in the native desktop app, and/or browsers. Multiple clients can attach to the same session.
- It talks to the usual model zoo. Claude Code (via CLI or API), OpenAI (codex plan or API), Gemini, Ollama, OpenRouter, Z.AI, Deepseek, etc.
Getting started
Download a build from the Releases page or via juggler.studio.
Each download contains the same two moving parts:
- Juggler app — the native desktop app. Works like you’d expect it to.
juggler— the headless command-line server. Run this from a terminal for long-lived, remote, or network-accessible sessions. It has no window of its own, but you can typewinto its terminal to open the desktop app, or use the browser URL it prints.
The desktop app, browser tabs (on local or remote machines) can all be clients viewing the same server session.
Installing
- macOS — download the
.dmg, open it, and drag Juggler to Applications, then launch it. The app and its server are bundled together, so the server starts automatically. The first time you open it, macOS Gatekeeper may block the download: right-click (or Control-click) the app → Open → Open, or go to System Settings → Privacy & Security → Open Anyway. After the first launch it opens normally. - Windows — download
Juggler-<version>-setup.exeand run it. It installs the desktop app and the matchingjuggler.execommand-line server together in one directory (and can addjugglerto your PATH), so the two never drift apart. - Linux — download the
jugglerserver binary and run it from a terminal, then connect with a browser or the desktop app.
The desktop app and the server always ship and install as one unit — see docs/distribution.md.
Running the server directly
For a headless session, just run:
juggler # prints everything you need to connect from a browser
By default, the server opens a web UI and prints its URL plus a QR code for easy connection.
The server is localhost-only by default — nothing off your machine can reach it. To let other devices on your network connect, press p in the terminal (or launch with --public). LAN access has no password: anyone who can reach the address can drive the agent, so only enable it on networks you trust.
Access from beyond your LAN isn’t built into this repository: a build from this source is local + LAN only. The official Juggler binaries from juggler.studio additionally include WAN access modes for reaching your server across the internet (see LICENSING.md on components not in this repo).
What makes it different?
Your conversation is an editable tree, not a chat history
Most agents give you a single linear transcript and if you’re lucky you can rewind it.
Juggler gives you a tree. Any point can branch into a sub-thread. Sub-threads can branch again. You can navigate, inspect, and edit the structure directly.
The UI uses Miller columns: root on the left, selected items expanding into properties and children to the right. (If you’ve used Finder’s column view, you already understand the basic move).
Everything is an extension
The core app manages the document and orchestration. Almost all the objects that make up the document are defined by JavaScript extensions:
- Context items — every item type in a conversation (
read-file,replace-text,bash, …) controls both how it talks to the LLM and how it appears in the UI. - Strategies — high-level LLM loops such as
plan,research, or your own fever-dream inventions are plugins too. - Commands — slash commands like
/clearand/compactare all just plugins that manipulate the session document.
Not every LLM workflow wants to live as a headless Python script skulking in a terminal. If an orchestration idea needs its own UI, controls, or visualisation, Juggler is a platform for that.
A desktop app with a multi-client architecture
Juggler looks like a native desktop app, but underneath it is a local webserver serving a live collaborative session. The app is just one client. A browser tab can be another. A different machine can be another.
That means you can run the server where the code lives — local workstation, dev box, server farm - and attach views from wherever is convenient.
Model support
Juggler connects to the usual suspects: Claude Code (via CLI or API), OpenAI (codex plan or API), Gemini, Ollama, OpenRouter, Z.AI, Deepseek, etc. It’s easy to add more providers, so if yours is missing, ask your friendly neighbourhood LLM to add it as a PR.
Building from source
The idea with the extensions system is that most people won’t need to actually build the app. If you do, see CONTRIBUTING.md for the full setup. The short version:
git clone --recurse-submodules https://github.com/juggler-ai/juggler.git
cd juggler && make build
If you already cloned without --recurse-submodules, fetch them with:
git submodule update --init --recursive
Windows binaries cross-compile from any host with make build-windows; the Linux desktop app must be built natively.
Tech stack
Juggler is a simple native app without baggage. The backend is Go, using Wails for windowing. The UI is HTML/JS served by the Go backend. Session documents are stored and synchronised with Yjs. Extensions are JavaScript. It doesn’t use electron.
The frontend is type-checked JavaScript rather than TypeScript: types live in JSDoc and are enforced in CI with strict static linting. There’s no build step between source and what ships.
Contributing
See CONTRIBUTING.md for setup, test commands, and project conventions. For security issues, please use the private channel described in SECURITY.md rather than the public issue tracker.
License
Juggler’s application code is licensed under the GNU Affero General Public License v3.0 or later. The extension SDK (web/sdk/) and the bundled extensions (web/extensions/) are licensed under Apache-2.0, so you can build extensions — including closed-source ones — with no copyleft obligation. See LICENSING.md for the full map.
For the AGPL parts you’re free to use, modify, and redistribute — but any modified version you distribute or host as a service must also be released under the AGPLv3. If you want to do something closed-source with it, contact me to discuss commercial licensing.
Similar Articles
@CryptoTied: Guys! An AI coding agent emphasizing visual control has arrived. Juggler is an AI coding agent focused on "controllable, visual" interaction, with the core goal of giving developers true control over the LLM's interaction with the codebase, rather than black-box operation. It offers a native desktop app + Mi...
Juggler is an AI coding agent that emphasizes visualization and controllability. It provides a native desktop app, session tree structure, plugin architecture, and supports multiple models like Claude, OpenAI, Gemini, etc., allowing developers to clearly view and edit every step of the LLM's actions.
1jehuang/jcode
jcode is an open-source coding agent harness designed for multi-session workflows with low resource usage, offering CLI installation and performance improvements over existing agents like Claude Code and Cursor Agent.
Koder: browser UI based harness for coding and computer use
Releases Koder, a local/offline AI coding and computer use harness with browser UI, supporting llama.cpp, MCP, visual models, and multi-task planning.
Show HN: Y – A malleable coding-agent desktop app built with Electron
Y is a malleable, chat-first desktop app that runs local coding agents like Claude Code and Codex in parallel, with a self-modifying UI capability.
Show HN: Jacquard, a programming language for AI-written, human-reviewed code
Jacquard is a programming language designed for AI-written, human-reviewed code, with built-in effect tracking, probabilistic simulation, and canonical identity to help humans trust AI-generated programs.