Yet another code-agent app, but done properly.
Yes, folks — it's another AI coding agent.
I've been a C++ developer for 30+ years, and the best things I created in that time happened when I got annoyed at some tool I had to use, and decided to write my own version. I ended up with a DAW, a C++ framework, and a DSP language that way.
Luckily (so far) enough people have shared my taste that those projects found a decent audience.
I had a similar moment with AI agents. It's pretty clear that the rest of my coding career will involve staring at some kind of code-agent tool all day, so I decided to have a stab at making one that feels right to me. I may be arriving at the AI agent party after all the booze has been drunk, but maybe there's still some fun to be had..
In this case, the annnoyances I was trying to improve on were the UX, and the data model under the hood.
-
It's a real GUI
Using a code agent involves editing big chunks of multi-line text, and getting hosed with information that you need to scan, navigate, and understand. I find a terminal UX horrible for all those tasks, and Juggler is resolutely a GUI app.
-
The session is a tree, not a doom-scroll
Juggler conversations are documents, not log files. Any point can branch into a sub-thread, recursively. Navigate, drill down, backtrack, duplicate, delete, undo/redo, re-open or branch threads.
-
The session is a persistent state-machine
A session lives in a database, not in memory. Quitting, relaunching the server, or losing/regaining a connection just resumes everything exactly where it was. If an agent was paused mid-run, waiting for you to approve its next step, it'll still be waiting for you after a relaunch, even if you connect from a different device.
-
Everything is inspectable
Tool calls, approvals, thread structure, item properties, even the raw context JSON message — all part of the document.
-
Everything is an extension
Context items, LLM loop strategies, slash commands, and their UIs are JavaScript plugins you can inspect, fork, or replace. Even the basic tools like read/write/bash etc are plugins that you can swap out.
-
Runs locally, remotely, or both at once
A juggler session can have any number of simultaneous client views. The native juggler desktop app and browsers connecting locally or remotely all see the same thing. You can run a headless juggler session on a remote machine and still get the same UI you'd have locally.
-
It supports the usual LLM providers
Claude Code (via CLI or SDK), OpenAI, Gemini, Z.ai, Ollama, OpenRouter, Deepseek, etc. Obviously this list will grow as people request others.
It's open-source
My business plan for this project is "release it and see what happens". The extensions API and juggler core extension are Apache-2.0 licensed. The main app itself is AGPLv3. There's no signup needed - it's just an app, so hopefully should be friction-free for the agent-curious to kick the tyres. If there's enough interest, I might add a 'pro' version with extra features to fund the project.
This is very much an alpha/beta version at the moment! It's taken me about 6 months of heavy churn to get to this point, and the app will need to do some growing-up out in the real world.
Constructive feedback is welcome - come and say hello on the Discord - but be gentle! This is not being developed by a huge team at a trillion-dollar AI company, it's a one-man side-hustle, and I have several other jobs :)
No terminal required
The desktop app
Download it, drag it to Applications, and double-click. It launches the agent and its local server for you — nothing else to set up.
For long-lived & remote sessions
The terminal server
Run the headless juggler binary on the machine where your code lives, then attach from any browser or the desktop app — on your laptop or across the network.
A single Go binary - no node or electron.