Juggler
A desktop app for working with AI coding assistants where you can see exactly what the AI is doing step by step, rewind and branch off from any earlier point in the conversation, and edit what it already said — instead of scrolling through an endless term
🔗 Visit Juggler
Description
Command-line AI coding agents work, but they're a black box in motion: once the agent has made ten tool calls, it's hard to see what actually happened at step three, and if you want to try a different approach from that point, you're usually starting over. Juggler was built by someone who spent years inside CLI-based AI coding tools and decided the interaction model itself needed a visual rework, not just a better model underneath.
Juggler is a native desktop application (macOS, Windows, Linux, built with a Go backend via the Wails framework and an HTML/JavaScript frontend) that presents AI coding sessions as tree-structured, branching conversations you can edit and backtrack through, with a Miller-column file navigator and inspector views into every tool call the agent makes. It supports Claude Code, OpenAI, Gemini, Ollama, OpenRouter, Z.AI and Deepseek as backends, syncs state across multiple clients (desktop, browser tabs, remote), and persists sessions across restarts. It's open-source under AGPL-3.0 for the core app and Apache-2.0 for its extension SDK, runs localhost-only by default, and offers a commercial license for anyone wanting to build closed-source extensions on top of it.
💬 Our review
The short version: if you find yourself fighting the linear, scroll-and-lose-context nature of CLI-based AI coding agents, Juggler's branching, inspectable, editable session model is a genuinely different and often better way to work with the same underlying models — built by someone with real credibility in developer tooling (JUCE, Tracktion, Cmajor).
It doesn't replace Claude Code or other agents so much as sit on top of them as a richer interface, which means its value depends entirely on how much friction you feel with the current CLI/chat paradigm — if you're happy scrolling a terminal, this adds a layer you may not need. The AGPL license on the app itself is worth flagging for teams: it's free to use and modify, but distributing a modified closed-source version triggers AGPL's copyleft obligations, which is exactly why a separate commercial license exists for that case. Multi-backend support (Claude Code, OpenAI, Gemini, Ollama, and more) is a real strength since it's not locking you into one vendor's agent. Worth trying if session branching and tool-call inspection sound genuinely useful to your workflow; skip it if a terminal already feels sufficient.
💰 Pricing
📊 Global score
🤖 AI-enriched data
AGPL-3.0 pour l'app, Apache-2.0 pour le SDK d'extensions ; licence commerciale sur devis pour usage closed-source
Pros
Conversations en arborescence, branchables et éditables (backtracking réel)
Compatible avec plusieurs backends IA (Claude Code, OpenAI, Gemini, Ollama, etc.)
Vues d'inspection détaillées de chaque appel d'outil
Sessions persistantes, synchronisées entre plusieurs clients
Cons
Ajoute une couche d'interface, pas une nouvelle capacité IA en soi
Licence AGPL à surveiller pour un usage closed-source
Projet jeune, écosystème/documentation encore en construction
Utile seulement si le modèle CLI/chat pose vraiment un problème pour vous