mmux vs Nimbalyst

What is Nimbalyst?

Nimbalyst describes itself as "the open-source visual workspace for Claude Code, Codex, and OpenCode" [2]. It is a TypeScript and Electron desktop app for macOS, Windows and Linux, with an iOS companion app [2][7]. You run several agent sessions side by side, each optionally in its own git worktree, and work on their output in visual editors for Markdown, mockups, diagrams, spreadsheets and code [1][5].

Claude Code and Codex are the main agents; OpenCode, GitHub Copilot, Gemini and a few others are listed as alpha [8]. By default Nimbalyst shows Claude sessions in its own transcript UI; running the Claude Code CLI in a terminal is an opt-in mode [8]. You bring your own subscription or API key [1], and sessions are stored in a local SQLite database [9].

Nimbalyst is the successor of Crystal (stravu/crystal), a desktop app for running Claude Code and Codex sessions in parallel git worktrees. Crystal's README marks it deprecated as of February 2026 and replaced by Nimbalyst; its last release is v0.3.5 from 2026-02-26 [12][13].

Where Nimbalyst is better

• Visual editing next to the agent. WYSIWYG Markdown, mockups, Excalidraw, Mermaid, CSV and data models sit in the same window as the session, and the agent works on them with you [1]. In mmux, that is your own editor in another pane.
• Approving edits before they land. Each AI edit is shown as a diff to inspect and approve, and a sidebar lists every file a session read or wrote [1]. mmux shows diffs in its git panel after the fact, with whole-file staging only.
• Task tracking built in. A tracker with kanban and list views that agents read and update as they work, plus the nim CLI for scripting it [1][10]. mmux has no task tracking.
• Windows and iPhone. Native builds for Windows 10+ and an iOS app that relays to your desktop with end-to-end encryption [2][7]. mmux runs on macOS and Linux and has no mobile app.
• No terminal needed. Everything is point and click, and the docs include a section on code and git basics for non-developers [11]. mmux assumes you are comfortable in a terminal.
• Team features. Teams adds shared documents, trackers and team chat; it is free during its beta, then listed at $20 per user per month [1][4]. mmux is single-user.

Where mmux is better

• Sessions outlive the window. Everything runs in an invisible per-directory tmux session, so closing the terminal or losing SSH does not stop agents, terminals or dev servers; mmux or mmux attach rejoins. After a full quit or restart, Claude, Codex, Pi and Grok agents resume their conversation. Nimbalyst's docs say quitting stops terminal processes [6].
• Remote machines. mmux is a terminal program, so it runs on a server or VM over SSH, and its desktop notifications still reach the machine you are sitting at.
• Agents are the real CLIs. An agent is a command in your config (claude, codex, or anything else) in its own terminal pane, so you get the tool's own interface, flags and slash commands, and a new CLI agent works without waiting for an integration.
• Dev processes next to agents. Dev servers and scripts are Processes you start, stop, restart and read logs from. Across worktrees, only one checkout runs a given process at a time, so every branch reuses the same ports.
• Agents can drive it. mmux ls, mmux read, mmux ask and mmux start let an agent check its siblings, read a dev server's logs, hand a prompt to another agent, or cut a worktree. Nimbalyst's nim CLI covers tracker items [10].
• Small and terminal-native. One Rust binary that needs tmux (and git for the git panel), shown in whatever terminal you already use.

When to use which

Moving from Crystal or Nimbalyst to mmux

1. Install mmux. It needs tmux on your PATH (Homebrew installs it for you).
2. Add your agents once, for every project, in ~/.mmux/config.yaml.
3. Optionally add dev servers as processes in the project's mmux.yaml.
4. Run mmux init in the repo to choose a setup, then mmux to open it.
5. For Crystal-style parallel work, press w for a new git worktree. It opens as its own project box with its own agents, terminals and git panel; M merges it back into the branch it came from.

# ~/.mmux/config.yaml
agents:
  - name: Claude
    cmd: claude
  - name: Codex
    cmd: codex

# ./mmux.yaml
name: my-project
processes:
  - name: Dev server
    cmd: npm
    args: ["run", "dev"]

cd ~/some/project
mmux init
mmux

mmux has no importer for Crystal or Nimbalyst sessions, so agents start new conversations. You can also keep Nimbalyst for docs and mockups and run agents in mmux on the same repository: Nimbalyst keeps its content as plain files on disk [2].