QwenLocal

A tiny macOS menu bar switch for a local llama.cpp server.

See if your model is off, loading or running. Turn it on or off with one click.

Why

A large local model eats most of your RAM. You want it running when you use it and gone when you don't, without opening a terminal every time.

QwenLocal sits in the menu bar and does exactly that. Nothing more.

• Manual only. The model never starts on its own. Not when the app opens, not at login. Only when you click Turn On.
• No colored dot. The icon is a plain monochrome symbol. A green or red dot in the menu bar looks like a recording indicator when you share your screen.
• No state of its own. It asks launchd and the server every few seconds, so it always agrees with whatever you do in the terminal.

The icon

Icon	Meaning
cpu outline	Off
cpu outline, pulsing	Loading the model
cpu filled	Running and ready
exclamationmark.triangle	Still not ready after 5 minutes. Check the log.

The menu

• Turn On / Turn Off loads or unloads the launchd job.
• Open Chat opens the llama.cpp web UI at http://127.0.0.1:8080/.
• Copy API Address copies http://127.0.0.1:8080/v1 (OpenAI-compatible).
• Open Log opens the server log in Console.
• Open at Login starts the app at login. Only the app, never the model.
• Quit closes the app. The model keeps running if it was on.

Setup

1. Run llama-server as a launchd job

Save this as ~/.local/llama.cpp/mp.kaio.qwen.plist, with your own paths and model. Keep it out of ~/Library/LaunchAgents, so macOS never starts it by itself.

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
    <key>Label</key><string>mp.kaio.qwen</string>
    <key>ProgramArguments</key>
    <array>
        <string>/Users/YOU/.local/llama.cpp/build/bin/llama-server</string>
        <string>-m</string><string>/Users/YOU/models/your-model.gguf</string>
        <string>-ngl</string><string>99</string>
        <string>--host</string><string>127.0.0.1</string>
        <string>--port</string><string>8080</string>
    </array>
    <key>RunAtLoad</key><true/>
    <key>KeepAlive</key><true/>
    <key>StandardOutPath</key><string>/Users/YOU/Library/Logs/qwen-server.log</string>
    <key>StandardErrorPath</key><string>/Users/YOU/Library/Logs/qwen-server.log</string>
</dict>
</plist>

Want a different label, path or port? Change the constants at the top of Sources/QwenLocal/Server.swift.

2. Install the app

git clone https://github.com/kaiomp/qwen_local.git
cd qwen_local
./scripts/install.sh

This builds a release, wraps it into QwenLocal.app, signs it locally, copies it to ~/Applications and opens it. Needs Xcode or the Swift 6 toolchain.

How it works

Question	How the app answers it
Is the job loaded?	launchctl print gui/<uid>/mp.kaio.qwen
Is the model ready?	GET /health: 200 ready, 503 still loading
Turn on	launchctl bootstrap gui/<uid> <plist>
Turn off	launchctl bootout gui/<uid>/mp.kaio.qwen

It checks every 2 seconds while loading and every 5 seconds otherwise.

The same two launchctl commands work in a shell, so a terminal alias and the app can drive the same job side by side.

Develop

swift run QwenLocal   # run from source
swift test            # state logic tests

License

MIT