Flip between local LLM runtimes from your menu bar. One click to activate. One click to stop everything.
Running local models on an Apple Silicon Mac usually means a sprawl of terminal windows, half-remembered launch scripts, and no clean way to see what's actually running. Model Switchboard puts llama.cpp, MLX, Ollama, vLLM, and custom launch scripts behind one menu bar panel. Click Activate, every other model stops, and the one you picked comes up at an OpenAI-compatible endpoint.
No terminals. No orphan processes. No "green dot" lies.
Activate stops every other profile and brings the chosen one up. No more forgetting to kill -9 a 24 GB process before starting the next one.
Profiles are marked ready only after a real health check passes. By default, the app checks /v1/models, and you can swap in a custom HTTP probe. If it says green, it means green.
Built with SwiftUI and MenuBarExtra. No Electron, no bundled inference engine, no resident background worker pegging your CPU.
CPU and GPU utilization live in the header, so you know what your machine is doing without dropping into Activity Monitor.
Benchmark All runs the fleet. Reopen Last jumps straight back to what you had up. Sync Droid pushes your managed profiles into Factory Droid's custom-model settings, so the model you just activated is the one Droid uses. It is the first of several planned sync adapters (see contributing).
The inline panel reads the latest run and shows TTFT, Decode, E2E, and RSS per profile in one place.
Tap Export CSV and you have a portable report. Every run lands as both JSON and Markdown under Controller/benchmark-results/, so it is easy to diff, commit, or feed into another tool.
Same codebase, two apps. Pick at install time. They live side by side as Model Switchboard.app and Model Switchboard Plus.app under ~/Applications/.
| Base | Plus | |
|---|---|---|
| Profile list with live status | ✓ | ✓ |
Activate / Start / Stop / Restart |
✓ | ✓ |
Refresh / Stop All |
✓ | ✓ |
Launch At Login + attached Settings / Help |
✓ | ✓ |
| CPU / GPU utilization badges | - | ✓ |
Benchmark All + per-profile Benchmark |
- | ✓ |
| In-app Benchmarks panel + CSV export | - | ✓ |
Reopen Last |
- | ✓ |
Sync Droid and future integration adapters |
- | ✓ |
- macOS 14 (Sonoma) or later
- Apple Silicon recommended. Intel Macs run the app fine, but MLX models require Apple Silicon
- A running controller that exposes the controller contract. This repo ships a reference controller under
Controller/
Signed DMG (recommended). Grab the latest from Releases:
Model-Switchboard-<version>.dmg(Base)Model-Switchboard-Plus-<version>.dmg(Plus)
Open, drag to Applications, launch.
Maintainers can cut the next release with:
python3 Scripts/bump-version.py patchPush that version-bump commit to main and GitHub Actions will build, notarize, and publish the release automatically.
From source.
git clone https://github.com/AdityaVG13/Model-Switchboard.git
cd Model-Switchboard
./Scripts/install.sh # Base
APP_VARIANT=plus ./Scripts/install.sh # PlusThe installer places a fresh build under ~/Applications/, registers it with Launch Services, and forces a Spotlight import so Raycast and Alfred pick it up immediately.
Model Switchboard is the control surface. It does not run models itself. You need a controller that knows how to launch and health-check models.
1. Install the reference controller:
./Controller/install-model-switchboard-controller.sh2. Drop a profile manifest into ~/.model-switchboard/model-profiles/ (the exact path is shown in Settings). A minimal llama.cpp example:
DISPLAY_NAME=Qwen 3.5 35B Local
RUNTIME=llama.cpp
MODEL_PATH=/path/to/model.gguf
PORT=8080
REQUEST_MODEL=qwen35-local
SERVER_MODEL_ID=qwen35-local3. Open the menu bar icon. Your profile appears. Click Activate.
Every profile must resolve to a unique endpoint. Reusing the same HOST:PORT or BASE_URL across two profiles is a configuration error, and the controller doctor will flag it.
Using your own runtime or launcher? Any backend that honors the controller contract works. MLX, Ollama, vLLM, and custom-command examples live in SETUP.md.
If something looks off, these labels tell you what the app is seeing right now:
| Label | Where it appears | Meaning |
|---|---|---|
RUNNING |
Profile card badge | Process is currently running. |
NOT RUNNING |
Profile card badge | Process is not running. |
STARTING / STOPPING / RESTARTING / ACTIVATING |
Profile card badge | Action is in progress for that profile. |
STALE |
Footer chip near the clock | Last successful status refresh is older than ~45 seconds. |
CACHED |
Footer chip near the clock | Controller was temporarily unavailable and the app is showing last cached status. |
ERROR |
Footer chip near the clock | Latest refresh or action returned an error. |
RUNNING / READY |
Plus Benchmarks panel | Benchmark job is active / no benchmark currently running. |
All the deeper material lives in one place so this README stays skimmable:
SETUP.md: profile formats, supported runtimes, health checks, controller API contract, build-from-source flow, release pipeline, Raycast power-user notes, troubleshooting, and known limitations.
CHANGELOG.md: release-by-release changes and distribution hardening notes.
The app's Help button opens the same doc.
PRs, issues, and profile recipes are welcome. A few ground rules that keep the project reusable:
- Keep the app generic. Runtime-specific behavior belongs in the controller or a profile manifest.
- The controller HTTP contract is the stability boundary. Make additive changes only.
- External tools stay optional integrations, never required features.
- Ship a runnable example with any new adapter.
Sync Droid is currently Factory-Droid-specific because that's the agent I run. The integration slot is generic, but the adapter is not. PRs that add sync adapters for other local-model terminals or agentic tools are very welcome, including but not limited to:
- Cursor / Windsurf: push the active profile into the OpenAI-compatible provider settings
- OpenAI Codex CLI: update
~/.codex/config.tomlmodel entry - Zed: update
~/.config/zed/settings.jsonassistant provider - Continue (
~/.continue/config.json) - Aider: point at the active endpoint
- LM Studio / Ollama chat frontends / any OpenAI-compatible consumer
If you build one, follow the shape of Controller/sync-droid-local-models.py and register it under Controller/integrations/ so it shows up in the Plus menu automatically.
Before opening a PR:
swift test && ./Scripts/check-cycles.py && ./Scripts/build-app.shFor maintainers, release prep is scriptable instead of hand-editing version files:
python3 Scripts/bump-version.py patch # or minor / major / x.y.z
./Scripts/release-preflight.shIf you already downloaded the app and just want the controller and profiles wired up, give your favorite AI this prompt:
I already downloaded Model Switchboard on my Mac.
Set up the reference controller for me, create working model profiles for the runtimes I actually have installed, and make the configuration portable instead of hardcoding your own assumptions.
Rules:
- This is macOS-only, and that is intentional.
- Do not hardcode Homebrew paths, repo-local build paths, or personal directories unless you first verify they exist on this machine.
- Prefer profile-driven config:
- `MODEL_PATH` or `MODEL_FILE` with `MODEL_ROOT` for llama.cpp
- `MODEL_DIR` or `MODEL_REPO` for MLX
- `SERVER_BIN` when a runtime binary is not already on PATH
- `START_COMMAND` only for truly custom launchers
- Use the controller contract and profile format documented in this repo's `SETUP.md`.
- Put profiles in the controller's `model-profiles` directory.
- Verify that each profile can be started, health-checked, and stopped cleanly.
- If something is missing, inspect the machine and ask me only the minimum necessary question.
End state:
- Model Switchboard opens with valid profiles visible
- `Activate` works
- health checks go green only when the endpoint is actually ready
- nothing is tied to one specific Mac beyond what is truly installed here
MIT © 2026 AdityaVG13
