Codex Model Switcher is a small macOS utility that lets you switch the Codex desktop app between local Ollama models and the original OpenAI Codex mode without typing terminal commands every time.
It is built for developers who use Codex daily and want a fast local-model toggle for Ollama-powered coding sessions.
When you want Codex to use a local Ollama model, the command is easy to forget:
ollama launch codex-app --model gpt-oss:20b --yesAnd when you want to go back to normal OpenAI Codex mode:
ollama launch codex-app --restoreThis app wraps that workflow in a clean native UI.
- Detects installed Ollama models with
ollama list - Shows models in a dropdown
- Remembers your last selected model
- Launches Codex with the selected Ollama model
- Restores original OpenAI Codex mode
- Quits Codex safely
- Shows simple logs and errors
- Includes model helper labels
- Clean dark-mode friendly macOS interface
- Includes a lightweight Windows PowerShell GUI version
On startup, the app auto-selects the best available model in this order:
- Your last selected model, if still installed
gpt-oss:20bdevstral-small-2:latestdevstral-small-2gemma4:26b- First available model
| Model | Label |
|---|---|
gpt-oss:20b |
Recommended for Codex / balanced |
devstral-small-2:latest |
Coding fallback |
gemma4:26b |
Hindi / general chat |
gemma4:latest |
Light general use |
Open:
Mac/Codex Model Switcher.app
Or share:
Mac/Codex Model Switcher.zip
The macOS app is ad-hoc signed. For a completely warning-free public distribution on other Macs, Apple Developer ID notarization is required.
Open:
Windows/Launch Codex Model Switcher.bat
The Windows version is a lightweight PowerShell GUI with the same main actions.
CodexModelSwitcher/
├── Mac/
│ ├── Codex Model Switcher.app
│ ├── Codex Model Switcher.zip
│ └── README-Mac.txt
├── Windows/
│ ├── Codex Model Switcher.ps1
│ ├── Launch Codex Model Switcher.bat
│ ├── README-Windows.txt
│ └── Windows Codex Model Switcher.zip
└── Project/
├── Sources/
├── Assets/
├── Docs/
├── Tools/
└── Package.swift
The app only uses local commands for this workflow:
ollama list
killall Codex || true
ollama launch codex-app --model MODEL_NAME --yes
ollama launch codex-app --restore
open -a CodexNo sudo. No model deletion. No manual modification of Codex files.
cd Project
swift build -c releaseMain source files:
Project/Sources/CodexModelSwitcher/CodexModelSwitcherApp.swiftProject/Sources/CodexModelSwitcher/ContentView.swiftProject/Sources/CodexModelSwitcher/ShellRunner.swiftProject/Sources/CodexModelSwitcher/OllamaModelService.swiftProject/Sources/CodexModelSwitcher/CodexLauncher.swift
- macOS 13 or newer for the native app
- Ollama installed
- Codex desktop app installed
- At least one Ollama model pulled
Recommended:
ollama pull gpt-oss:20bThis project is intentionally small and boring in the best way: it is a button-based wrapper around a few explicit local commands. It does not delete models, edit unrelated files, or require admin privileges.
If this saves you from retyping the same Codex/Ollama commands every day, a star helps other developers find it.
