Codex Profile Launcher

Codex Profile Launcher is a small Windows-only utility for using OpenAI Codex Desktop with multiple model backends, including local or self-hosted LLMs exposed through Codex profiles.

Its main value is simple: it makes the Codex Desktop GUI practical to use with a local Ollama setup or another self-hosted OpenAI-compatible endpoint, without manually editing config.toml every time you switch.

It reads profile definitions from Codex config.toml, can append basic profile and model provider sections, saves the selected profile as the active default, and launches or restarts Codex Desktop.

The goal is intentionally narrow: a fast, local, practical Windows UI utility for switching Codex profiles so that Codex Desktop can be used comfortably as a GUI for both OpenAI-hosted and local/self-hosted models.

The problem

A common use case is running Codex Desktop as a GUI while switching between:

• an OpenAI-hosted model
• a local or self-hosted LLM such as Ollama behind an OpenAI-compatible endpoint

Codex Desktop supports this through profile-driven configuration in config.toml, but switching that setup is not currently exposed as a first-class desktop UI flow.

If you use Codex Desktop with both cloud and local model backends, editing config.toml by hand gets old quickly. This launcher removes that friction with a small set of targeted config.toml edits, without becoming a full configuration manager.

What it does

• Reads the active profile from top-level profile = "...".
• Reads available profiles from [profiles.<name>].
• Shows one button per profile.
• Can create a basic [profiles.<name>] section from the UI.
• Can create a basic [model_providers.<name>] section from the UI.
• Highlights the current profile.
• Safely updates the active profile.
• Creates a timestamped backup before modifying config.toml.
• Opens Codex Desktop with the selected profile.
• Offers to restart Codex if it is already running.
• Provides quick actions to open the config file, open the config folder, and copy the config path.

What it does not do

• It does not provide a general-purpose editor for arbitrary existing profile fields.
• It does not provide a general-purpose editor for arbitrary existing model provider fields.
• It does not create advanced nested provider config for you.
• It does not add telemetry, networking, or cloud services.
• It is not cross-platform.
• It is not intended to replace Codex Desktop settings.

Example config

Example: using Codex Desktop GUI with the default OpenAI profile and an Ollama-backed local or self-hosted profile.

profile = "openai"

[profiles.openai]
model = "gpt-5.4"
model_reasoning_effort = "high"

[profiles.ollama]
name = "Remote Ollama"
model = "qwen3.6:35b"
model_provider = "remote_ollama"
model_reasoning_effort = "medium"

[model_providers.remote_ollama]
name = "Remote Ollama"
base_url = "http://your-ollama-host:11434/v1"
wire_api = "responses"

[windows]
sandbox = "elevated"

Each [profiles.<name>] section becomes a launch button.

If a profile contains name, display_name, or displayName, that value is used as a friendly label in the UI.

Config location

The launcher looks for:

• CODEX_HOME\\config.toml if CODEX_HOME is set
• otherwise %USERPROFILE%\\.codex\\config.toml

Safety

When updating config, the launcher:

1. Reads the current config.toml.
2. Updates the top-level profile = "..." setting or appends a basic [profiles.*] / [model_providers.*] section.
3. Writes the new content to a temporary file in the same folder.
4. Replaces the original file using File.Replace.
5. Creates a timestamped backup such as config.toml.20260423120000.bak.

This preserves the rest of your config, including comments and provider sections.

Requirements

• Windows
• OpenAI Codex Desktop installed from the Microsoft Store
• .NET 10 SDK for local builds

The repository includes global.json to prefer the stable .NET 10 SDK line and roll forward to newer stable .NET 10 feature bands when available. Published Windows x64 builds are self-contained, so end users of the packaged EXE do not need to install a separate .NET runtime.

Build

dotnet build .\\CodexProfileLauncher.csproj -c Release

Run from source

dotnet run --project .\\CodexProfileLauncher.csproj

Test

dotnet run --project .\Tests\CodexProfileLauncher.Tests.csproj -c Release

Publish as a single EXE

dotnet publish .\\CodexProfileLauncher.csproj `
  -c Release `
  -r win-x64 `
  --self-contained true `
  -o .\\publish `
  -p:PublishSingleFile=true `
  -p:IncludeNativeLibrariesForSelfExtract=true `
  -p:EnableCompressionInSingleFile=true `
  -p:DebugType=None

Known limitations

• Profile section names are expected to be simple keys such as [profiles.openai] or [profiles.ollama].
• The parser is intentionally scoped to the subset of TOML this launcher needs: top-level profile and [profiles.*] sections.
• Codex Desktop is a single-instance app, so applying a changed profile usually requires a restart.
• Restarting Codex first attempts a graceful close, then falls back to killing Codex Desktop processes if needed.

Support

If this project is useful to you and you want to support it, donations are welcome.

See SUPPORT.md.

Project structure

• global.json - pins the project to the stable .NET 10 SDK line with stable feature-band roll-forward
• MainForm.cs - WinForms UI
• CodexConfigService.cs - profile discovery and safe config writes
• CodexLauncherService.cs - Codex process detection, launch, and restart
• CodexProfile.cs - profile model
• CodexConfigSnapshot.cs - parsed config state
• Tests/Program.cs - lightweight regression test runner for config parsing and writes

License

MIT