Codex DeepSeek Switcher

Why this tool exists

Trying another model should not require repeatedly overwriting the configuration you already use. The practical questions are how to preserve the existing sign-in and settings, how to select your own DeepSeek API credentials, and how to return to the original environment.

Codex DeepSeek Switcher addresses these questions with separate configuration directories. You continue working in the existing Codex extension. The switcher selects the appropriate profile, and newly started processes use that selection after the window is reloaded.

Switch providers. Keep your configuration.

Selecting OpenAI uses the normal .codex directory. Selecting DeepSeek uses the separate .codex-vscode-deepseek directory. The original auth.json is not copied into the DeepSeek home; DeepSeek uses your own API credentials.

The current direct configuration points to api.deepseek.com and uses the Responses API, without MoonBridge or a local relay. The extension manages provider profiles. It does not translate API protocols, supply model accounts or provide API credits.

To return, run Codex Source: Use GPT (OpenAI) and reload. This selects the original configuration directory; its existing contents still determine which provider actually runs.

Local development and Remote SSH

The implementation targets Windows x64 local VS Code and VS Code Remote SSH connected to a Linux host. For remote use, install both the switcher and the OpenAI Codex extension on the SSH extension host, not only on your local computer.

Provider selection is scoped to the operating-system account's home directory. Separate local and remote homes can keep separate choices. Editor windows, SSH aliases or NFS mounts sharing the same home share the provider choice. This is not per-window isolation, and running tasks are not hot-swapped.

Getting started

Install the OpenAI Codex extension, openai.chatgpt, on the appropriate local or remote extension host, and initialize your existing Codex configuration. Then install dzr.codex-local-model-switcher from the Marketplace, or run:

code --install-extension dzr.codex-local-model-switcher

Save your work and run Codex Source: Use DeepSeek Direct. Enter your own DeepSeek API key in the masked prompt when requested, then reload the window. Inspect the configuration with Codex Source: Show Current Codex Source Status.

A configuration check does not establish that a live request will succeed. Test a minimal request in a disposable folder separately, checking connectivity, model access and account balance. Model requests may incur provider charges.

Marketplace, GitHub Releases and development branches can carry different versions. For a newer VSIX, use “Extensions: Install from VSIX...” and select the actual downloaded file. Consult the dated publication record.

Read the step-by-step guides to use DeepSeek, switch back to OpenAI, or set up Remote SSH.

Important boundaries

The current implementation does not support local macOS / Linux, WSL, Dev Containers, Codespaces or browser VS Code. It is not a universal switcher for arbitrary model providers. The launchers locate the binary bundled with the Codex extension and do not preserve custom CODEX_HOME layouts.

The DeepSeek API key is stored in plaintext in the isolated configuration file. Masked input is not encryption. Switching back does not erase credentials or retained backups. The derived configuration may also retain unrelated provider / MCP values from the source profile. Never publish complete configurations, authentication files or unredacted logs.

Setup changes chatgpt.cliExecutable; remote setup also changes a VS Code Server compatibility setting. When removing the switcher, follow the project guide to restore the original settings rather than deleting .codex. See the privacy and security notes and the usage guide.

Open source and future updates

The source is available under the MIT license. Implementation details, usage guides, troubleshooting and version history are maintained in the public repository. Read the source before installing and evaluate whether the tool fits your environment.

Future versions are defined by the actual GitHub and Marketplace releases. This article introduces the project; it does not assert successful live model calls on every client version or treat a website update as a new extension release.