Why this matters
Many proxy setups make Codex behave like a generic OpenAI-compatible client. That can be enough for plain chat, but it may break the parts of Codex Desktop that expect the official OpenAI provider shape and signed-in account state.
MuxLayer uses a different pattern: keep the official-looking client entry, move the actual model entry to localhost.
Codex Desktop
-> OpenAI provider entry in Codex config
-> local MuxLayer base URL
-> DeepSeek / Xiaomi MiMo / OpenAI / Kimi / GLM / DashScope / other provider
Codex Desktop still sees an OpenAI-style provider entry, while MuxLayer receives the model request locally and routes it to the selected upstream provider.
What keeps working
| Area | What MuxLayer preserves |
|---|---|
| Codex Desktop sign-in | Keeps the official auth.json account tokens instead of replacing them with a local API key. |
| Plugin and account features | Keeps Codex Desktop on the OpenAI-authenticated provider path that plugin and account features expect. |
| Local model entry | Sends model requests through MuxLayer first, then converts or passes through to DeepSeek, MiMo, OpenAI, Kimi, GLM, DashScope, or another configured provider. |
| One-click restore | Saves the original Codex config so you can switch back to official behavior from the MuxLayer UI. |
Plugin availability still depends on Codex Desktop, your signed-in account, and the upstream feature itself. MuxLayer's role is to avoid breaking that official path while making the model request locally controllable and traceable.
Quick Setup
- Download MuxLayer from Releases and open the app.
- Add at least one provider, such as DeepSeek or Xiaomi MiMo.
- Start the gateway from Overview or Gateway. The default endpoint is
http://127.0.0.1:9090. - Open Clients and click Apply Config on the Codex card.
- Keep your Codex Desktop account signed in.
- Send a message in Codex Desktop and check MuxLayer's request logs to confirm the selected provider was used.
Chinese Notes / 中文说明
如果你搜索的是“Codex 桌面端 第三方 API”“Codex Desktop DeepSeek 插件”“Codex 桌面端 MiMo 插件”，这个能力是 MuxLayer 和普通代理最不一样的地方之一。
普通代理通常只是把 Codex 请求改到一个 OpenAI-compatible 地址。这样能让模型请求跑起来，但 Codex 桌面端里依赖官方 OpenAI provider 形态、登录态、账号能力的部分可能会失效。
MuxLayer 的做法是：
- Codex 配置里仍然使用 OpenAI provider 入口。
base_url指向本地 MuxLayer 网关。auth.json里的 ChatGPT / OpenAI 登录态保留，不被本地 token 覆盖。- 本地访问 token 放在 Codex 配置里，由 MuxLayer 网关校验。
- 真正的模型请求由 MuxLayer 路由到 DeepSeek、小米 MiMo 或其他 Provider。
结果是：Codex 桌面端可以继续保持它熟悉的官方 provider 语义，同时你可以在 MuxLayer 里切换第三方 API。
How it differs from a simple proxy
| Simple proxy | MuxLayer |
|---|---|
| Usually treats Codex as a generic OpenAI-compatible client. | Preserves Codex Desktop's OpenAI provider path while changing the local base URL. |
| May require replacing auth with a proxy API key. | Keeps official account tokens and stores the local gateway token in config. |
| Often focuses on one provider. | Supports route profiles, failover, model mapping, and multiple providers. |
| Harder to switch back cleanly. | Provides one-click switch back to official Codex config. |