Does the remote client need the host's provider keys?

No. Shimex keeps provider keys and Grok, Codex, and Cline sessions on the host. The client only uses the host gateway with its scoped token.

Use Grok 4.5 from another machine in Codex Desktop — without copying secrets.

Q: Does Shimex modify the original Codex app?

No. Shimex creates a managed app copy and isolated profile, leaving the original Codex Desktop installation untouched.

Q: Can I use Kimi K3 in Codex Desktop?

Yes. Shimex exposes Kimi K3 in the Codex Desktop model picker through an authenticated ClinePass profile, with text input, streaming responses, and Codex tool calling.

Q: Can a remote machine use a host's Grok 4.5 profile?

Yes. Pair a remote Tailscale client to a Shimex host with a one-time invite. The client receives a scoped revocable token while the Grok session stays on the host.

Q: Does Shimex support multiple provider profiles?

Yes. Shimex keeps multiple ChatGPT/Codex and ClinePass profiles separate, then exposes profile-scoped models in the Codex picker so requests use the selected profile.

Q: Where do API keys live?

Provider secrets live in a local git-ignored .env file and are referenced from shimex.yml by environment variable name.

Shimex keeps your provider profiles on a host, then bridges a remote client over Tailscale with a one-time invite and revocable token. Your original Codex install stays untouched.

GitHub →

Install Clone, configure .env, and launch the managed Shimex app. Providers Bring hosted APIs, local models, OpenAI-compatible endpoints, and CLI-backed sessions. Host/client bridge Use Grok 4.5 from a remote Tailscale machine while credentials stay on the host. Documentation Understand managed app copies, model catalogs, routing, and secrets. FAQ App safety, supported providers, API keys, and local data.

Shimex admin control plane showing providers, models, and managed-app controls — The local admin control plane at `https://shimex.localhost/admin` (with `http://127.0.0.1:5413/admin` as fallback).

Kimi K3 for Codex Desktop

Kimi K3 is now supported in Shimex through ClinePass. Connect a Cline profile, restart Shimex to refresh the model catalog, and select web: Kimi K3 in the Codex model picker.

Start with ClinePass

Open the local Shimex admin UI, choose Sign in with Cline, then run npm run stop:all && npm start after the profile is connected.

Get a working Codex agent

Choose web: Kimi K3 for text conversations, streamed replies, and Codex tool calls. Shimex normalises strict Moonshot tool schemas before routing.

Use host profiles from a remote Tailscale machine

Grok 4.5 is now available to remote Shimex clients. Start the host once, create a one-time client command, and use the host’s provider profiles from another machine without copying API keys or Grok sessions.

Host: shimex host up → admin page → Create client command → Client: run the generated curl ... | bash command.

shimex host up
shimex host code
# paste the generated command on the Tailscale client

Credentials stay on the host

The host keeps Grok, Codex, and Cline sessions. Clients receive a gateway URL and a scoped, revocable token.

Same model picker, remote machine

Open the managed client app and choose Grok 4.5. Requests travel through the host bridge over your private Tailscale network.

Read the host/client bridge guide →

What it does

Codex Desktop can only use the providers it knows about. Shimex gives it a local OpenAI-compatible gateway and generated model catalog so your own providers appear in the picker.

Run Kimi K3 in Codex Desktop through an authenticated ClinePass profile, with text input, streaming responses, and Codex tool calling.
Use a host’s authenticated Grok 4.5 profile from a remote Tailscale client through a one-time pairing invite and revocable token.
Use hosted APIs, local model servers, OpenAI-compatible endpoints, and external CLI-backed providers from Codex Desktop.
Keep multiple ChatGPT/Codex and ClinePass profiles separate, with profile-scoped models in the picker so each request uses the selected account.
Keep your original /Applications/Codex.app untouched by running a managed Shimex.app copy with an isolated profile.
Run a local OpenAI-compatible gateway at https://shimex.localhost/v1: /v1/chat/completions, /v1/responses, /v1/models, and a Codex picker catalog. http://127.0.0.1:5413/v1 remains available as a direct fallback.
Normalises Anthropic Messages, OpenAI Responses, OpenAI Chat, and Cursor Composer behind the same inbound shape. Streaming SSE everywhere; non-streaming upstreams are wrapped in a single-chunk SSE so the consumer always sees SSE.
Discovers models dynamically with a written cache so normal catalog reads don’t require a live network call; refresh is best-effort on startup.
Holds capabilities to what each provider can actually do. Text-only models reject image input with a typed error instead of dropping parts silently.
Keeps secrets out of source. API keys live in the git-ignored .env; shimex.yml only references environment variable names.

Free

Shimex is free and runs on your machine. Bring your own provider keys, local models, or external CLI sessions.

Free local app

Run Shimex and the managed Codex app locally. There is no hosted subscription or Shimex usage meter.

Bring your own providers

Use your own API keys, Ollama models, LM Studio server, ClinePass session, Cursor session, or OpenAI-compatible endpoint.

Install

macOS with Codex Desktop installed at /Applications/Codex.app, Node 20+, npm 10+.

git clone <repo-url> ~/Projects/shimex
cd ~/Projects/shimex
npm install
cp .env.example .env   # add your API keys; .env is git-ignored
shimex host up         # keep the host gateway online
npm start              # prepare and launch the managed Shimex.app

Then open the managed Shimex.app and the admin UI at https://shimex.localhost/admin. The direct fallback is http://127.0.0.1:5413/admin.

CLI

Run Shimex from npm. These commands manage the daemon, inspect health, and list configured provider models.

npm run status                                 # PID, log path, health
npm run shimex -- doctor                       # prerequisite check
npm run shimex -- providers list               # configured providers
npm run shimex -- models list                  # discovered models
npm run shimex -- host clients                 # paired remote clients
npm run shimex -- app stop                     # quit the app without stopping the host

Documentation

Shimex is configured by shimex.yml and a local git-ignored .env. The managed app copy, isolated Codex profile, model catalog, and local gateway are all created from those files.

Want to add a provider? Point your coding agent at the provider rules or the Adding a Provider docs. The required path is explicit: add the provider module, manifest, adapter, registry entries, shimex.yml example, and tests.

Managed Codex app

Shimex copies Codex Desktop into a managed app and points that copy at the local gateway.

Provider adapters

Adapters translate provider-specific request, response, model, and streaming behavior into Codex-ready shapes.

Agent-friendly provider docs

If Shimex does not support a provider yet, give your agent the provider docs and the Shimex provider rules. The architecture tells it exactly where each change belongs.

FAQ

Does Shimex modify the original Codex app?

No. Shimex creates a managed app copy and isolated profile so the original Codex Desktop installation stays untouched.

Which providers can Shimex use?

Shimex can route to ClinePass, DeepSeek, LM Studio, Cloudflare Workers AI, Anthropic, Ollama, Cursor Composer, OpenAI-compatible endpoints, and other configured providers.

Can I use Kimi K3 in Codex Desktop?

Yes. Connect a ClinePass profile in the Shimex admin UI, restart Shimex with npm run stop:all && npm start, and select web: Kimi K3 in the Codex model picker. Shimex supports text input, streaming responses, and Codex tool calls for Kimi K3.

Can a remote machine use a host’s Grok 4.5 profile?

Yes. Run shimex host up, create a one-time client command in the admin page, and run it on the Tailscale client. The client can then select Grok 4.5 without receiving the host’s provider credentials.

Does the host need Shimex.app open for bridging?

No. shimex host up runs the persistent gateway separately. Use shimex app start and shimex app stop to control only the managed desktop app.

Does Shimex support multiple provider profiles?

Yes. Keep multiple ChatGPT/Codex and ClinePass profiles separate. Shimex exposes profile-scoped models in the picker, so each request uses the profile you selected.

Where do API keys live?

Provider secrets live in a local .env file that is ignored by git. shimex.yml references those secrets by environment variable name.

How do I update when Codex Desktop releases a new version?

Just run npm run stop:all && npm start. Shimex re-copies the upstream Codex app into the managed Shimex.app, reapplies patches, and restarts the gateway with the latest version.