Codex models setup

How it works

Codex Proxy is a small app that runs on your machine and exposes your ChatGPT subscription as a standard OpenAI-compatible API on http://localhost:8080.

Kismet’s model picker has a dedicated Codex group. Selecting any model from that group sends the request to the local proxy — and therefore to your ChatGPT subscription — so there is no metered OpenAI key and no per-token cost. The other GPT entries (GPT-5.5, GPT-5.4, …) continue to use the real OpenAI API as before.

Routing is decided per model, not globally: a Codex model always goes to the proxy and a plain GPT model always goes to the metered API, so you can switch between them in the same session. Kismet points the Codex route at http://localhost:8080 with the default proxy key automatically — there are no environment variables to set.

Requirements

Before starting, make sure you have the following.

Step 1 — Install Codex Proxy

Codex Proxy is a desktop app. Install it on the same machine that runs Unreal Editor.

1. 1.

   Download the installer

   Get the latest release from the Codex Proxy releases page. On Windows download Codex Proxy Setup x.x.x.exe; on macOS download Codex Proxy-x.x.x.dmg.

2. 2.

   Install and open the app

   Run the installer (Windows) or drag the app to Applications (macOS), then launch Codex Proxy. It starts a local server on port 8080.

3. 3.

   Sign in with your ChatGPT account

   In the app, start the login flow. A browser window opens for the ChatGPT sign-in; complete it and return to the app.

4. 4.

   Confirm the dashboard

   Open http://localhost:8080 in a browser. You should see your signed-in account. On a fresh install the proxy API key is pwd — which is the value Kismet uses by default, so you usually don’t need to touch it.

Step 2 — Verify the proxy

Confirm the proxy answers before touching Unreal. You should see streaming text in the response.

macOS (Terminal):

curl http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer pwd" \
  -d '{"model":"gpt-5.5","messages":[{"role":"user","content":"Hello!"}],"stream":true}'

Windows (PowerShell):

curl.exe http://localhost:8080/v1/chat/completions `
  -H "Content-Type: application/json" `
  -H "Authorization: Bearer pwd" `
  -d '{\"model\":\"gpt-5.5\",\"messages\":[{\"role\":\"user\",\"content\":\"Hello!\"}],\"stream\":true}'

Step 3 — Pick a Codex model in Kismet

There are no environment variables and no editor restart dance — just pick a model.

1. 1.

   Make sure Codex Proxy is running

   The dashboard at http://localhost:8080 should load and show your signed-in account.

2. 2.

   Open the Kismet panel and the model picker

   You’ll see a Codex group with a teal badge, listing the available Codex models.

3. 3.

   Select a Codex model

   Pick GPT-5.5 (Codex) for the strongest coder. Reasoning effort is set to High automatically on selection.

4. 4.

   Run a small test edit

   Ask for something small and verifiable — for example, add a Print String after Begin Play — and let Kismet compile the Blueprint. If the edit completes, the chain (Kismet → proxy → your subscription) is working.

The Codex model lineup

The Codex group mirrors the models your ChatGPT plan exposes through Codex. All of them run on your subscription via the proxy.

Codex vs. the metered API

Kismet lists two ways to reach OpenAI models, and the picker keeps them visually distinct:

The two are independent — picking a Codex model never touches your OpenAI key, and picking a metered model never touches the proxy. You can move between them freely.

Advanced configuration

The defaults (http://localhost:8080 and key pwd) match a fresh Codex Proxy install, so most setups need nothing here. Override them only if your proxy differs.

These are read from the Unreal Editor’s environment at startup. On Windows, set them with setx (then restart the launcher); on macOS, launch the editor from a Terminal that exported them, or use launchctl setenv.