togetherlink
Codex in ChatGPT Desktop · alpha · 8 min

Use Open Models in Codex for the ChatGPT Desktop App

Add open models beside Sol, Terra, Luna, and your other native GPT choices in the Codex picker. A local router sends each model to the right provider, while an automatic backup and one restore command keep the change reversible. Ordinary ChatGPT conversations are not changed.

By TogetherLink

TLtogetherlinkCODEX DESKTOP PROVIDERUse open models in Codexfor ChatGPT Desktop$togetherlink chatgptOpen models · Responses APIChatGPT DesktopPOWERED BY
Alpha integration

Native GPT and Together models stay in one picker

TogetherLink preserves the built-in OpenAI provider, merges native and Together model metadata, and routes requests by the selected model. The router configuration persists because ChatGPT Desktop runs separately from the CLI. TogetherLink backs up the previous file first.

Before you start

Keep ChatGPT signed in

Install ChatGPT Desktop with the Codex coding workspace and complete the normal OpenAI sign-in. That session continues to serve and bill native GPT choices. Together AI serves and bills only the Together models you select.

This guide currently targets macOS. The code path and generated configuration were checked against ChatGPT 26.715.61943 on July 21, 2026. App behavior may change while this integration remains alpha.

Setup

Configure the desktop coding workspace

1

Install TogetherLink

$curl -fsSL https://togetherlink.vercel.app/install.sh | bash
2

Save your Together API key

$togetherlink configure

Press Enter to skip the optional Exa key if you do not need web search.

3

Configure and open ChatGPT Desktop

$tchatgpt

If the app is already open, TogetherLink asks before restarting it. If you decline, quit and reopen the app manually so it reloads the provider and model catalog.

What gets changed

Codex configuration

Adds a local OpenAI-base-URL router and generated catalog path in ~/.codex/config.toml. Your native provider and default model are preserved unless you pass --model.

Routing

Native GPT slugs keep their ChatGPT authentication; Together slugs use the Together API key. Credentials are never forwarded across providers.

Model catalog

Writes ~/.codex/togetherlink-codex-app-models.json with native GPT metadata plus the available Together models.

Backup

Stores the previous configuration under ~/.togetherlink/backup/codex-app/.

Active provider

Keeps the built-in OpenAI provider active so native GPT models remain selectable.

Verify before opening a task

Check the additive router without exposing your key

This command prints the managed routing settings. The provider should remain openai (or be absent, which means the built-in default), your existing model should remain selected unless you passed --model, and the catalog path should point to the generated JSON file.

$grep -E '^(model|model_provider|model_catalog_json|openai_base_url) =' ~/.codex/config.toml

In ChatGPT Desktop, open the Codex coding workspace and check that the model picker shows the same model before starting a repository task.

Undo TogetherLink

Restore your previous app configuration

TogetherLink restores the backup, removes its generated model catalog and session registration, and clears the app's model cache. If the app is open, approve the restart or quit and reopen it manually so it loads the restored configuration.

$tchatgpt off
Optional model choice

Make another model the default

The generated catalog lets you switch between GPT and Together models in the app. To make a Together model the default, rerun setup with --model after chatgpt.

GLM 5.2
$tchatgpt --model zai-org/GLM-5.2
Qwen 3.7 Max
$tchatgpt --model Qwen/Qwen3.7-Max

If the models do not appear

Fully quit and reopen ChatGPT Desktop. To undo setup, run the restore command instead of manually deleting configuration blocks.

Common questions

FAQ

Can the Codex coding experience in ChatGPT Desktop use Together models?+

Yes. TogetherLink's alpha integration merges Together models into the normal Codex picker. Native GPT models keep using your ChatGPT sign-in, while Together model requests go through a local Responses-compatible router to Together AI. It does not change ordinary ChatGPT conversations.

Do I need a Together API key?+

Yes. Together AI serves and bills the selected model. A ChatGPT subscription or OpenAI API key does not cover these requests.

Is this change temporary like the CLI wrappers?+

No. ChatGPT Desktop launches separately from TogetherLink, so the additive catalog and local router stay in its Codex configuration until you run tchatgpt off. Native GPT models remain available while it is enabled.

Can I restore my previous app configuration?+

Yes. TogetherLink backs up the affected configuration before changing it. Run tchatgpt off, then restart the app if it was open.

Is the ChatGPT Desktop integration stable?+

It is currently labeled alpha because ChatGPT Desktop behavior can change between app releases. TogetherLink includes a backup and restore path, so keep the restore command nearby while testing it.

Related guides