ChatGPT Provider

ChatGPT is integrated as a first-class MangoStudio provider for personal ChatGPT subscription accounts. It uses browser-based OAuth sign-in instead of an API key, then talks to the ChatGPT backend with the same account-oriented flow used by tools such as Codex CLI and OpenCode.

Provider Type

Provider ID: chatgpt
Authentication: Sign in with ChatGPT from Settings
Wire format: ChatGPT backend Responses protocol
Continuation: Stateless replay within MangoStudio; no durable server-side continuation cursor is stored across turns
Tools: MangoStudio tool registry, executed by the API

Open Settings -> Connectors.
Choose Sign in with ChatGPT.
MangoStudio starts a temporary loopback callback server on 127.0.0.1:1455.
Complete ChatGPT sign-in in the browser opened by the API host.
After ChatGPT redirects to http://localhost:1455/auth/callback, the API exchanges the authorization code, stores the token bundle, and lists the connector like any other provider.

Only one ChatGPT sign-in can use port 1455 at a time. Close other login flows, including codex login, if MangoStudio reports that the port is busy.

Token Storage And Reauth

ChatGPT access tokens expire and refresh tokens rotate. MangoStudio stores the full token bundle in the OS secret store through Bun.secrets; it does not support config.toml or .env storage for ChatGPT connectors. The SQLite database stores only UI-safe metadata such as connector name, account label, plan label, enabled models, and reauth status.

When a refresh token is rejected or revoked, MangoStudio marks the connector as requiring reauthentication. Re-run Sign in with ChatGPT for the existing connector to replace the token bundle without losing enabled-model settings.

Model Availability

The model catalog combines a static ChatGPT baseline with best-effort discovery from the account. Availability still depends on the ChatGPT plan attached to the signed-in account, and plan rate limits are enforced by ChatGPT.

If a model appears but the account cannot use it, the backend may reject the generation request. Reauthenticate after plan changes so MangoStudio can refresh account and plan metadata.

ChatGPT Vs OpenAI API Key

Aspect	ChatGPT connector	OpenAI connector
Credential	Browser sign-in, rotating tokens	API key
Account scope	ChatGPT subscription account	OpenAI API organization/project
Storage	OS secret store only	OS secret store, `config.toml`, `.env`
Rate limits	ChatGPT plan limits	API account/project limits
Continuation	Stateless replay, no durable cursor	Responses continuation where supported
Intended usage	Personal subscription access	Programmatic API access

Use the OpenAI connector when you need organization/project scoping, API-key automation, or API billing controls. Use the ChatGPT connector when you want MangoStudio to use the models available through your ChatGPT subscription.

Remote Hosts

The OAuth redirect always targets localhost:1455 in the browser. If the API runs on a remote host but the browser runs on your laptop, that redirect will hit the laptop unless you forward the callback port to the remote API host.

Example:

ssh -L 1455:127.0.0.1:1455 user@remote-host

Keep the tunnel open while signing in, then open the MangoStudio UI through your normal remote access path. If the UI itself is also remote-only, forward the API/frontend port as usual in a separate tunnel.

Debug Overrides

The production endpoints are built in. Test and smoke harnesses can override them with:

[chatgpt]
auth_base_url = "https://auth.openai.com"
api_base_url = "https://chatgpt.com/backend-api/codex"

The same values can be supplied with MANGO_CHATGPT_AUTH_BASE_URL and MANGO_CHATGPT_BASE_URL. These overrides are for tests and debugging only; connectors are still created through the Settings sign-in flow.