How can I connect my chatgpt subscription?

[Helixoo]

I searched for it like 1 hour but could find and connect my subscription to litellm. I'm hosting on dokploy. I tried creating a config.yaml file and adding models in it then ran litellm --config config.yaml and it showed me a login link and I loged in but nothing happened.

[x-tahosin]

Quick heads-up that will save you a lot of time: a ChatGPT Plus/Pro subscription doesn't include API access — they're two separate products with separate billing.

• chat.openai.com — your subscription product (UI chat, GPTs, Sora, etc.)
• platform.openai.com — the API, billed per-token from a separate prepaid balance

There's no way to "connect" the subscription to LiteLLM (or to any other API client), because the subscription doesn't authenticate against the API endpoints LiteLLM calls. The login link you got was probably from LiteLLM's own SSO/admin login flow, not an OpenAI auth flow.

To use OpenAI models through LiteLLM you need:

1. Go to https://platform.openai.com/api-keys and create an API key.
2. Add prepaid credit at https://platform.openai.com/settings/organization/billing/overview (minimum $5).
3. Either set OPENAI_API_KEY=sk-... in your environment, or put it in your config.yaml:

model_list:
  - model_name: gpt-4o
    litellm_params:
      model: openai/gpt-4o
      api_key: os.environ/OPENAI_API_KEY

If you'd rather not pay separately for API credit, LiteLLM also supports a bunch of other providers (Anthropic, Gemini, Groq, OpenRouter, local Ollama, etc.) — same client, different litellm_params.model value. The supported providers list has the full set.

[Helixoo]

I know that, but that's not I'm looking for. In the other services you can log in with your codex account and use codex models thats what I'm trying to solve.

[thugilthedeveloper]

Im not really sure on if that's true cuz the LiteLLM docs does have a guide, but not a clear one

[reallyticsai]

Hey there, I can see the frustration with getting your ChatGPT subscription hooked up to LiteLLM. We've run into similar hiccups when integrating API-based models in production, especially with config setups on hosted environments like Dokploy. Let's walk through this step by step to make sure you're on track.

First, ensure your config.yaml is correctly formatted with your OpenAI API key (tied to your ChatGPT subscription). It should look something like this:

model_list:
  - model_name: chatgpt
    litellm_params:
      model: gpt-3.5-turbo
      api_key: sk-your-openai-api-key-here

After saving this, running litellm --config config.yaml should initialize the proxy server, and the login link you mentioned is likely for the UI dashboard. If nothing happens post-login, check if the server is running properly by hitting the /models endpoint (e.g., curl http://localhost:4000/models) to confirm the model is listed. If it’s not showing up, double-check your API key and ensure Dokploy’s network settings allow outbound requests to OpenAI’s API. We've seen cases where hosting platforms block certain API calls by default, so you might need to tweak firewall or egress rules. Drop a note if you’re still stuck—I can dig into logs or error messages with you.

[omni-front]

You need to make sure you're using an API key from the OpenAI API, not a ChatGPT subscription. Here's how you can integrate it with LiteLLM using a config file:

models:
  openai:
    enabled: true
    api_key: "your-openai-api-key"

Run litellm --config config.yaml and it should connect using the OpenAI API key. The ChatGPT subscription is separate and doesn't offer API access.

[NeoMetric]

The responses here makes no sense, the official docs say it's possible, but folks here say it's not!.

https://docs.litellm.ai/docs/providers/chatgpt

[omni-front]

It seems there might have been updates. If the official docs now support ChatGPT subscription, follow their guide. The older responses might be outdated.

[anthony-spruyt]

I have been battling for a few hours following the docs and attempting variations, I am able to attempt the device flow via the logs after adding the model and if I am quick I to auth I can avoid a restart loop, but then it still doesnt work. I get invalid model/doesnt exist. If I am a bit slow with the device flow then the pods go into restart loops.

Edit: I finally got it working, you can only do it via config, not via UI. You have to check the logs and very quickly do the device flow on all pods (I run it on my talos homelab). It only works for responses.