Error invoking model response: invalid x-api-key

Hi @alexmoreno - invalid x-api-key is usually what Anthropic throws as an error when the API key is wrong. Did you perhaps setup Anthropic before you setup OpenAI?

What you can do is the following:

1. Install AI API Explorer and visit /admin/config/ai/explorers/chat_generator. There you can pick any OpenAI model and write a chat message and see if the error persists or if it works.
2. If it works and you did setup Anthropic first, could you visit the default models and setup so OpenAI gpt-4o is the default provider and model for Chat and Chat with Complex JSON. You can do that here: /admin/config/ai/settings. These are the providers/models the chatbot in Drupal CMS uses by default.
3. After that try the chatbot again

um, could be this the issue?

You exceeded your current quota, please check your plan and billing details. For more information on this error, read the docs: https://platform.openai.com/docs/guides/error-codes/api-errors.

I actually did not setup Anthropic. Is that a requirement?

No, you should not need to configure it - its just that this specific error message "Error invoking model response: invalid x-api-key", points to an Anthropic authentication key.

If you have the time, could you check the following to figure out if this is a bug in Drupal CMS installer:

1. Check if there is a key setup here for Anthropic: /admin/config/system/keys
2. Check if Anthropic is setup here: /admin/config/ai/providers/anthropic
3. Take a screenshot of the following page: /admin/config/ai/settings

Could you also answer the following questions:

• Did you setup from the repo or via some service?
• Version of Drupal CMS, if you know it
• Did you apply the recipe via Drush or via Project Browser?

Thanks!

Hi Marcus,

1. No, there was no key there, I added my own trying to fix this.
2. there was not, but same, I added one after the errors, and it did not fix anything
3. Screenshot Attached

I did install the site via UI in Pantheon, then I installed the AI recipe with this command:

terminus drush drupal-cms-from-platform-upstream-alexmoreno.dev recipe ../recipes/drupal_cms_ai

Drupal Version - 11.1.1

Thanks a lot

That looks about right on the settings page. The gpt-4o-mini and o1-preview-mini as chat model are untested with the agents, but I don't think that has anything to do with your issue at hand. gpt-4o is the mostly test.

You don't need to add Anthropic - that was just based on the error message.

So with this setup the issue is still that:

1. The chatbot is visible and you can send a message.
2. When you send a message, you complete error with the retry button.
3. When you look in the log, the error message is "Error invoking model response: invalid x-api-key"?

Yes, exactly. Added a couple more screenshots, one with the error, the other with the log.

Interestingly, now the error is that I have exceeded the quota. I have a paid account, is it possible to send some parameters to use my own user in openai?

You exceeded your current quota, please check your plan and billing details. For more information on this error, read the docs: https://platform.openai.com/docs/guides/error-codes/api-errors.

Ah, that error makes more sense at least, since its an OpenAPI error.

Could you verify that you have money on that account and if you do, can you check this page what kind of limits you have: https://platform.openai.com/settings/organization/limits, also which tier you are on.

It should be however that tier 1 should be enough to ask ~5-10 requests per minute without any issues.

If you are on the free tier, that will unfortunately not work with the agents.

If you want to change which user you use, you would have to create a new api key with the other user to use.

Also to save money, I would suggest that you use gpt-4o intead on o1-preview.

ok, that's it, it was simply a question of adding money on my account.

Tests started to work in admin/config/ai/explorers/chat_generator and the chat errors dissapeared for a couple of queries...

But then, when I asked to create a content type I think I broke it haha

Error invoking model response: Unsupported value: 'messages[0].role' does not support 'system' with this model.

Ah, that is a known limitation for o1 models. For the production release of the OpenAI provider we have a work around for it, see 📌 O1 does not allow system role Active . So either you can:

1. get the production release of the OpenAI provider (which will be in Drupal CMS release the 15th)
2. or a simple fix is that you can switch on the settings page to use gpt-4o as the default Chat with Complex Json model. You do it here: /admin/config/ai/settings. o1-preview is a lot more expensive and also slower, and the results might not be that different

I hope that makes it work :)

um, in settings I only have 01-preview available.

hmm, do you see gpt-4o model in the chat explorer?

oh yes, you are right. From the chat explorer I can select admin/config/ai/explorers/chat_generator

but not for the settings in admin/config/ai/settings

Actually, can you look for gpt-4o, not gpt-4o-mini and see if you see that in chat explorer?

Nop

Hmm, so there are two things I can think of here to further debug:

• Try a manual curl call to see what models are given back: https://platform.openai.com/docs/api-reference/models/list?lang=curl
• Check if some models are disabled for the project the API key is running on. You can do that, if you have access to the account behind it, by going to https://platform.openai.com/docs/overview. Click on the Manage Projects and click limits. There you can see if some models are disabled. gpt-4o is by far the best combination of speed, reliancy and cost.

The simple solution is to update the OpenAI Provider → to 1.0.0. Then you should be able to use o1-preview.

I also read up on the tiers, and it looks like tier1 is after paying $5. So if you put in less than that, it could also be the issue.

Sorry, about all the messages back and forth :)

I dont see any limits, gpt-4o is there for me

ok, I took option two, and composer updated the provider. That worked... However, things I'm finding on my first tests, and I realise that this may be misconfiguration on my side.

When I create a content type, that works fine.

When I ask to create a field, that fails with a timeout

Location	https://dev-drupal-cms-xxxx.pantheonsite.io/api/deepchat

Error invoking model response: cURL error 28: Operation timed out after 30000 milliseconds with 0 bytes received (see https://curl.haxx.se/libcurl/c/libcurl-errors.html) for https://api.openai.com/v1/chat/completions

Yeah, that becomes a problem with the o1 models - they are reason models and need time to think, so it's a little bit overkill for such "easy" tasks as most agents perform.

It is guzzle (so the http client) timing out in this case after 30 seconds. You could raise that time in settings.php, but I'm not sure if thats the right approach. See here for instance: https://www.jeffgeerling.com/blog/2016/increase-guzzle-http-client-reque...

Do you know how to use curl? One thing that would be interesting is if this is a bug in the OpenAI provider or not. If you run:

curl https://api.openai.com/v1/models -H "Authorization: Bearer {apikey}" | grep "\"gpt-4o\""

with replacing {apikey} with your api key. Do you see any results?

ok, that worked finallyyy haha :-D.

I added this in my settings.php (Located in sites/default/settings.php in Pantheon):

/** * HTTP Client config. */
$settings['http_client_config']['timeout'] = 1800;

And the I repeated the ask to ChatGPT, but this time I got this answer:

For that curl you sent me, this is the reply I got:

...  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100  6233  100  6233    0     0    893      0  0:00:06  0:00:06 --:--:--  1653
      "id": "gpt-4o",

Automatically closed - issue fixed for 2 weeks with no activity.