I can’t complete assistant setup up in Reality Composer Pro 3 beta

I can’t complete assistant setup up in Reality Composer Pro 3 beta. I filled in the set up fields for OpenAI and clicked done. However, when I click connect nothing happens.

Answered by Vision Pro Engineer in 891106022

Hey @reflexx,

Can you make sure that there are credits on your account? Depending on the provider, if there's no credits associated with the API key it may fail. The expected URL for Anthropic is https://api.anthropic.com/v1.

If you're still experiencing an issue please open a bug report, and include the model provider URL, API Key Header, and a redacted version of the API Key field, and post the FB number here once you do.

Here are the values that work for three popular providers:

OpenAI
URL: https://api.openai.com/v1
API Key Header: Authorization

Anthropic
URL: https://api.anthropic.com/v1
API Key Header: x-api-key

Gemini
URL: https://generativelanguage.googleapis.com/v1beta/openai
API Key Header: Authorization

Hope this helps,
Michael

Hey @aandyzoom01,

Please follow the process outlined in Add a model provider. Are you sure you're providing the correct values in the dialog?

If you're still experiencing an issue please open a bug report, and include the model provider URL, API Key Header, and a redacted version of the API Key field, and post the FB number here once you do.

Bug Reporting: How and Why? has tips on creating your bug report.

Thanks,
Michael

I've followed the steps for Add a model provider. I have added the correct values in each field of the dialog as best I understand based on latest ChatGPT instructions. I am adding OpenAI as the provider. I am able to create the provider, but it has a red dot saying not connected. I click the Connect button but nothing happens and the provider continues to show not connected. I will try a different provider, but anything else to check before I submit a bug ticket. Seems like the basics of creating an assistant should work without having to do a bug.

By the way, I added a different provider (OpenRouter) and it worked. Not sure why OpenAI provider setup did not work (would not connect).

Accepted Answer

I was able text get OpenAI provider to work. The exact field entries for the url, API key and API Key Header were not easy to figure out. It took offline (from Apple documentation) to find out the correct expectations for the fields for OpenAI to allow a model assistant connection.

I'm in the same situation. Is there any way of troubleshooting the connection problems?

I'm using Claude. For the API, I've tried https://api.anthropic.com as well as appending /v1 and /v1/messages, all without success.

I've also tried a cURL call (using the /v1/messages version) and do get a response.

Hey @reflexx,

Can you make sure that there are credits on your account? Depending on the provider, if there's no credits associated with the API key it may fail. The expected URL for Anthropic is https://api.anthropic.com/v1.

If you're still experiencing an issue please open a bug report, and include the model provider URL, API Key Header, and a redacted version of the API Key field, and post the FB number here once you do.

Here are the values that work for three popular providers:

OpenAI
URL: https://api.openai.com/v1
API Key Header: Authorization

Anthropic
URL: https://api.anthropic.com/v1
API Key Header: x-api-key

Gemini
URL: https://generativelanguage.googleapis.com/v1beta/openai
API Key Header: Authorization

Hope this helps,
Michael

I was able to get it to connect with ollama, however when it began a conversation, it worked once and then came back with errors when I tried to have it generate an image.

By the way, the field for API key has a placeholder that says optional. However, at least for OpenAI, I don’t believe the API key is optional. I think it’s required, at least it was required for me when I tried to create a provider for OpenAI. I think I had to put the API key in before it actually worked. Can someone from Apple comment on API key field being optional or not for various API vendors?

I believe the interpretation is that Apple doesn't require the user to enter it, as they may be using self hosted model providers etc.

Any provider who is charging for usage would require one for token usage purposes.

It's like Mail says / said a SMTP password is optional, but you won't find any mainstream email hosts who allow unauthenticated users.

If you're still experiencing an issue please open a bug report, and include the model provider URL, API Key Header, and a redacted version of the API Key field, and post the FB number here once you do.

Have filed report FB23043902.

Regarding your suggestion of being out of tokens, as mentioned, I was able to execute a cURL call with the same details and receive a reply from the model.

Hey @reflexx,

I was wrong earlier. You do need to provide x-api-key as the value for the API Key Header. There is no default value for this field. Can you try again with this provided value and let me know if it works?

Thanks,
Michael

Now that's interesting. It worked with the combination of having the /v1 in the url, and x-api-key typed in.

However, once it connected, it stayed valid (including after a restart) even if the v1 and key header were removed again.

Another issue I'm seeing (don't know if anyone else is) is the settings window presents behind the main window so is hidden unless the user moves things around. I've included that bit in by FB but wanted to mention it here as it might throw a few people out.

I can’t complete assistant setup up in Reality Composer Pro 3 beta
 
 
Q