xAI Grok Transcription API Key Permissions

[loyer]

Does anyone know what Models and Endpoints are required for the xAI Grok API key to work with 3CX voicemail transcription? I am looking to lockdown the API to the minimum set of permissions needed.

In the model dropdown it offers: Chat, Image, Video
In the endpoint dropdown it offers about 10 options (see screenshot).




I am hoping to avoid trail and error by randomly choosing.
Thank you!

 

[TheodorosG_3CX]

It will depend on the model selection as well:



Grok Voice will give you Realtime, TTS, STT, voice-agent, and custom-voice docs since you need realtime voice, speech-to-text, text-to-speech, or custom voices.

 

[loyer]

Ok, I adjusted the settings and found the configuration below to work.

On the 3CX Side:
- Analysis Text Model: grok-4.3
- Stereo Mode: checked (on)

On xAI API Key:
- Models: Video
- Endpoints: Voice
- Tokens per minute: 2000 *****
- Requests per minute: 100 *****
- Expiry: No expiry



***** I am only using xAI/Grok for voicemail transcription. You may need to adjust your limits if you use this API key for other purposes. I based my requests per minute based upon the max voicemails I think my phone system could get in one minute. I noticed that there seems to be a 16:1 token to request ratio ---- meaning for every voicemail that gets transcribed, it uses 16 token. So to keep the math easy and to add in some padding, I set the token per minute equal to the requests x 20.


My goal is to find the least permissions required to enable transaction while limiting the API key's exposure level.

3CX team, please let me know if I am missing anything or got something wrong.

Thank you.

 

[nikolascx]

From what I see, you are on the right track. Keep us posted