For our calls the analysis has been coming out empty {} lately. I have tried all the know fixes i have seen. The issue seems to come and go since some calls a couple of days ago worked with no change. Upon inspecting the affected call logs i see in all the analysis items some errors: 1. Analysis Summary error All model completion attempts failed. Last error: TypeError: Cannot read properties of undefined (reading '0') 2. The API deployment for this resource does not exist. If you created the deployment within the last 5 minutes, please wait a moment and try again. There are also some issues apparently with Anthropic So something appears to be broken internally at least with our constellation. I attached a screenshot of how i am creating the analysis plan for the assistant creation using the [vapi sdk](https://github.com/VapiAI/server-sdk-typescript) assistants.create method- Here are some of the affected call ids: 58fe2415-bae1-4e6b-b6bc-c75908785e0c 335bd57c-1b17-426e-88b0-240f084ebce0 a1a75dbc-8b12-40c5-aade-a63a681a0b45 This is blocking us massively so we would appreciate some help. https://cdn.discordapp.com/attachments/1413655590600839359/1413655591204814869/Screenshot_2025-09-06_at_00.39.27.jpg?ex=68bcb8dd&is=68bb675d&hm=3919493b4ac499b2ac0a2391ce2092b54cc686ab27c5ec3b73f7d5d12c916cb2&

Sounds like the issue isn’t on your end since the errors you’re seeing (undefined property, missing deployment, Anthropic failures) usually mean something is off in how the analysis plan or model reference is being handled inside Vapi. I can help dig into how you’re setting up assistants.create with the SDK and make sure the analysis plan is structured right. Quick question: are you noticing this only with Anthropic or across OpenAI too? That’ll help narrow it down. @Steyt

I am not sure, we use the openai provider with GPT 4o cluster. I do not have more information than what i provided in my post.

Since you’re on GPT-4o, those errors usually come from how the analysis plan is being passed into the assistants.create call. I can jump in, review the way your assistant is being created in the SDK, and adjust the analysis plan so it runs cleanly on the GPT-4o cluster. If you share the snippet of how you’re building the assistant (without sensitive keys), I’ll handle the debugging and fix directly. Would you like me to help you directly @Steyt ?

Yes i would like you to help me directly debug this. I attached the way i create the assistant using the sdk. Please keep in mind that this has worked in the past with no issues, it seems to come and go. Let me know if you require more information. https://cdn.discordapp.com/attachments/1413655590600839359/1414196283312181268/vapiAssistantCreation.ts?ex=68beb06c&is=68bd5eec&hm=6199d9576927a24c24fe3e8b5081a3b989907986a495ebdee2c49bff329930b2&

Hey! To help track down this issue, could you share: \- When exactly this happened (the timestamp) \- What response you expected to get \- What response you actually got instead This would really help us figure out what went wrong!

I put all the data i have as well as the call id's. I just simply expect to get a proper analysis not a empty one {}. And if it fails i would like to more clearly know why.

We will gather all the data from the info you provided and give an update shortly