Please respond. This is the second post on this topic. Please see assistant id 03555353-b9ea-4825-b5fc-6102b84dc756 example call id 019a074e-1e13-7ff4-9b17-2253d85954a5 the liquid js is not rendering, but when i put it in to the liquid js renderer on liquidjs.com , it is rendering correctly. Please advise how to fix, or implement fix on your end. last post: https://discord.com/channels/1211482211119796234/1430211274855415898/1432367763250221190 Additionally , for this call log: 019a0776-cf29-7ddb-a181-e266fd1158a0 We called an assistant that had previously been working, and now it is not. I believe you have had a regression in your liquid js templating and it is not not evaluating correctly for assign variables. Please fix - we need this badly and cannot really use the product without it for our use case assistant id for the message directly above: d4348686-0758-47a8-ab67-e71536d077e2 working old call log example: ee756faf-b4ce-4b94-abcf-28900640cdf2 where it is rendering correctly and responding 33bee5fb-4792-42be-b2a0-ab7b5a126e02 another one form early september where it is working and one more b6f68b49-aea3-46d2-9cfb-61ff50d8dba9

Try isolating the issue with this minimal LiquidJS template in your Vapi assistant prompt:

liquid
{% assign testVar = "Hello from LiquidJS!" %}
{{ testVar }}

If this does not render as expected in Vapi, it confirms a platform-level regression with

assign

. In that case, please report this minimal failing example (along with your call IDs) to Vapi support for immediate escalation, as suggested in the [issue reporting guide](https://docs.vapi.ai/issue-reporting). Source: - [Issue Reporting Guide](https://docs.vapi.ai/issue-reporting)

@Shubham Bajaj @User @User

Hey , thanks for sharing your details. Our team will look into it and get back to you soon with an update.

thank you

@jsieber-eip I think i got the same issue, is it when you assign something like at the top of your prompt and you can't use it at the end for example right? I thought i was going crazy because this made no sense, my injected variables work anywhere but the assignments don't

I’m experiencing it that any assignments don’t work - you’re correct. Injected variables are working but the “assign” function in liquid js is not

Hi could you please provide some recent call ids made within the last >7 days? We tested it on our side as far as assistant-request webhooks with assistantOverride.variableValues as liquidJS and assignment within the prompt then calling those values in a phone call and both worked.

I provided the specific information on the support email chain thats going on - moving it to that

REOPEN

Just tested this in my prompt and it worked. Could you please try again and send us the call id if it doesnt work?

Still doesnt work for me call id: 019ae6ad-1dbe-733c-9918-7807322b2654 Here's how I debug: - open call logs, filter by models - select Azure OpenAI HTTP request - check attributes.request.messages:

[ { "role": "system", "content": "# AI Phone Assistant System Prompt\n\n- Current date: \n- Day of the week: \n- Current time: \n ........

- currently showing/replacing with empty string instead of the actual time. Can confirm that {{now}} and {{customer.number}} works

Could you try using OpenAI GPT-4.1 directly instead of Azure?

I already set the model as OpenAI GPT 4.1 in the assistant. but somehow in the call logs, it use azure-openai? Can you have a look? https://cdn.discordapp.com/attachments/1434935239381352509/1446272811210506362/image.png?ex=694139c4&is=693fe844&hm=ec1b81dce9ae0579b8788a0783f4f150d70052613d87592eb1f16a97f15762b8&

Hi there, We’re checking with our team to confirm that the assistant is using the correct provider configuration. Once we verify that everything is aligned on the backend, we’ll share an update with you. Thanks for your patience while we look into this. Regards, Vapi Support

Is there any updates on this?

Azure OpenAI will be selected as a fallback if the deployment has better latency

Is that going to affect the liquidJS or variables mapping?

it shouldnt. the liquidJS implementation is mostly embedded within VAPI and not the LLM

So, what the root cause of my issue, and also what's the workaround?