{{now}} and {{customer.number}} variables not avai...
# support
v
Hi everyone! 👋 I just joined the Discord — hope you're all doing well. I'm currently using a Workflow with the Cartesia voice (Sonic 2 – French language) and I have a few questions: Dynamic variables like {{now}} and {{customer.number}} don’t work : - The voice agent doesn’t recognize the current date or the incoming caller number, even though I’m using a Vapi free number and calling from a real phone. Has anyone faced a similar issue and found a fix ? Also, the agent pronounces long numbers digit by digit For example: - 67100 => should be pronounced (in french) “soixante-sept mille cent”, but it is spelt number by number instead “six sept un zéro zéro” Same issue with emails: the “@” and “.” are not pronounced naturally in French Even with instructions in the Global Prompt or node-specific prompts, pronunciation rules are ignored... I tried guiding the agent with custom instructions, but it still pronounces things strangely. Has anyone managed to solve these issues ? Would mean a lot if you could please share tips and feedback 🙂 Thanks !
r
Following! For context - I'm having a similar issue with the {{now}} date issue. I can't quite get the assistant to understand what day it is today. I've tried multiple ways, including {{now}}. In my case I want this for appointment booking so that if someone says "I want to book for Thursday" (meaning tomorrow) it could say back "got it! So you want to book for April 17th" or similar.
j
I've used this in my prompt : Date et heure en ce moment : {{ "now" | date: "%A, %B %d, %Y %H:%M %Z", "Europe/Brussels" }} That seems to work, my assistant is aware of day, date and time
v
Thank you. Will try right now and report if any issue.
Does it work with {{customer.number}} too ? Are you using Workflows ?
Just tried it. It worked the first time (okay ...) but then stops working. Still using October 2023 date. On one call, it said we were October 13th. On the next call, it said we were October 25th.
j
I'm using openai gpt-4o
Maybe that does influence
v
Are you using functions ? I am currently implementing a POST request. In workflows, you can specify the HTTP verb (GET vs POST), the JSON response (properties, variables), and the body. But in normal functions, I struggle implementing it. Do you use the VAPI UI to implement functions or code ?
j
To clarify, the "code snippet" {{ "now" | date: "%A, %B %d, %Y %H:%M %Z", "Europe/Brussels" }} is put in the system prompt of my assistant. Functions are a different things, I used them but not related to the dates. If I'm asking my assistant what day / hour it is, it answers correctly based on the prompt (actually it says the start hour of the call, as it's when the variable is instantiated).
Didn't try customer.number, and I'm not using workflows
k
Hey, I was out of office for a couple of days, so couldn't reply to you earlier. Checking if this is resolved for you?
v
Hey, its not 🙂
k
can you please provide some call ids for these instances where the date is incorrect?
v
Yes,
2f598fcc-5fc6-4b46-97f4-7bae2f493d13
7124ce65-ee63-4982-8246-a7b3a3633638
5cfe6e52-b4ac-4579-b6d6-a9feec3a4637
k
following, {{customer.number}} does not return a value when used in system prompt. she just says your phone number is customer.number
k
looking into it.
@Virgile, could you check the call IDs again? I guess you have said the wrong call ID. The assistant named Eva doesn't have any customer numbers {{customer.number}} inside it in its prompt. So I will suggest you take a look again for the call IDs. Let me know if you need any clarification or help with this.
Your queries are important to us, so we consider them as high priority. You can even DM me if required. Happy to help everyone here.
v
Hi @Shubham Bajaj, just did call
ded1a3d7-2cd6-46de-94fb-4b2aaec263a6
. I've not worked on the bot for a whole week. It seems that the service has improved since then. I really was about to give up. My problem is a lack of knowledge. I gave the right prompts, the right functions, but it seemed that the voice agent didn't care. There are still pronunciation problems in french though. It feels not polished enough, and clearly detectable as an AI. Our goal would be to have a voice agent, and only have a minimal number of customers realize they talk to a bot
I'd love it if you could point me towards resources / guides, or hop on a call with me to tell me what I'm doing wrong. I've spent a few weeks working on the bot now and feel like I reached a glass ceiling of what I'm about to do with Vapi.
s
Hi @User , I appreciate you sharing your experience with Vapi. To help solve these issues properly, I'd like to collect some specific details about the problems you're encountering. Could you please share: - Call IDs for a few examples where you experienced issues - Specific timestamps when pronunciation problems occurred - The exact prompts/functions you provided - What response you expected vs. what you actually received - Examples of French phrases with pronunciation problems This information will help me investigate exactly what's happening and provide targeted solutions
v
Since last week, our experience with Vapi has improved. She is aware of the current day and of the customer phone when injecting the variables in the prompt. She accesses the knowledge-base successfully. Most of our problems are gone. Now our goal is to make the voice more natural sounding
k
Glad it's working as expected. Can you elaborate on what you mean by a natural-sounding voice? What have you done so far to achieve it? I can suggest something aligned with your existing positions or something new.
6 Views