I think there is a serious issues with custom functions in Conversation flow agents. Function has static text. Speak During Execution is On Wait for Result is On Call_id: call_b746f2779fa99b634c418cd1e6b Static text is 'played' after functions is executed, and not during execution to fill the silence gap. Is it me who interpretates it wrong or it is a genuine issue with custom functions. How I understand it should work : 1. Customer makes a request 2. custom function is fired 3. Static text is played/or prompt 4. response is received after static text or during static text 5. conversation is moved to the next node based on the response. Where am I wrong? https://cdn.discordapp.com/attachments/1460607322358546567/1460607322572193843/image.png?ex=69795452&is=697802d2&hm=68a07c55445d632c7a3c52d5265e68b8a3c268fc1a2245a159a07dfabbe08ddc& https://cdn.discordapp.com/attachments/1460607322358546567/1460607322987565193/image.png?ex=69795452&is=697802d2&hm=8eeea3419d28a8121082dd167dc7864e137da5e09bde25fcbedb15d59ff9ce99&

Hello @Q4 , does this happen often , if yes , can you share more call ids? and I will report it as bug.

call_b0447092e93500aa1524f467189 https://cdn.discordapp.com/attachments/1460607322358546567/1461289082192793641/image.png?ex=696a0302&is=6968b182&hm=baf5a5c4cca72790970e918e76f7f07bcbbfc19fcda17a78daad305f2302b8e9&

I see it a latency issue, try to use faster model like , GPT 4.1

that agent version and function nodes had gpt 4.1 set on them already. To push to better performance I had to set them to gpt4.1 Fast and remove speech during execution and make them speak before or after (depends if I need wait for the result). So for this example when I retrieving some data I need to speak and tel the customer it will take some time so what i did, i added another conversasion node with thect before function. Which works but its a workaround for the something that should work out of the box - and here I am talking about "function" nodes.

Can you share a recent one , because their a over 100 new version.

do you mean recent agent or recent call?

A recent call

no, I have all function on on gpt4.1 Fast with "Speak During Execution" == false.

@User @User Having same issue. Cannot get ANY nodes to speak on execution BEFORE the function completes. Also cannot even get a Conversation Node to say static sentence with skip user response if the next node is a function node. It only says the sentence of the prior conversation node, AFTER the next function node completes. How do we fix this? I have also even created a dummy function with speak on execution for it before the real function, and it will not even speak on execution the dummy function. They all execute only after the main function completes.

my workaround which is not great but works so far: custom function (no speak, no wait for response on gpt4.1 fast, with Transition set on 'else' -> conversation node, where AI speaks static/prompt with skip response -> move to the node you need to be. Issues with custom functions that I know: - if Speak during execution is off && wait for response is on --->> this shit will speak if your function has much data to get 🙂 - if Speak during execution is On && wait for response is off --->> this shit still waits for response to start their part to speak 🙂 - if Speak during execution is On && wait for response is on --->> as above 🙂 Best option don't speak and don't wait for response just make use next node don't need data that you retrieving into conversation lol

@Q4 @Omar Ashraf | Community Lead It makes no sense at all. What is the point to skip waiting user response if it won't even speak? We can't even get the function node to speak on execution either, it only speaks at end of the function running. We simply want to add a small sentence that says please hold a moment

thats my workaround until Retell will fix the issue 🙂 I just move to another node to say that exact sentence 🙂