Hello, I am using controlUrl to add messages to the assistants message history so it can track call time:

if (currentDurationTotalSeconds > 30 && callData.checkCount % 6 === 0) { 
            console.log(`[CALL TRACKING] Adding elapsed time system message: ${elapsedMinutes}m ${elapsedSeconds}s`);
            await fetch(controlUrl, {
              method: 'POST',
              headers: { 'Content-Type': 'application/json' },
              body: JSON.stringify({
                type: 'add-message',
                message: {
                  role: 'assistant',
                  content: `[Call time: ${elapsedMinutes} minutes ${elapsedSeconds} seconds]`
                },
                // No response needed for this informational message
                triggerResponseEnabled: false 
              })
            });
          }

This is using the **add-message **command, not the say command Sometimes the assistant correctly does not say anything, but then sometimes the assistant will speak out the message that was added Could you assist please?

Hi there! 👋 I understand your frustration with the assistant sometimes speaking out messages that were added via the

add-message

command, even though you've set

triggerResponseEnabled

to false. I'm unable to pinpoint the exact cause in your specific implementation without more details, but I can definitely help with some potential causes and solutions based on similar issues other Vapi users have experienced. Based on the information I found, here are some potential causes and solutions: ## Potential Causes 1. **System Message Processing**: According to [Discord discussions](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1338730919904673853), the issue might stem from how system messages are processed in the conversation buffer and immediately sent to the LLM stream, potentially triggering a response even when they shouldn't. 2. **Message Role**: I notice you're using

role: 'assistant'

in your message. This might be causing the assistant to treat it as its own message that needs to be spoken. 3. **Configuration Interference**: Other settings in your assistant configuration might be overriding the

triggerResponseEnabled

flag. According to [Discord discussions](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1330092161835733013), unexpected behavior can sometimes be caused by interference from other settings. 4. **Timing Issues**: The timing between message playback and action execution might not be properly synchronized, causing the assistant to speak unexpectedly. ## Potential Solutions 1. **Change Message Role**: Try changing the message role from

'assistant'

to `'system'`:

javascript
message: {
  role: 'system',
  content: `[Call time: ${elapsedMinutes} minutes ${elapsedSeconds} seconds]`
},

2. **Use Mute Control**: If the assistant starts speaking unexpectedly, you could use the

mute-assistant

control to stop it: ```javascript vapi.send({ type: 'control',

call ID: 9d18746d-e43e-454e-b5b2-33d5685a17a6 When i tried using the system role, the system is not aware of the call time

Hi, Can you provide time duration in the recording when the issue occurred?

I just realized i think Claude is seeing the pattern of call times and also returning the call times in its response:

call ID: 9d18746d-e43e-454e-b5b2-33d5685a17a6

16:51:08:372
[LOG]
Model output: [Call time:

16:51:08:373
[LOG]
Voice input: Any sensations, thoughts or emotions coming up? [Call time:

16:51:08:379
[LOG]
Model output: 1 minutes 53 seconds]

Looking into it

thank you

Hey Ethan, try adding this into your prompt: The current date and time is: {{"now" | date: "%b %d, %Y, %I%M %p", "America/Chicago"}} Change America/Chicago based on your location. Than prompt it to remember the exact time the call started. If this doesn't work, I'll look into it. Let me know.

hey @masondoberstein0 Thanks for the suggestion. This will give it the start date, but how would the LLM know how long it's been since the call began?

checking if this is resolved/solved for you?

thanks @Shubham Bajaj no, still need help with this. It's very odd. Even when I use add-message with the user role, the assistant doesn't seem to be aware of it at all:

// Send a message to the assistant about the client's local time
        await fetch(controlUrl, {
          method: 'POST',
          headers: { 'Content-Type': 'application/json' },
          body: JSON.stringify({
            type: 'add-message',
            message: {
              // Using 'user' role to mimic a system note injected into the conversation context
              role: 'user',
              content: `[Current time is ${userTime}]`
            },
            triggerResponseEnabled: false // Don't make the assistant respond to this
          })
        });

Call ID: 044adb5e-8efd-4a64-932a-0bb5945043ca https://cdn.discordapp.com/attachments/1361390885467848805/1365828659973460049/Screenshot_2025-04-27_at_00.13.43.png?ex=680eba90&is=680d6910&hm=4ab77e9600a6a46a2e1778a403cf1952b42e70dff9b0634d7e4efe495405f2ef&

Hey Ethan, the assistant never gets to know about time when you add it as a user role. It only uses the user messages to generate its next response. In case you want to pass some information, either pass it as role 'assistant' and try and make it work using it. If it doesn't work, then pass it as role'system'. Give her two to three test calls, try and let me know how it goes for you.

Hi Shubham I got it to work using role assistant , thank you. I'm wondering why it isnt able to see it if it's a system or user message though?

when pushing a message with role user, its consider as input to generate the next response. When pushing a message with role system or assistant, it's used to generate the next response.