What we need is: assistant 1 greets customer (assistant-speaks-first) When conversation returns ro assistant 1 from assistant 2: assistant must speak first but with a content gwnerated mossage

empty first message\*

Hi! I think this is not working as expected

I have the assistant set with speak first but because i am using the squad, the assistant is waiting for the customer to speak (or is not speaking first)

17495b6a-9307-4579-816e-a0bb96945a9e

When we talk to customers this is really having a negative impact on experience

Please compare with this one that uses only the first assistant of the squad (5e79414c-6929-48c2-8576-6e8c3daf6cb9)

Call ID: 17495b6a-9307-4579-816e-a0bb96945a9e starts talking at around 0:03 (using the squad) Call ID: 5e79414c-6929-48c2-8576-6e8c3daf6cb9 starts taling since 0:00 (using only the assistant)

It seems like the AI waits and if customer doesn't talks, the AI then talks

Hi! Did you have a chance to check the calls?

Call ID: 17495b6a-9307-4579-816e-a0bb96945a9e https://storage.vapi.ai/17495b6a-9307-4579-816e-a0bb96945a9e-1740150689627-f88d5058-3a88-4fa6-95ee-19b3ea540c97-mono.wav 🟢 15:11:14:336 CallId

17495b6a-9307-4579-816e-a0bb96945a9e

Call Picked Up. 🔵 15:11:14:652 Deepgram First Transcript Received: 🔵 15:11:15:997 [user CHECKPOINT] Model sent first output token 🔵 15:11:16:512 Voice Input Formatted: "Hola, soy Celeste de Kavak. Noté que estás interesado en comprar un auto.", Original: "Hola, soy Celeste de Kavak. Noté que estás interesado en comprar un auto." "firstMessage": "Hola, soy Celeste de Kavak, ¿Tengo el gusto con ?", "firstMessageMode": "assistant-speaks-first-with-model-generated-message", -------__----------____________________________---------------------___________ Call ID: 5e79414c-6929-48c2-8576-6e8c3daf6cb9 https://storage.vapi.ai/5e79414c-6929-48c2-8576-6e8c3daf6cb9-1740410009940-f4ac2d01-f824-4e65-9d3e-1117be115879-stereo.wav 🟢 15:13:12:171 CallId

5e79414c-6929-48c2-8576-6e8c3daf6cb9

Call Picked Up. 🔵 15:13:12:171 Say Request...61 (say: "Hola, soy Celeste de Kavak, ¿Tengo el gusto con mateo bernal?", interruptible: false, audio: false, endCallOnError: true, addToModelOutput: true, endCallAfterSpoken: false) 🔵 15:13:12:181 [user LOG] Voice cached: Hola, soy Celeste de Kavak, ¿Tengo el gusto con mateo bernal? 🔵 15:13:12:181 Say Request Saying 61... (say: Hola, soy Celeste de Kavak, ¿Tengo el gusto con mateo bernal?)

@Juanjo As evidenced in the logs and information provided above for the squad call ID, the model is generating the response, which is causing it to take approximately three seconds more than the required time. In contrast, when using only the assistant call ID, the first message was utilized, and it was cached, resulting in a voice response within a sub-second of the call.

I mean: I need that the call starts with my assistant talking but to keep the configuration of assistant-speaks-first-with-model-generated-message on the transfers

I need the assistant to greet the customer at the beggining of the call

Example: 0:00 Assistant 1: Hello I am your assistant, how can I help you? 0:01 Customer: Hello I want help on this task this task is done by assistant 2 Transfer to assistant 2 0:03 Assistant 2: Sure lets do that... . . . 0:30 Customer: Great now I need this This task is done by assistant 1 0:32 Assistant 1: Sure let me help you

Here's the detailed configuration for each: 1\. **Assistant_1A (Initial Greeter)**:

{
  name: "Assistant_1A",
  firstMessage: "Hello I am your assistant, how can I help you?",
  firstMessageMode: "assistant-speaks-first",
  model: {
    messages: [{
      role: "system",
      content: "You are the initial greeter. After greeting, evaluate the customer's need and transfer to the appropriate assistant."
    }]
  }
}

2\. **Assistant_2 (Specialist - Unchanged)**:

{
  name: "Assistant_2",
  firstMessage: "",
  firstMessageMode: "assistant-speaks-first-with-model-generated-message",
  model: {
    messages: [{
      role: "system",
      content: "You handle specific tasks. When task is complete, transfer back to Assistant_1B."
    }]
  }
}

3\. **Assistant_1B (Continuation)**:

{
  name: "Assistant_1B",
  firstMessage: "",
  firstMessageMode: "assistant-speaks-first-with-model-generated-message",
  model: {
    messages: [{
      role: "system",
      content: "You are the continuation of the conversation. Generate contextual responses based on the conversation history."
    }]
  }
}

The error that I see is `pipline-error-anthropic-llm-failed`:

Model request failed (attempt #2, reason: (messages.7: `tool_use` ids were found without `tool_result` blocks immediately after: toolu_01TYNqJ7H3ynwcsFeznbephM. Each `tool_use` block must have a corresponding `tool_result` block in the next message.))

Actually, this seems to fail every single time a tool is used, irregardless of the stage of the conversation. Id: ff3771b2-c241-41e1-892f-ebdb573083f7