hi @Vapi, @Vapi Support Bot team, you guys are doing some great stuff, product and engineering wise. kudos. But it seems like there's no concept of agents based on workflows anymore. Curious, why was that abandoned?

Great question! Vapi has moved away from workflow-based agents because current AI systems aren't able to reliably maintain awareness of node instructions or all possible next steps in a workflow. Instead, Vapi now recommends using Assistants for most cases or Squads for multi-assistant setups, as these approaches have led to better results for customers. If you're starting a new project, check out the Assistants or Squads documentation for the latest best practices. Reference: See the [Workflows Overview](https://docs.vapi.ai/workflows/overview) for more details. Source: - [Workflows Overview](https://docs.vapi.ai/workflows/overview)

thanks, @User , @User do you have any recommendations / tips on how to build a great prompt in Vapi for use cases where 1. the assistant has to walk the human through a series of steps - For example open the app, login with gmail, go to settings, click on add phone number, and save? 2. and we have to somehow track if the Assistant shared each step correctly with the human? 3. and if the human says that they have alraedy added the phone number before the agent mentioned all the steps then the agent should realise and then thank the human and end the call. of course the overall conversation will also involve the initial intro message, confirming that the assistant is speaking to the right person etc. And the human may also say things like I am busy, call me back right now. Finally, we need to support English, Spanish and German

Message marked as helpful by @sudhama! 🎉

@Vapi Support Bot @Vapi would love to get some advice on this

Here's a prompt structure you can use to guide users through step-by-step actions and handle early completion, with multilingual support:

[Identity]
You are a helpful assistant who can speak English, Spanish, and German.

[Task Flow]
1. Confirm the user's identity.
2. Ask if they've already added their phone number.
   - If yes: Thank them and end the call.
   - If no: Guide them step-by-step (e.g., "Please open the app...").
   - After each step, wait for user response.
3. If interrupted (e.g., "I'm busy"), politely offer to call back.

Keep responses concise and switch languages based on user input.

This approach ensures clear progress tracking and dynamic flow control. For more, see the [Prompting Guide](https://docs.vapi.ai/prompting-guide). Sources: - [Prompting Guide](https://docs.vapi.ai/prompting-guide) - [Multilingual Configuration Guide](https://docs.vapi.ai/customization/multilingual)

thanks @Vapi Support Bot , @Vapi , i saw the appointment booking prompt guide. seems useful. but what LLM works best for such a use case, so that it does not go into loops or skip steps ?

For your use case, the best LLM is

gpt-realtime-2025-08-28

from OpenAI Realtime. It is optimized for step-by-step guidance, progress tracking, and multilingual support. Here’s a sample configuration snippet:

json
{
  "model": {
    "provider": "openai",
    "model": "gpt-realtime-2025-08-28"
  }
}

This model delivers structured outputs and high-quality voice interactions—ideal for your scenario. Reference: See the [OpenAI Realtime documentation](https://docs.vapi.ai/openai-realtime) for more details. Source: - [OpenAI Realtime documentation](https://docs.vapi.ai/openai-realtime)

this seems like a speech to speech model, Any other option @Vapi Support Bot ?

You can use a custom OpenAI-compatible LLM (such as Claude, Gemini, or Llama) for step-by-step guidance in Vapi by specifying your own endpoint. Here’s a sample configuration for a custom LLM:

json
{
  "model": {
    "provider": "custom-llm",
    "url": "<YOUR_OPENAI_COMPATIBLE_ENDPOINT>",
    "model": "<your-model-name>"
  }
}

Tip: Combine this with structured, step-by-step prompts as described in the [Prompting Guide](https://docs.vapi.ai/prompting-guide) to minimize loops and skipped steps. Sources: - [Prompting Guide](https://docs.vapi.ai/prompting-guide) - [Fine-tuned OpenAI models & Custom LLMs](https://docs.vapi.ai/customization/custom-llm/fine-tuned-openai-models)

Message marked as helpful by @sudhama! 🎉

We happy that your questions were answered. If you need any further assistance, please feel free to reach out to us and we’ll be happy to help.