Feature Requests

More SOTA TTS models like Kokoro-82M. Sesame new CSM, F5-TTS, Zonos (Zyphra)

If someone asks the agent to slow down or repeat something have a way to adjust the voice speed.

Im experiencing a technical problem with our outbound calling agent's ability to navigate IVR systems. While using Vapi's built-in DTMF functionality, the system logs indicate successful function calls with the correct digits, but many IVR systems fail to detect these inputs. This issue has been observed across multiple IVR systems, though it functions correctly with some. Affected Call IDs dde63818-b547-41eb-bfd3-9e587cddd2ab df13e533-4b27-4cd7-9c07-54e7f61a807f ea0db2d8-a59f-4ae9-9b50-d49116efc056 1f5dd938-b0e6-40f2-b549-d9a6280a3606

Currently, users deploying VAPI through Azure OpenAI in the EU region face limitations due to unavailable model-region combinations. Specifically, attempts to use GPT-4o-2024-08-06 and GPT-4o-2024-11-20 in the Sweden Central region result in errors such as "Pipeline no available mo...". To comply with EU data protection regulations, many users require a reliable EU-based deployment (or alternatives in Norway, Switzerland, or the UK). However, VAPI does not yet support these newer model versions in Sweden Central, despite their availability in Azure OpenAI. Additionally, the VAPI dashboard (Web UI) lacks an Azure OpenAI option in the Model → Provider dropdown, requiring users to manually trigger API calls to connect models. Requested Features: Enable support for GPT-4o-2024-11-20 in the Sweden Central region. Expand model availability in other EU-compliant regions (France, Germany, etc.). Provide a clear, updated documentation page listing supported Azure OpenAI models per region. These improvements will unblock EU users who need to follow strict data protection rules while optimizing their VAPI-Azure OpenAI deployments. Your support in prioritizing this feature would be greatly appreciated!

11labs has an unbelievable new language detection tool in their conversational AI. I successfully tested it in their playground. It detects language change in two ways : Once asked by the user (in any language) Once the user talks in another language Is there a way to get this to VAPI quickly ? 😬 Game changer for Belgium where multiple languages are spoken. https://www.linkedin.com/posts/elevenlabsio_conversational-ai-now-supports-automatic-activity-7310336522229096449-QXUu

We have a lot of times where the ai and the person start talking at the same time. then they both stop, then they both go ahead and it repeats over again. In normal conversation, if that happened, a human would say, "No, you go ahead". A feature that allowed you to set the number of interruptions that were detected, and then if it was above a certain threshold would say "Go ahead" and wait a few seconds while the person finished.

There should be a scroll feature to scroll assistants column and also to put them in categrories or columns, archive them, and save them per service

We have much better results using Open AI for international languages like portuguese, spanish, etc.

It would be great if Zyphra ZONOS model is added to VAPI. Zonos-v0.1 is featuring two expressive and real-time text-to-speech (TTS) models with high-fidelity voice cloning. https://www.zyphra.com/post/beta-release-of-zonos-v0-1 https://www.zyphra.com/

Right now it's extremely restrictive and silly. Because it's presuming what will be most helpful for the assistant without knowing the context of how the assistant is engaging with an end-user. Also there's no details within the UI that states that the assistant will randomly select from the array of strings that is to be the response to be said upon silence. It confuses the user/developer (us). Idle messages can be described instead as: "What do you want the assistant to say to break the silence if there is one?" "How long should the assistant wait, before breaking the silence?" Etc. Lots of descriptions are too "technically" oriented it's not helpful nor is it tailored to those who may be using the assistant to create value for the end-user. Just an opinion.