Feature Requests

Could we please get Gemini Flash 3 as a model option for both chat and calls

European server for VAPI, so that European developers can adhere to GDPR rules and can minimize latency for Voice calls. Would highly improve call quality and integration process. Data residency included in EU. Would also help with downtime, if US servers go down, the EU server could take over.

I would like an option to configure a sort of "no-op" assistant, one which will speak the first message but then provide no further interaction on the call. This would allow me to easily create a "voicemail inbox" style assistant, leveraging VAPI's existing fantastic call recording and webhook event features. I am unable to get this to work properly now, as I have not found a way to restrict the assistant from interrupting the caller and speaking while the caller is leaving their voicemail. Thanks in advance for the consideration of my feature request!

Please add native support for Google’s Gemini Live API models (e.g., gemini-2.5-flash-native-audio-preview-09-2025 and gemini-live-2.5-flash-preview) so we can run real-time, low-latency, voice + text conversational agents in Vapi. Why this matters Google’s Live API enables low-latency, bi-directional streaming conversations (voice and text), with interruption/“barge-in”, function calling, and—on certain models—native audio behaviors designed for natural spoken dialogs. Bringing these models into Vapi would let us use Vapi’s telephony, tools, test suites, analytics, and guardrails with Google’s realtime stack.

The button "submit bug report leads to the page feature requests" https://docs.vapi.ai/support

Please, can we get Grok 4.1 Fast (both in reasoning & non reasoning), as well as the upcoming Grok 4.2 Fast. I assume it is as easy as adding a new endpoint to the UI and would be very useful to be up to date with all the main providers, not only OpenAI which seems to be slowing down in progress.

Can we get an option to turn on thinking mode in Gemini 2.5 Flash. I was informed by Vapi support this week that Gemini 2.5 does not use thinking mode currently and there is no way to activate it in the Vapi agent configuration. Lot of startups have Google Cloud credits and would prefer to use Gemini 2.5 Flash with thinking. Without thinking mode, I have seen some critical lapses, which I do not get in Google AI studio with thinking mode on.

I’m working on a client project and need some help setting up a proper warm transfer flow. What we’re trying to achieve: The assistant should introduce the caller first (warm handoff). Then it should attempt to transfer the call to the human agent. If the human agent does not answer, the assistant should stay on the line, return to the caller, and offer an alternative—such as booking an available time. The issue we’re facing: After the transfer is initiated, the call rings the human agent. When the human agent doesn’t pick up, instead of the assistant coming back in, the entire call ends. In one test, the tool even switched sipVerb back to "refer" on its own, even though we originally set it to "dial". This means the assistant is not staying on the call, and the caller gets hung up on—which breaks the workflow. Goal: We want the assistant to remain active during the transfer attempt and resume the conversation if the call to the human agent fails. Has anyone successfully implemented this warm transfer behavior? Is there a specific configuration or method to ensure the assistant stays in the call when the human agent doesn't answer? Any guidance or examples would be greatly appreciated. Thanks! If you want, I can also

Call: 019b0edc-fea5-7ff2-a144-23bde81ed2db Model: GPT Realtime Tool call finishes, but assistant only says: "I'm checking on that for you right now. Let's wait just a moment for the response." Very frustrating. { "id": "log-28", "level": 30, "time": 1765480937089, "timestamp": 1765480937089000000, "severityText": "INFO", "severityNumber": 9, "body": "Tool calls finished", "attributes": { "category": "tool", "callId": "019b0edc-fea5-7ff2-a144-23bde81ed2db", "orgId": "9fc10dfb-0f59-4ced-8157-345511dd806a", "runDurationMs": 5272, "totalResults": 1, "errorsCount": 0, "results": [ { "name": "get_response_2", "toolCallId": "call_Dujyv9wGmxRLsp1U", "type": "function", "serverUrl": " https://trustful-uncherishing-braiden.ngrok-free.dev ", "result": "Our dentures are made of high quality materials.", "metadata": {} } ] } }

As per this discussion https://discord.com/channels/1211482211119796234/1448919423204327545 , it seems like vapi doesn't include ordering on 'model-output' webhook message thus rendering it unusable for most applications. Please help to include ordering in the response too