Feature Requests

We have invested deeply into testing and making it work though logic/gating between steps vary a lot on AI model. Please do not deprecate this critical feature for Vapi.. We no longer recommend Workflows for new builds. Use Assistants for most cases or Squads for multi-assistant setups. See Assistants and Squads. Existing workflow content remains for reference.

During the vapi sip down time we faced bad reviews from our clients that thire customers is calling the ai but its keep ringing then i told them about the server down they said that when its down then why call was not forwarded to us so its a bad impression for us. Please make this feature as soon as possible

Enable the Google event scheduler integration to set the description, location, and video conferencing option of a calendar event

Setting voice.model to a Cartesia snapshot (e.g. sonic-3-2025-10-27 ) returns a 400 Bad Request error. Only base models like sonic-3 or sonic-2 are currently accepted. ( https://docs.vapi.ai/api-reference/assistants/create#request.body.voice.CartesiaVoice.model ) Please add support for snapshot identifiers. Using only base names can cause inconsistencies when the underlying model updates. Below a list of snapshots for some base model names taken from cartesia docs: sonic-3: sonic-3-2025-10-27 sonic-2: sonic-2-2025-06-11, sonic-2-2025-05-08, sonic-2-2025-04-16, sonic-2-2025-03-07 sonic-turbo: sonic-turbo-2025-03-07 sonic: sonic-2024-12-12, sonic-2024-10-19

Should be able to add and remove currency lines via API

Hi, I'd like to request the ability to configure call concurrency when setting up a campaign. In certain cases, I need to ensure that calls are executed sequentially rather than in parallel—especially when my clients’ telecommunications infrastructure cannot handle the concurrency limits defined by the plan. When this happens, we often encounter issues where calls are not properly connected to agents, resulting in muted calls—neither the agent nor the recipient can hear each other. Thanks in advance for your support. Best regards, Erich

I posted this in discord but never heard back. This is a platform bug making voicemail configuration not work properly. https://discord.com/channels/1211482211119796234/1432727023125532763/1432727023125532763 I'm using Vapi voicemail detection provider with the recommended settings https://docs.vapi.ai/calls/voicemail-detection#vapi-provider-recommended . The docs https://docs.vapi.ai/calls/voicemail-detection#important-beepmaxawaitseconds-configuration say that beepMaxAwaitSeconds is the maximum it will wait until reading the voicemail script. My interpretation of this is that it is meant as a stopgap for when we dont detect that its a voicemail using the provider. What I'm seeing in practice is that any value other than 0 makes this wait the whole wait period every time, even if a voicemail is detected in the logs. For support and example call is 019a28a3-d338-7aa8-a38c-eb55f0897e6b. At 5s the voicemail recording stops, at 6s google returns Customer detected as voicemail by google, then silence until 25s when the max wait seconds is reached and the voicemail script is read. In contrast, if I set the beepMaxAwaitSeconds value to 0 which the docs recommends against I see the correct behavior 019a2856-f3cc-722b-a8ab-4d887ab9b20d: at 7s google detects voicemail and it reads the script immediately.

Would be great if the Langfuse integration set the call id as session_id to group together traces from the same conversation - https://langfuse.com/docs/observability/features/sessions

I created a very basic Eval for a squad, when executed it shows running and the status in the API is "queued" but it never changes (65h 50m running for now) Example eval run: "id": "3f2b58a1-fb64-440b-89d7-5170056a83a1", "orgId": "3423c3fe-782c-4f7c-9f1f-6cd1a20d6c72"