The web app is built. Here’s what connects real voices, calls, and payments.
Studio, microphone recording, browser rehearsal, saved invitations, mobile PWA
Clone creation and browser voice tests use Cartesia. Phone calls and checkout remain disabled until their connections are verified.
Not connected
01 / GPT-Live 1 · ready-made voices
OpenAI
Marin is the generic voice option: no voice recording required. Add a project API key privately, verify access to gpt-live-1 and the gpt-5.6-luna backend, then enable the connection after a live test. The key stays on the server. Phone calling requires its own integration.
Sonic 3.6 handles cloning and preview speech. Managed Agents handle listening, replies and interruptions in the browser voice test. Record 30–60 seconds, create your clone, then talk to it. Phone calling is a separate connection.
Create a Verify service for phone ownership checks. Users also complete Twilio’s caller-ID verification call. The existing ElevenLabs phone adapter imports that verified number. Cartesia and GPT-Live phone routing still need integration and a real call test. Arbitrary caller-ID spoofing and anonymous calling are not offered.
Encore ($9 for 10 minutes) and The whole family ($19 for 30 minutes) are created as one-time USD products under Face Search AI. Their prices are saved in hosting. Only a paid, signature-verified webhook adds minutes. Checkout is disabled until the operator enables it.
This release is an installable web app for iPhone and Android. Use the studio’s Install app button for instructions. There is no native store binary yet.
Apple’s guideline 1.1.6 expressly rejects apps enabling prank or anonymous phone calls. A native release needs a genuinely opt-in comedy or education product and App Review, not simply a new label.
Configure secrets in hosting, never in this page or a public repository.
Verify recipient sign-in and webhook delivery on prenk.org before enabling calls or payments.
APP_ORIGIN is https://prenk.org. Configure Stripe’s signed webhook at https://prenk.org/api/billing/webhook.
Verify a real own-voice clone, caller ID, recipient opt-in, stop behavior, and one-minute limit on controlled test numbers.
Enable CALLING_ENABLED and BILLING_ENABLED only after those checks. Payments still require Stripe credentials and verified callback delivery.
Open-source foundations considered
The existing phone adapter uses ElevenLabs’ native Twilio integration to avoid operating an audio server. The React/Vinext app remains your own source. For a self-hosted audio backend, the strongest alternatives are:
LiveKit Agents — Apache-2.0 orchestration, WebRTC, SIP; requires a separate agent process.
Pipecat — BSD-2-Clause audio pipelines and phone transports.
Chatterbox and Qwen3-TTS — MIT / Apache-2.0 cloning options, with GPU hosting to evaluate.
Capacitor — MIT native client packaging after a store-eligible product is established.
F5-TTS’s official pretrained weights are noncommercial, so they were excluded from the paid-product foundation.
The joke has a boundary.
Clone only your own voice. Recipients accept a specific invitation, verify the number they control, and request their own call. Calls open with an AI disclosure and processing notice. We don’t offer urgent money requests, emergency scenarios, credential collection, threats, or official impersonation.
Raw voice samples stay in the browser until you submit them to Cartesia. The app stores your provider voice ID, consent timestamp, verified phone, sessions, and credits. Removing your clone deletes it from the provider through the app. The app does not save browser-test audio; Cartesia account retention settings apply. The legacy phone adapter disables audio saving; transcript deletion is configured after one day. No call recording is published or shared.