Questions people ask.

Nothing to run it. There is no subscription and no per-minute fee from Auralis: you run it on your own computer. The code is MIT-licensed and shared on request.

Costs that can appear, and who charges them:

• A Groq key is optional. It makes replies faster. Groq sets its own limits and prices.
• Phone minutes are billed by your telephony provider (Exotel, in our tests), not by Auralis.
• The voice models keep their own licences, listed at the bottom of this page. One of them, DhVaani, needs its authors' written approval for commercial use.

On an Apple Silicon Mac, yes. After the one-time model download, speaking, cloning, listening, answering from your documents, watermarking and receipts all run on the Mac, with no internet and no keys.

The network is used only for things you switch on yourself:

• Groq, for faster listening and answers (your audio or text goes to Groq).
• A hosted Supabase setup, for sign-in and cloud storage.
• Exotel for phone calls, Google Calendar for meeting times, WhatsApp for voice notes.
• The first download of each model, from Hugging Face.

On Windows or Linux the agents need a Groq key or a local OpenAI-compatible server such as Ollama.

Not through Auralis. Before a voice is cloned, the speaker says a one-time sentence that contains a random code. Auralis checks the code was spoken, and checks the recording is the same person as the sample (in our test, the same person scored 0.88 and different people 0.14).

The consent recording is stored with the voice. One-click withdrawal deletes the voice and its recording, and the withdrawal is written to a signed public log. A clone made with some other tool is outside what Auralis can stop.

Every clip and agent reply gets an inaudible watermark, a signed receipt and an “AI-generated” label inside the WAV file. Anyone can drop a clip on the Verify page to check it, and the receipt signature can be checked offline with the public key.

The watermark was found in 70 of 70 test cases over MP3 at 32 kbps, a WhatsApp-style chain and 8 kHz phone audio. It is evidence, not proof: speed or pitch changes, heavy noise or re-encoding with an AI codec can remove it.

• Speech in 22 Indian languages, with Indic-Speak preset voices.
• Your own consented voice speaking Indian languages, with DhVaani-0.5 from ARTPARK-IISc. Hindi can also use Chatterbox.
• Cloning in ten more languages, English among them, with Qwen3-TTS.
• Agents understand and answer in English, Hindi and Hinglish, and follow the caller if they switch.

Live phone and agent turns use a preset voice, because cloned Indian-language speech is too slow on a laptop for a live call.

Our numbers come from an Apple M4 with 16 GB. With every model loaded Auralis uses 10 to 12 GB, so 16 GB works but swaps under heavy use. An 8 GB Mac and a Windows laptop with an RTX 3050 Ti have not been tested yet.

The Can you run it? page lets you pick your machine and see what works, what to switch off and how sure we are.

By default, no. In local mode your voices, documents and conversations stay in a folder on your own computer. This page has no trackers or analytics.

Data leaves your machine only when you switch a service on: Groq receives the audio or text of the turns it handles, Supabase stores what you put in a hosted setup, and Exotel carries phone calls.

Partly. An agent can place a call over Exotel from the Agents page. It speaks first to say it is an AI and stops when the person talks over it. One live call has been tested, on an Exotel trial account.

Receiving calls on your own number is set up but not yet tested. Only call people who agreed to be called.

The full list is in the project's status file (shared on request). The short version:

• WhatsApp voice notes and Google Sheets logging are built but tested only against fake services.
• Inbound phone calls and real Google Calendar booking are untested.
• Windows is untested. Hosted cloud mode is tested against fakes only.
• The Hindi watermark is weaker than English under heavy noise (common channels are fine).
• A 16 GB Mac swaps 3 to 5 GB with every model loaded.
• Groq's free tier runs out after a day of heavy use; replies then come from the slower local model.

Auralis builds in what India's rules on synthetic media and personal data are concerned with: spoken consent before cloning, a retention period, one-click withdrawal, an AI-generated label and watermark on everything it makes, and an AI disclosure at the start of every agent call.

That is a design choice, not a legal opinion. We are a student project, not lawyers, so this is not legal advice. What is lawful for your use, for example calling people, is for you to check.