AIVA for Voice

Talk to AIVA.It talks back.

AIVA listens and speaks — free with the Mac’s built-in voices, or richer with OpenAI or ElevenLabs. Five activation modes, natural interruption, and follow-up listening make it feel like a conversation, not a command line.

What AIVA can do

Concrete help with Voice.

Speak with the voice you choose

Free Apple voices need no key. Add OpenAI (Cloud or Local) or ElevenLabs (Local) for richer character — switch providers any time.

Start a conversation five ways

Wake word, push-to-talk, the composer mic button, automatic follow-up, and talk-over interruption. Mix them to fit your environment.

Interrupt naturally

Start talking while AIVA is speaking and it stops mid-sentence to listen — useful when it’s heading the wrong way or you’ve thought of more.

Stream for responsiveness

Speech can start playing while it’s still being generated, so replies feel fast and the back-and-forth stays natural.

Example prompts

Just say it in plain language.

Hey Claire — what’s my first meeting today?

(push-to-talk) Draft a reply to Sam and read it back to me.

Turn on wake word and follow-up listening.

Use the ElevenLabs voice from now on.

Stop — actually, make it more formal.

Switch to push-to-talk; I’m heading into a meeting.

Actions available

The tools AIVA actually has.

Speak

Apple, OpenAI, or ElevenLabs voices.

Listen

Wake word, push-to-talk, or mic button.

Follow up

Keep listening for a few seconds after speaking.

Interrupt

Talk over AIVA to redirect it mid-sentence.

Stream

Begin playback before audio finishes generating.

Permission model

You stay in control of every consequential action.

Read-only work happens silently. Anything that changes the world waits for your approval — and destructive actions always ask prominently.

Safe — does silently

  • Listen and respond by voice
  • Read drafts and summaries aloud

Asks first

  • Any action a voice request triggers (send, post, create) still gates that action

Always asks

  • Voice-triggered destructive actions still ask prominently

Common workflows

How people put it to work.

01

Always-on assistant

Wake word + follow-up + interruption turns AIVA into a hands-free companion you can talk to across the day.

02

Meeting-safe mode

Switch to push-to-talk in noisy rooms or meetings so AIVA only listens while you hold the key.

03

Voice and text together

Speak when convenient, type when you can’t — both work simultaneously, no mode switch required.

Without your approval, AIVA won’t

No surprises.

These never happen on their own. AIVA drafts and asks; you decide.

  • Send, post, or create anything triggered by a voice command — the action still asks
  • Take destructive actions by voice without a prominent confirmation

Local-first notes

Your data, on your machine.

Apple voices run entirely on-device and free; no cloud key required.

API keys for OpenAI/ElevenLabs are stored encrypted locally.

There’s no always-on recording to a server — listening modes are yours to enable and disable.

Related workflows

AIVA works better together.

Core feature

AIVA for Screen Awareness

Ask AIVA to look at your screen and it captures a single screenshot, reads it with a vision model, and helps — the error, the article, the PDF, the email you’re reading. Only on demand, only after you grant permission, never on a timer.

Core feature

AIVA for Memory

A normal chatbot forgets when you close the tab. AIVA keeps four kinds of memory — durable facts, workflow recipes, live context, and behavior rules — so it knows you, gets faster over time, and never makes you repeat yourself.

The app

AIVA for Mac

AIVA is a native Mac app, not a browser tab. It runs locally on 127.0.0.1, keeps its data on your machine, and stays on in the background — remembering, following through, and only surfacing what’s genuinely useful.

Get started

Let AIVA handle Voice for you.

Create your account on the web — it carries straight into the Mac app.