A voice that listens. Properly.

The pitch

Tap the orb. Hold it for a beat. The orb glows.

Now you're in a conversation. You speak. It listens. It speaks back. Naturally, at human pace, without the awkward "processing" beat between turns.

While you talk, Amanu is still capturing in the background. Every todo, every event, every person, every mood, pulled out and filed. The conversation isn't a tangent from your capture. It is the capture.

Why this is different

It isn't Siri. Siri is a one-shot command line. You ask, it answers, it forgets. Talk-back holds the thread. You can change your mind mid-sentence. You can say "no, the other Tuesday". You can ramble.

It isn't ChatGPT voice. ChatGPT voice is a chatbot with a microphone. The session is a chat session. It doesn't know your calendar, your reminders, the people in your life, what you said last week. Talk-back does, because Amanu does.

It isn't a meeting bot. Otter and Granola are built for rooms with other people in them. Talk-back is built for the meeting with yourself.

Built on real voice infrastructure

Talk-back runs on OpenAI's Realtime API for the language layer, and our own self-hosted LiveKit cluster for the audio layer. That means the audio leg of your call doesn't fan out to three vendors you've never heard of. The pipeline is short, and we own most of it.

When the conversation ends, the transcript and the items it produced live in your private account. The audio doesn't get used for training. Not by us, not by OpenAI.

What people use it for

A walk after a hard day, talking through what actually happened.

The drive home from a client, dumping every action item before they evaporate.

The five minutes after a difficult conversation, untangling what you actually feel about it.

A morning planning session, talking through the week before you've opened a single app.

It isn't a friend. It doesn't pretend to be. It's a scribe that happens to listen well.