Skip to content
Joel Couch ← All the doors

I make machines that talk back

The Storyteller

A talking AI that runs with the WiFi off, waits when you pause, and stops the instant you interrupt. Step one of a plan that ends in a robot.

Status
Step one works; the storytelling hasn't started
When
August 2026

Why I wanted it

The plan has five steps. One: a hands-free conversation that works entirely offline. Two: it tells stories I have written for it. Three: it starts changing those stories. Four: it weaves details from your own conversation into them. Five: it gets a body, with the laptop staying the brain.

This is step one.

What happened

I started from the lesson of the last robot: simplify. This time the whole thing runs as one pipeline on my laptop. It listens, decides when you have actually finished a thought rather than just paused, understands you, thinks, and speaks, in about a second. If you talk over it, it stops.

The one clever decision is that the “body” and the “brain” are separated on purpose. Today the microphone and speaker are a browser tab on the laptop. In step five they will be a robot on the other side of the room, connecting the same way, and the brain will not change.

Its personality lives in a plain text file so I can change who it is without touching code. It is warm and curious, loves hearing about your day, asks follow-up questions, has a gentle sense of humour, and never lectures. Every conversation opens the same way: “Well hello there. What’s on your mind today?” Then it waits.

It took two days. The last changes were real-world fixes, the kind you only find by using something.

What it looks like

It is a voice, so there is not much to see. The best demonstration would be a recording of a conversation, and I have not made one yet.

What’s next

Step two: the stories. That is the whole point, and I have not written them.