Dictate into any field. Ask an agent that sees your screen, uses your tools, and draws its answers as live diagrams and videos. For macOS on Apple silicon.
Generative UI
Hold fn+ctrl and say “show me how a request reaches a Durable Object, and record it.” Nineteen seconds later the agent has built a live diagram on your desktop and saved a video of it.
01 · SCENES
The agent describes boxes, wires and steps. t2t’s scene kit turns that into a clean, animated diagram that lights up one step at a time.
02 · WINDOWS
Charts, tables, forms or 3D: the agent can write any HTML into a window. Move it, cover it, close it. It comes back after a restart.
03 · VIDEO
t2t renders one loop frame by frame, offscreen, to a crisp 30 fps mp4 in ~/Movies/t2t. Ready to drop into a doc or a slide.
The same scene kit draws every explainer, so what the agent makes looks like this: one font, one accent, steps that light up in order. No design tool, no screen recorder, no editing.
Hold fn in any editable field. T2T records while you hold the key, transcribes the audio locally with Whisper, and inserts the result into the field you were using.


Hold fn+ctrl and ask. The agent sees your screen, answers out loud as it thinks, draws boxes and arrows over your apps, and opens small windows when words are not enough.
t2t reads the MCP servers you already use from ~/.pi/agent/mcp.json. Turn one on and the agent can call its tools. Tools run on your Mac, and you can ask t2t to show Allow / Deny before each call.

With the hand MCP server on, the agent reads and changes Mac apps through Accessibility.
hand sets values directly. It does not move your mouse, type on your keyboard, or bring apps to the front.
A labelled cursor marks each control the agent touches. Buttons that look destructive are blocked unless you confirm.