About Ditto AI
Ditto AI is a Mac app that builds an AI agent from one recording of you doing a task. It's built by Amrit Preet Singh in Berlin, Germany, around one idea: you shouldn't have to describe a task you already know how to do. Show it once, read what the agent will do, and hand it off.
Why it exists
Most of the work people hand to an AI agent is work they already do by hand, every week, across the same two or three apps. Writing that down as a prompt is slow, and it's the easy part. The hard part is everything around it: finding a tool for each app, getting a credential into each one, and deciding what the agent may touch.
Ditto AI does that part from a recording. You do the task once. It writes down the steps, finds and checks a real tool for each one, and sets the permissions from a fixed table. You read all of it before the agent runs.
The rules it's built on
These hold throughout the app, and the security page explains each one:
- Your recording stays on your Mac. Only the frames, clicks and transcript of a recording go to the AI provider you choose, so it can read the task.
- The model doesn't set permissions. Access comes from a table, and anything unclassifiable is treated as a write.
- Checking a tool never changes anything. One real, read-only call.
- It doesn't make up integrations. Tool servers come from the MCP registry and npm.
- Every step shows its evidence, or it's dropped.
Who it's for
People who repeat the same multi-app task every week: developers who already use Claude Code or Kiro and would rather not hand-write another skill and its MCP config, and anyone in operations or finance whose week includes the same invoice, report or tracker update.
Who's behind it
Ditto AI is built and run by Amrit Preet Singh, in Berlin. It's early: the app is at version 0.1.0, and what changed recently is on the changelog.
Get in touch
Email [email protected]. Bug reports, questions and descriptions of tasks it got wrong are all welcome. Postal details are on the legal notice.