Mushin Docs / Building Conversationally

Building Conversationally

The quickest way to build an application in Workbench is to describe it in a chat. In a conversational project you work with a Copilot — an AI assistant that turns your requests into your application’s design and code. This page walks through that flow.

Working with the Copilot

Open a design stream in your conversational project and you get a Copilot chat. You describe what you want in plain language — “an app to track tasks, with a title, a due date, and a done flag” — and send it. The Copilot responds with the design or code for what you asked, as the output of that exchange.

You build the application up one request at a time:

  • Ask for the data you want to store.
  • Ask for the pages and behavior you want.
  • Refine anything that is not quite right by asking for the change.

Each exchange is kept, so the conversation is a record of how the application came to be.

Guiding the Copilot with memory

As the conversation grows, you can pin a good result so the Copilot keeps it in mind on later requests — carrying forward the shape you have settled on rather than drifting. This keeps a longer session coherent, so later requests build on the design you have already approved.

From conversation to a running application

When the design is where you want it, you generate the application from it (see Code Generation & Templates ). Generation produces a complete, working codebase — a Go backend and a TypeScript/React frontend — that you can build and run. The conversational flow is designed to take you from an idea to a running application quickly, which makes it the best place to start and to prototype.

For building something larger and more deliberate — where you want to review every change, branch, and keep an auditable specification — see The Neuro-Symbolic Approach .

← PreviousProjects