# Quickstart

Generate your first diagram with AI, then make it yours. About five minutes.

[Sign up](https://app.eraser.io/auth/sign-up?ref=docs) if you have not already,
and create a file from the [dashboard](https://app.eraser.io/dashboard/all)
(`^ N`).

## 1. Generate a diagram from a prompt

In the canvas, press `/` to open the insert menu, choose **Diagram as Code** then
**AI Diagram** – or press `⌘J` / `Ctrl+J` to go straight there.

Describe the system in plain language. Something like:

```
A web app behind an API gateway, with a Lambda writing to a Postgres
database and publishing events to a queue that three workers consume.
```

Click **Generate**. You do not have to pick a diagram type – Eraser chooses the
one that fits, though you can override it. More in [AI diagrams](/ai-diagrams).

## 2. Refine it

The output is [diagram-as-code](/diagram-as-code), so you can edit the text
directly and the canvas updates as you type. Add a node, rename one, or attach an
[icon](/icons):

```
Analytics [icon: aws-redshift]
```

You can also keep prompting to change the diagram, or drag things on the canvas –
layout stays clean either way.

## 3. Start from something real

Rather than a prompt, point Eraser at what you already have:

- an [existing image](/image-import) of a diagram, including a photo of a
  whiteboard
- a [file](/file-to-diagram) such as Terraform or a database schema
- a [Git repository](/connecting-git-repositories), to generate
  [codebase diagrams](/codebase-diagrams) that [Eraserbot](/eraserbot) keeps
  current as the code changes

## 4. Make output consistent

Generic diagrams are the usual problem with AI. [Templates](/templates),
[references](/references), [rules](/rules) and [custom styles](/custom-styles)
ground generation in your own architecture and conventions.

## Next

Connect the [MCP server](/mcp) so coding agents can create and update diagrams
directly, or read [what is Eraser?](/what-is-eraser) for the wider picture.