Potato

Local · Open-Source · Zero Cost

🥔 Potato

File

Edit

Tools

View

Layout

Shapes

AWS

Azure

GCP

Start your first diagram

Three ways in. Pick any one.

Connect nodes by dragging from the dots that appear on their edges. Select a node to edit its label, theme, and description.

Properties

🖱️

Click a node or arrow to edit its properties

Double-click text to edit inline

Ready

0 nodes

0 connections

x:0 y:0

Ctrl+S Save · Del Delete · Double-click Edit · Drag Connect

✏️ Edit Label

📋 Duplicate

⬆️ Bring to Front

⬇️ Send to Back

📄 Copy

📌 Paste

🗑️ Delete

Help

🕘 Recent diagrams

Click to reopen. Last 8 saves are kept in this browser.

✨ Start from a template

Pick a starting point. Everything is editable after.

💰 Cost Estimator

Rough monthly USD estimate based on the services in this diagram. Pick a workload size to see a baseline number.

Workload:

Estimated monthly total

—

⚠️ Estimates are reference-grade, not authoritative. Excludes data egress, support plans, NAT-gateway processing, and many add-ons. Always verify with the official AWS / Azure / GCP pricing calculators before committing.

📝 Play Sequence Editor

Set the order, narration, and which connections appear in the Play animation. The LLM seeds this — edit freely.

🤖 AI Import — generate a diagram with any LLM

Three steps. Works with ChatGPT, Claude, Gemini, Mistral, Llama — any model that returns HTML.

1Copy

→

2Paste into ChatGPT / Claude / Gemini, describe your architecture

📚 Example prompts & output format details

"Draw an AWS serverless API: API Gateway → 3 Lambdas → DynamoDB + SQS"
"Show a RAG chatbot: S3 docs → embedding pipeline → vector DB → Bedrock"
"Diagram a Kubernetes app with ingress, 2 deployments, Redis + Postgres"

The LLM replies with a Potato-format HTML block (a <script type="application/json" id="potato-data">). To keep it for later, save the reply as my-diagram.potato.html and use File → Open.

3 Paste the LLM's full Potato-format HTML output here, then Import

The whole reply — from <!DOCTYPE html> through </html>. Raw JSON works too if the LLM skipped the wrapper.