PantryAtlas Free · Open source · Runs on a small computer in your kitchen

Your kitchen's smart pantry.
Lives in your kitchen.

A small computer you set up once. It helps you cook delicious food from what's already on your shelf — whether you're feeding four people or four hundred. Free, open source, and nothing leaves your kitchen.

Built with Gemma · Google's open model, running on-device
How to set it up See it on GitHub →

It's a computer the size of a deck of cards that you keep in your kitchen.

Plug it into a power outlet and your Wi-Fi. From any phone or laptop on your home network, open a browser and go to pantryatlas.local — that's the address of the little computer. Type what's in your pantry, or snap a photo of your shelf with your phone's camera. PantryAtlas reads what you have and shows you real recipes you can cook with it, ranked by what uses the most of what you've already got. The Gemma 4 model on the device combined with the Epicure flavor-pairing research makes the matches genuinely delicious — not just technically possible.

Two starting presets — and everything in between

Most kitchens land somewhere between "family dinner" and "feeding hundreds." Pick the preset closest to yours; PantryAtlas adapts ranking under the hood and you can switch anytime.

Home Kitchen

4 to 10 servings

A weekday family meal. You open the fridge, see what's there, and PantryAtlas finds five recipes that use most of it without making you go shopping. Add "expiring soon" tags so you cook the kale before it wilts.

Community Kitchen

50 to 500 servings

You run a soup kitchen, food pantry, or shelter meal program. A donation truck dropped off mixed produce, dry goods, and proteins. PantryAtlas ranks recipes that scale linearly, use bulk-friendly ingredients, and need minimum specialized equipment.

You pick once on first launch. You can switch anytime from the top bar. School cafeterias, refugee kitchens, monasteries, mutual-aid coalitions, a college house — all of these are "in between." Pick whichever preset feels closer.

Set it up in four steps

About 30 minutes the first time. Then it just sits there working.

  1. STEP 1

    Pick a small computer

    PantryAtlas runs on two kinds of small computer. Either is a one-time purchase of around $120–$200 — you keep the hardware forever.

    Option A — Raspberry Pi 5 (8GB) default

    The friendlier choice. A Pi 5 kit from Amazon, Adafruit, or CanaKit includes everything you need: the computer, microSD card, USB-C power supply, and a small case. Around $120. Fast enough for everything PantryAtlas does.

    See a recommended Pi 5 kit →

    Option B — Coral Dev Board alternative

    Google's small computer with an on-board Edge TPU chip. Same software, same setup, around $130. PantryAtlas doesn't yet use the Edge TPU directly, so a Pi 5 is the faster default today. Choose Coral if you already own one, prefer Google hardware, or want to help us port the vision model to the Edge TPU.

    See the Coral Dev Board →

    Both boards run the same PantryAtlas image. Your data lives on the microSD card, so you can move between boards anytime.

  2. STEP 2

    Copy PantryAtlas onto the microSD card

    Download our pre-built image from GitHub. Use the free Raspberry Pi Imager app on your computer to write the image to the card. It takes about 8 minutes.

    Download image Get Raspberry Pi Imager

    Already comfortable with Linux? You can also clone the repo and run our installer script — see the GitHub README.

  3. STEP 3

    Plug it into power and Wi-Fi

    Put the microSD card into your Pi 5 (or Coral Dev Board), plug in the power supply, and either connect an ethernet cable to your router OR follow the on-first-boot Wi-Fi prompt. After about a minute, a small light on the board turns solid green — that means PantryAtlas is running.

  4. STEP 4

    Open pantryatlas.local in any browser

    On your phone, tablet, or laptop — anything on the same Wi-Fi — open a web browser and go to pantryatlas.local. The first time, you'll pick Home Kitchen or Community Kitchen. Then you're done. Type ingredients, snap a photo of your shelf, get recipes.

    On iPhone: tap Share → Add to Home Screen. On Android: tap menu → Install app. It feels like a real app.

Stuck? Open a GitHub issue or read the docs.

What it does

Snap a photo, or just type

Type ingredients one by one, or point your phone's camera at your shelf and let Gemma 4 read it. Gemma is a small but capable AI model from Google that runs entirely on your kitchen computer — no photo ever leaves the device.

Delicious matches, not just possible ones

Every recipe shows a coverage ring — 8/10 means you have 8 of the 10 ingredients. PantryAtlas combines coverage with the Epicure flavor-pairing research so the matches taste good, not just technically work. Expiring ingredients get a small boost so nothing goes to waste.

From a family of four to four hundred

In Community Kitchen mode, PantryAtlas favors recipes that scale linearly to hundreds of servings, use bulk-friendly ingredients, and need minimum specialized equipment. School cafeterias, refugee kitchens, mutual-aid coalitions — everyone in between is welcome.

Yours, not ours

PantryAtlas is free software. It runs entirely on the small computer sitting in your kitchen. There is no cloud account. We don't see your pantry. We don't track what you cook. We don't sell anything to anyone. The code is Apache 2.0 — read it, fork it, audit it, run it. Recipe data comes from the public RecipeNLG corpus (CC-BY-NC-4.0) curated by Bień et al. Ingredient and flavor data comes from FlavorDB (CC-BY-NC-3.0) curated by Garg et al. Built for kitchens everywhere, by people who think food access is a human concern, not a market.

Ready to set up your kitchen?

DIY in about 30 minutes

Buy a Pi 5 kit (or Coral Dev Board), download our pre-built image from GitHub, flash the microSD card with the free Raspberry Pi Imager, plug it in. The four steps above walk you through every part. You don't need to know Linux or use a terminal — just a web browser.

Download image Setup guide ↑

Clone the repo

Comfortable with Linux and git? Clone the repo, run the installer script, or contribute. Apache 2.0. Pull requests welcome. The project is small, the codebase is reasonable, the issues are friendly.

GitHub →