Drybread Studios

I make products , tools , games and other things I want to exist . I started Drybread Studios because some ideas are better made than explained.

Some become things you can use. Others remain experiments. I write here sometimes, too.

Why Drybread? Most would call it toast Drybread leaning against a toaster. I beg to differ.

Keep scrolling ↓
Drybread hanging completely dry on a sunny clothesline
Drybread outsideStandard drying procedure no. 04

Current work

and .

ToolAvailable

Eido

I made Eido because ideas change shape faster than you can write a prompt. You arrange your notes on a canvas, and the lines you draw between them become the instructions for the AI.

Open Eido ↗
GameAvailable

Deepbite

A pixel sim digging game inspired by Motherload (2004) and Noita. Built from scratch with AI for my nephews to see if digging deep and balancing fuel risk is still as fun as it was back then.

Play Deepbite ↗
SkillAvailable

Explain Work Visually

An agent skill that turns hours of coding logs into a single readable page. It shows what changed, calls out the assumptions it made, and lets you highlight lines to send notes back.

Get the skill ↗
AppAvailable

Karapop

Turns four phones into live microphones connected to a single room speaker over local Wi-Fi. Built for house parties so everyone can sing without buying dedicated hardware.

Start Singing ↗
21 OCT · BLUE MOUNTAINSA colder path home
AppEarly beta

Juno

A field journal that knits together the thoughts you jot down, the places you visit, the people you’re with, the photos you take, and small things you notice along the way into your very own wiki.

Join the beta ↗
Eido

Drybread Studios / Eido

Drybread connected at both sides by green strings inside a miniature Eido Board.
Context assembly, 10:14 amCards / Connections / Context

Think it out on a board, not in a prompt box.

Category
Tool
Status
Available
Type
Web app

I made Eido because my thinking changes faster than I can write a clean prompt. It is a whiteboard for the messy part. Write fragments on cards, drag the related ones near each other, and keep moving them as the idea changes shape.

The lines you draw between cards are the instruction. A strong line says use this. A faint one says keep this in mind. A crossed one says I tried that, do not go back there. You never write a prompt. The board is the prompt.

So the AI carries on your thinking instead of replacing it. And when it gets something wrong, you can see which card sent it that way, and move the line.

Deepbite

Drybread Studios / Deepbite

Drybread standing beside a tracked drilling rig on a glowing seam, deep inside a mine.
Drybread sitting comfortably close to a lava pitDepth: 1251m

Dig down, make the trade-off, and bring enough fuel to climb back out.

Category
Game
Status
Available
Type
Mobile web game

I’ve been vibe coding games for my nephews lately, and when they visited recently I found myself thinking about the games I played as a kid. I remembered playing Motherload (2004)—digging down, finding gold, and having to make sure I didn't run out of fuel getting back. Exploring deeper while managing calculated risk stuck with me; not many games back then forced you to make constant trade-offs every time you wanted to push a little further.

Now that AI agents are decent at building games, I wanted to make my take on the genre using a pixel sim style inspired by Noita. Everything in Deepbite was built from scratch by AI: the game engine, systems, assets, animations, and music. Everything is procedurally generated, so it loads super quickly.

AI can nail a nice looking demo, but turning it into an actually fun game your nephews will play through is a whole other challenge.

Interactive core sample

The ground remembers.

Tap to inspect. Drag to drill. The tunnel stays.

Explain Work Visually

Drybread Studios / Explain Work Visually

Drybread in an armchair, reading a wall of finished work on a large screen.
Review, 9:40 pmRead it once. Answer on the page.

The work is done. Judging it is the hard part.

Category
Skill
Status
Available
Type
Agent skill

Explain Work Visually is a skill for AI agents. When a run finishes, instead of dumping raw transcripts and changed files on you, the agent writes a single structured page explaining what it did, the decisions it took without asking, and the questions it still has.

You review the page directly: highlight any sentence, drop a note or screenshot, and send the file back. The comments stay pinned to the exact words they refer to, so you don't have to spend time explaining what you meant.

The real thing

Made by the skill, about the skill.

Live page. Select any sentence in it to leave a note.

Read the example
KARAPOP

Drybread Studios / Karapop

Drybread standing under a spotlight with a microphone in a dim, crowded karaoke bar.
Karaoke bar, 11:46 pmConditions remain acceptable.

Four phones. One speaker. No turns.

Category
App
Status
Available
Type
Mobile web app

I wanted karaoke at a house party without buying a karaoke machine. Karapop turns four phones into four microphones. Tap Go live and you get pitch correction, backing voices, echo and reverb. Your voice crosses the house wifi to a laptop plugged into the room speaker. The phone itself stays silent, so nobody sings against an echo of themselves.

I thought the hard part would be audio quality. It was latency. Thirty milliseconds felt comfortable. At forty I could hear the delay. At fifty, singing became difficult. So Karapop drops late audio rather than deliver it politely, and four voices arrive together or not at all.

The host gets four faders, four mute buttons and one button that silences the room. I added the last one after testing with children, who do not take turns.

Drybread standing unimpressed inside a laundromat dryer
Drybread doing some laundry.Standard drying procedure no. 72

Studio policy no. 01

Useful is good. Strange is also good. The best things tend to be both.

Breadcrumbs

Written occasionally

No breadcrumbs yet.

Drybread sitting under a vintage salon hair dryer
Drybread in salonStandard drying procedure no. 31