FiskarAI builds what you ask for. Everything else is on you.
The same problems come up again and again when you build apps with an AI coding agent. Every part of Fiskar exists to remove one of them.
You forget things
Sign-in, empty screens, error messages and offline only come up after testers complain. The Don't forget checklist puts them in the plan before Claude starts.
You can't see or steer it
See every screen within minutes. Tap what's wrong and Claude gets the exact file and line. Approve changes with one tap from a notification.
Your usage runs out
Free tools do the routine work, small tasks go to the cheapest model that works, and a receipt on every step shows what you used and saved.
It isn't yours
It runs on your own Mac with the Claude plan you already have. Code goes to your private GitHub. No credits, no lock-in, no extra bill.
Your phone steers. Your Mac does the work.
Fiskar is the remote for a coding agent that runs at home. The phone talks straight to your Mac: on the same Wi-Fi over HTTPS locked to your Mac's own certificate, and from anywhere else over Tailscale. Nothing passes through a server of ours.
Fiskar
Describe, approve, point at what's wrong, check and install. Works anywhere.
Tailscale
The bridge
Fiskar for Mac sets it up with buttons. It runs Claude Code, draws the screens, builds and runs checks with free tools.
account
Your repo
Every change is saved. Every build is a release you can install or share.
From an idea to an installed app, without the computer.
Five tabs: Home, Design, Screens, Checks and Usage. These are real screens from the current app, showing the sample bakery app "Crumb & Co". Switch your device to dark mode to see them dark.
Set up your Mac with buttons.
No Terminal. Fiskar for Mac installs Claude Code, Java, the Android tools and GitHub CLI into its own folder, signs you in to Claude and GitHub, keeps the Mac awake while it works, and shows a code your phone scans to pair.
Describe it. Then don't forget.
Answer a few questions by tapping. Fiskar suggests what apps like yours need: sign-in, empty screens, errors, offline, account deletion. Switch on what matters and each one goes in the plan with a check.
Mark features Must, Nice or Later. Claude builds the Musts first.
Watch it work. Approve with a tap.
See what Claude is doing as it happens. Safe actions are pre-approved on your Mac; only the rest reach you as Yes / No cards, in the app or right in a notification. Switch on Auto-accept edits and changes to the app's files go through by themselves, while commands still ask.
Lose signal mid-session? Nothing is lost. Messages wait and send when you're back.
Choose the look for free.
Pick a mood and a main colour. The palette, light and dark, is worked out on the phone. Change fonts, corners, shadow and spacing. Save, and your Mac writes the app's theme and redraws every screen, with no AI usage at all.
Point at what's wrong.
Every screen appears as a preview minutes after Claude starts. Tap the button that looks off, choose Shadow, Size, Colour or Text, and say what's wrong. Claude gets the exact file:line, not "the button on the second screen".
Collect several notes and send them as one. Then compare before and after, and say ✓ Good or ✗ Not yet.
Check it, then install it.
Each feature's "done when" line becomes a check in plain words, with a screenshot of what it saw. When they pass, tap Build new, then Install. The app lands on your phone and the release is saved on GitHub.
Know where your limit goes.
Rings for the 5-hour window and the week, cost per app and per feature, and what Fiskar saved you. Auto picks the cheapest model that has worked for that kind of job and moves up a model after a ✗ Not yet. When your limit is full, new work waits and starts by itself after the reset.
Small, precise tasks make cheap models enough.
Tested on Claude Code with real tasks. Fiskar makes each job small, so a cheaper model gets it right. The two savings multiply.
file:lineWant to try it?
Fiskar is free and in early beta. Demo mode shows every screen before your Mac is set up.