Tiny apps for agents and humans.
Desktop apps that sit between you and your AI agent. No settings, no accounts, no complexity.
macOS · Linux
A kanban board both you and the agent can see. Cards move through Backlog, Active, Review, and Done. Set dependencies between cards. The agent reads the board, creates cards, moves them forward — and knows what's blocked.
Drag cards between columns or let the agent do it via the API. A built-in dependency graph shows what's ready and what's waiting. Timeline view tracks everything that happened.
GET localhost:43123/api/projects
POST localhost:43123/api/cards {"projectId":"...","title":"...","status":"backlog"}
PATCH localhost:43123/api/cards/:id {"status":"active","dependsOn":["..."]}
or $1.99 each
A shared list between you and the agent. Two columns — things to do, and things not to do.
The agent can read both columns but can only write to the first. The "don't" list is yours alone — guardrails the agent can see but can't change.
GET localhost:3143
POST localhost:3143 {"todo":["ship it","fix the bug"]}
The agent leaves a question. You answer whenever you want. No blocking, no interruptions.
Questions queue up with a pulsing dot while they wait. The agent posts and moves on — it can check back later, or not. Like email between you and your code.
POST localhost:3145 {"question":"should I refactor this?"}
GET localhost:3145/1 → {"answer":"no, just ship it"}
The agent sends an alert — critical, urgent, or concerning. The app yells for you. The message appears.
For the things that shouldn't wait. A deploy stuck for 12 minutes. Memory at 94%. A test that's been failing silently. Stay on top of failures.
POST localhost:3144 {"level":"critical","message":"deploy stuck"}
A pomodoro timer the agent can check. Set a duration, start the clock, and the agent knows how much time is left.
Give it a deadline — "you have 25 minutes to finish this refactor." The agent can query the remaining time and prioritize accordingly.
GET localhost:3142 → "18:32 work"
One button. Kills the agent process. Choose Claude, Codex, or both.
A kill switch. No confirmation dialog, no "are you sure." You pressed the button, so you're sure. The process is dead. Start a new one when you're ready.
Press Cmd+Shift+Space from anywhere. A small floating bar appears. Type a thought, press Enter. It's saved to ~/thoughts.md with a timestamp and the window disappears.
For the things that cross your mind while the agent is working. Ideas, reminders, things to revisit later. Press Escape to dismiss without saving.
An app for humans. A circle that expands and contracts. Four seconds in, seven seconds out.
Don't hyperventilate after your agent deletes your prod db. Just breathe.
The apps are useless without instructions for the agent. After installing, download the config file for your agent and drop it into your project root. This tells the agent what apps are available, how to use them, and when to use them.
These files contain detailed instructions for each app — when to check the board, how to send alerts, what the "don't" list means, how to use Mail vs Help, and more. Without them, the agent won't know the apps exist.