Turns your thoughts into projects and tasks, hands them to your own AI agent — and brings the finished work back to you for review.
Works with Claude Code, Cursor, Codex — any agent that speaks HTTP.
Jot your thoughts in the notepad. Split them into projects and tasks — right here, in chat with an AI. Copy a link and hand it to your agent: it reads the tasks, does the work, and asks when something is unclear. You review and decide.
That is the real app playing next to it, not a mockup — click a step to jump.
Ideas about a project rarely arrive at your desk. You jot one down in notes or message it to yourself — and never open it again. A month later you do not even remember having it.
Not in a separate app you gradually stop opening. You pass them every time you come to the board — and any one of them can be picked up and thought through with an AI months later.
Give an AI ten items at once and by the middle of the list it repeats something already done, skips something else, or forgets where it was.
Tasks are deliberately small enough to fit whole — one title, one description, its own history. The agent never holds ten things in its head at once, only the one in front of it.
An AI reports back in a chat that it finished. To find out what it actually did, you scroll back through the whole conversation.
The agent moves the card into "In review" itself. Open the column — see exactly what is waiting for you, not what is still in progress.
You talk an idea through with an AI and end up with a real plan. A week later the conversation is unfindable — and even once you find it, the plan still has to be retyped into tasks by hand.
Give the agent a boot link once. It reads the project, discusses it with full context, and puts the outcome on the board itself — milestones on the roadmap, steps as tasks.
A project sits untouched for a month. You return and spend half your time reconstructing what was done, what was decided, and why it was decided that way.
Decisions live in layers — from why the project exists down to specific parameters. Every task carries its own history and last action. An agent restores the whole context from a single boot request instead of ten clarifying questions.
You run several projects and several AI agents. Keeping every project open in a tab is physically impossible, and an agent has no way to know which one it is working in right now.
Each AI role gets its own token. Pin a role to one project and it cannot reach the others; give it access to all of them and it names the project in every request. You see the results of all of them without switching tabs.
The agent worked for an hour. What it changed, what it deleted, and why it decided that way — you find out by reading a diff.
A person, or a specific AI role — every action carries a name. Progress comments render as headings, lists, and code, not a wall of text. Deleted tasks go to the trash instead of disappearing. And when a task is ambiguous the agent must ask rather than guess, handing the turn back to you so the question is not lost.
Because the project lives here and not in a chat, any agent sees its actual state at any moment. One boot link explains what the project is, what has been decided, what the standards are, and which API calls are available — no separate onboarding document to keep in sync by hand.
GET /api/boot?token=synaban_...
SYNABAN — BOOT-ИНСТРУКЦИЯ ДЛЯ ИИ. ПРОЕКТ: Randomizer
----------------------------------------------------------------
Ты подключён как ИИ-роль «Backend Dev AI»...
Слой 1 — Видение (зачем проект нужен)
Слой 2 — Архитектура (карта системы)
Слой 3 — Контракты (интерфейсы)
...
GET /api/tasks?idea_id=&column=&priority=
POST /api/tasks {idea_id, column_id, title, ...}
POST /api/tasks/{id}/comments {body}