Fractask
Vol. I — the Fractask method

Fractask

(fractal + task)

Get shit done with your AI partner.
Five rules. One open-source tool.

MIT licensedNode 20+CLI · Web · MCPBuilt in the open
§ 01 · The concept

What is
Fractask?

Fractask is a way to get things done with your AI — built for the hybrid world where humans and bots work side by side, not one behind the other.

Start with a goal. Anything you actually want to do — plan a wedding, ship a website, hire your first employee, move countries.

You and your AI break the goal into smaller parts. Then break those into smaller parts. Keep going until each part is something one person can finish in one sitting. Three at a time keeps it readable.

Anyone in the duo picks a part and does it. Human or bot. Notes, files, and links live on the part itself, so the next session — human or agent — picks up cold.

Together, you build something real.

§ 02 · What is a fractal?

Same shape,
every scale.

A fractal is a pattern that repeats at every level. Zoom in on a coastline and you see smaller coastlines. Zoom in on a tree and you see smaller branching trees. Zoom in on a snowflake and you see smaller snowflakes.

The same rule, applied again and again, builds infinite complexity out of one simple instruction.

Work breaks down the same way. A goal contains projects. A project contains tasks. A task contains sub-tasks. The same rule — break by ~3, until every part is actionable — applies at every depth.

That’s why it’s called Fractask.

§ 03 · The man who saw it

Benoît
Mandelbrot.

1924 — 2010

The Polish-French-American mathematician who coined the word fractal in 1975. He saw self-similarity everywhere — in cotton prices, mountain ridges, coastlines, lightning bolts, the human lung. Where most mathematicians saw mess, he saw patterns nested inside patterns.

His most famous discovery — the Mandelbrot Set — is the canonical fractal. Zoom in forever and you keep finding the same intricate detail. One equation, infinite depth.

“Clouds are not spheres, mountains are not cones, coastlines are not circles, and bark is not smooth, nor does lightning travel in a straight line.”
— The Fractal Geometry of Nature, 1982

He gave us the language to see the world as it actually is: recursive, repeating, nested. Fractask carries that vision into the work you do every day.

Inside the app

A whole tree. One screen. Either of you.

Fractask task tree view — sidebar with workspaces, central nested checklist, AI agent panel on the right
Goals on the left. Tasks and subtasks in the middle. Your AI partner on the right. Edits land everywhere instantly.
Fractask focus mode — single task with notes editor, subtasks, agent conversation
Drill into one task. Notes, files, and subtasks all stick to the part. Hand it to your AI, claim it back, watch it ship.
§ 04 · A day in Fractask

From a goal in your head to a thing that exists.

  1. 01

    Drop in your goal.

    One sentence. Anything. Fractask doesn’t care if it’s big or small.

  2. 02

    Break it down — together.

    Your AI proposes parts; you accept, edit, or rewrite. Three branches, three sub-branches, as deep as you need.

  3. 03

    Pick what you can do.

    The bot picks the boring parts. You pick the hard ones. Or vice-versa. Either of you can claim any part.

  4. 04

    Notes stick to the parts.

    A doc, a Loom, a Slack thread, a screenshot — all live on the part. Anyone opening it later starts informed.

  5. 05

    Done is done.

    Mark it done. Move on. The tree shrinks. The thing exists.

§ 05 · The five rules

The whole method, five lines long.

  1. Rule 01

    Start with a goal.

    One root. The thing you actually want. Everything grows downward from here.

  2. Rule 02

    Break it down as a duo.

    Decomposition is a conversation, not a monologue. You and your AI split it together — until every part is something you can actually do.

  3. Rule 03

    Chunk by three.

    Aim for ~3 sub-parts at each level. Three is what a brain can hold. Go as deep as the work needs.

  4. Rule 04

    Anyone in the duo picks a part.

    Human or bot. Pull the part you can do; let the other one pull what they can. Pulled work has an owner.

  5. Rule 05

    Everything lives on the part.

    Notes, files, links — they belong on the task itself. The next session picks up cold.

Review is a tool, not a rule. Use it when a part is risky or you want a second pair of eyes. Most parts ship without it.

§ 06 · The tool

Fractask is open-source.

MIT-licensed. Runs on your laptop or any server. Works with Claude Code, Cursor, Claude Desktop, and any MCP-compatible agent. CLI, web, and MCP all share one task tree.

  • · No accounts to start. Just clone, install, run.
  • · Your tree lives in a SQLite file you own.
  • · Sync across machines via Turso when you want.
  • · Six MCP tools — your agent does the rest.
fractask · cli
~/projects/launch  pnpm install && pnpm --filter fractask/cli build
✓ ready
~/projects/launch  fractask add "ship the new homepage"
+ created  9rEH4P  ship the new homepage
~/projects/launch  fractask ls --tree
○ 9rEH4P  ship the new homepage
  ○ 7Lm0Wq  hero copy
  ○ 8oNxRb  3D animation
  ○ 6PqdSe  open-source feel

Or run it as an MCP server. Your agent calls create_task; the result lands in the same tree your CLI sees.

Fractask — On the way, Vol. I
§ 07 · The book

On the way.

The Fractask method in 30 pages. Where the idea came from (Kasparov, 1998 · Miller, 1956). What GTD got right. What it missed in the age of AI. A worked walkthrough. A short call to action.

§ 09 · Hosted Fractask — for teams

Run Fractask for your whole team.

Everything in the OSS, plus the things only a team needs. We onboard each team by hand.

Team trees

Share goals across the org. Owners, members, and bots all see the same tree.

Slack & WhatsApp

Ping a part to chat and it lands as a block. Reply to update. The thread is the trail.

Scheduled agents

Run an agent every morning. It picks up the parts it can do and reports back.

Hosted, signed in

Google login. Daily backups. We host it; you stay out of the way of the work.

Book a demo →

Early access. We onboard each team by hand.