What is Mutsumi?

You don’t need focus. You need to never lose a thread.

“Focus” is last era’s religion.

Today’s developer switches between a dozen contexts every day — writing code, reviewing PRs, answering messages, running three agents, scanning feeds, writing reports. This isn’t a deficiency. This is how you work. You are multi-threaded.

The real problem was never “you do too many things at once.”

The real problem is: the moment you switch away, the previous thread starts fading from your brain.

An hour ago you had a critical bug to fix. Three context switches later, it’s gone. Not because you lack discipline — because human working memory holds about 4 items, and you have 12 threads open.

What Mutsumi Does

Mutsumi doesn’t ask you to stop. Doesn’t ask you to close your browser. Doesn’t lecture you about “deep work.”

She does one thing: keeps all your threads visible in the corner of your eye, always.

• Your AI agents write tasks into mutsumi.json as they work
• Mutsumi watches the file and re-renders instantly
• You glance, confirm direction, and continue to the next thread
• One hotkey to summon, one hotkey to dismiss

Switch contexts 40 times a day — it doesn’t matter. Every time you glance back, you see exactly what’s on your plate, what’s waiting, what your agents have already pushed forward.

Thread, not Task

In Mutsumi’s narrative, we say “thread” — an ongoing concern your brain needs to track. In the code and API, we say “task” — the data structure. Same thing, different lens.

Where Mutsumi Fits

Mutsumi is one component in a zero-friction workflow:

graph TD
    S["Summon Layer<br/>Quake terminal / tmux popup / Raycast"] --> V
    C["Control Layer<br/>AI Agents / CLI / scripts"] -->|"write"| M["Model Layer<br/>mutsumi.json"]
    M -->|"watch"| V["View Layer<br/>Mutsumi TUI"]
    V -->|"interact"| M

Layer	Role	Examples
Summon	Instant invocation	iTerm2 Quake, Windows Terminal Quake, guake, tmux popup
View	Visual thread table	Mutsumi TUI
Control	Thread creation	AI Agents, mutsumi add, shell scripts
Model	Persistent data	mutsumi.json — local, plain-text, Git-able

She fills the View gap. The ecosystem provides the rest.

Design Principles

1. Zero Friction

From summon to action complete < 2 seconds. No loading screen, no login, no network request.

2. Peripheral

She lives at the edge of your vision. Not center stage. Not hidden. Like a clock on the wall.

3. Agent Agnostic

Not bound to any LLM or Agent. Any program that writes JSON is a legitimate Controller.

4. Hackable First

Users can trivially hack the data structure, theme, keybindings, and views.

5. Local Only — Zero network dependency. Data is files, files are local.

Target User

The Multi-threaded Individual:

• Juggling browsers, chat groups, Reddit, Discord, forums simultaneously
• Running multiple agents on different tasks in parallel
• Terminal is the primary work environment
• Natural affinity for geek tools, loves DIY and customization

NOT for

• PMs who need team collaboration / shared boards (use Linear)
• Project managers who need Gantt charts (use Jira)
• Users who never touch a terminal (use Todoist)

Name Origin

Mutsumi (若叶睦) — from the Japanese “睦” (harmony, closeness). She doesn’t tell you what to do. She waits quietly, holding a sticky note covered with your threads. When you glance at her, she lifts the note a little higher. When you look away, she just waits there in peace.