How nonaAi works

Give one agent many jobs, many personalities — and run them as one team.

nonaAi is a desktop workbench. It turns the AI agents you already own into a governed team: each duty gets its own character and charter, one stand-in answers for you, and every decision is a card you keep. One subscription becomes a whole specialized workforce.

your project — nonaAi
Coder
head: Claude Code · hat: fast-builder
Reviewer
head: Claude Code · hat: skeptic
Researcher
head: Gemini · hat: digger
Legatus
runs the team the way you would
Your PCEverything runs on your own computer.
Your agentsThe AI programs you already own, under your accounts.
Your tools & MCPYour CLIs, configs and MCP servers, driven as-is.
One organizationAll of it governed as a single team.

The unique part · hats & heads

One head. Many hats.

A head is the agent underneath — Claude Code, Codex, Gemini. A hat is the character and job you give it: its personality, its specialization, what it does and what it's allowed to touch. Put a different hat on the same head and you get a different specialist. Pick a hat — watch the same agent become someone new.

The wardrobe — pick a hat

The head — pick an agent

🛠️
The Fast Builder
wearing this hat: Claude Code · a duty charter
7 heads × 6 hats
One agent subscription, worn as many specialists. Mix and match — the same head can wear every hat, and every hat fits any head.

A hat is a real artifact in the app — a duty charter with five parts: IDENTITY, OBJECTIVE, RESPONSIBILITIES, METHODOLOGY and an AUTHORIZATION PROFILE. Example charters shown; you write your own. Authorization can never exceed what the underlying agent is already allowed to do.

How you make a hat

Tailor it by hand — or let the agent tailor it with you.

Two ways to build a duty's character. Edit the charter yourself, or describe it in plain words and let a real agent draft, sharpen and rehearse it before it joins the team.

by hand
Duties Builder

Write the charter yourself

Open a duty and edit its five sections directly — who it is, what it's for, what it does, how it works, and what it may touch. Full control, nothing hidden.

with the agent
Smart Duties Builder

Describe it — the agent drafts it

Say what you want in plain words. A real, named agent drafts the charter, improves it from your notes, reviews it against the whole team, and rehearses it on a test case before you apply a thing.

describedraftimproveteam reviewauditionapply

The Smart Duties Builder is off by default; every draft is previewed with a before/after diff and the exact briefing before anything is applied. You always get the last word.

What makes it a team, not a pile of terminals

Four things run the team — so you don't have to babysit it.

Running several agents at once is common now. Running them under governance is not. This is the part that's nonaAi's own.

The Legatus · your stand-in

Answers for you, within the minute

At the end of every turn it reads each lane, answers what it can under your laws, and hands you only what it can't decide — as a card, recommended answer first. Deaths, silences and stalls reach it first.

Laws every agent reads

One rulebook, whatever the vendor

Your project's laws, objectives and the duty's charter ride the first message of every spawn — for Claude, Codex, Gemini, Kimi, Qwen, Hermes and opencode alike.

A Judge, independent of the workers

Checks the work, doesn't do the work

An optional Judge reviews each finished turn — on track, needs you, done, rule risk, stuck, next duty — and files a verdict card. Off by default; about a hundredth of a cent per question set.

A budget you set

It stops at your line and asks

Set a budget per task and per day. See what each teammate spends, in dollars. When a lane reaches your line, it stops and asks — it doesn't quietly run up a bill.

You decide · nothing silent

Every question is a card. The recommended answer is first.

When a teammate needs you, it doesn't wait in silence in some terminal you forgot to check. It puts a card in front of you with clear, numbered choices — and marks the one it recommends.

The rule: a job is never stopped on a question typed as plain text. If a teammate is unsure, it asks — as a card you answer in one click, from the desk or the phone.
your project — nonaAi
Reviewer asks you
Two totals don't match the source. What now?

Files win · a project is a folder

A folder you own, not a black box.

Plans, memory, rules and every conversation are ordinary files inside your project folder. The baton — what was done and decided — lives there too. Open them with anything. Move the folder, and your team comes with it.

your-project/ .agent/
📁 your-project/
📄 HANDOFF.md — the baton
📄 STATE.md
📁 .agent/
📁 plan/ — the board
📄 laws.md — every agent reads this
📁 duties/ — the hats live here
📁 src/ — your actual work

In motion

How one turn flows.

You give the team a job

Type it once. The Planner breaks it into steps on the board; each duty picks up the step that's its own.

Each hat works its lane

Its assigned agent runs on your machine, under your account, in character — and declares an outcome when the turn ends.

The Legatus reads every lane

It answers what it can under your laws, and routes the rest — a handoff to the next duty, or a card to you.

The Judge and budget watch

The Judge checks the finished turn; the budget meter counts the spend and stops the lane at your line.

Everything is written down

The board, the baton and every conversation land as files in your folder. Nothing happens off the record.

One head, many hats. One team, your machine.

Turn the agents you already pay for into a whole specialized workforce.