한국어Download
MoaEditor
Download

The code editor that runs AI as an org chart

Orders flow down.Blockers flow up.

MoaEditor puts multiple AIs to work as a CEO, team leads, staff and interns. Connect the subscriptions you already pay for through their official CLIs, and hand routine work to free intern models running on your own PC.

Download free for WindowsSee how it works

Windows 10 · 11 (64-bit) · Version 0.1.0 · Free for individuals

A CEO (Claude Code) leads two leads (Codex, a Gemini API key); under them are staff (Copilot, Cursor, Kimi) and interns running on your own PC (qwen3.5:9b, gemma4:e4b). Instructions flow down; blocked work flows back up.

  • Official CLIs onlyRuns tools like Claude Code and Codex unmodified
  • Your plan, your accountNo account sharing, no rotating between accounts
  • Respects every ToSNever reads other vendors' login token files
  • Korean-firstKorean UI from installer to menus — English too

How it works

Write one goal. The org splits it and gets it done.

The CEO breaks work down and delegates; the hands-on work happens below. When someone gets stuck, the level above resolves it — cheapest fix first.

  1. Instructions flow down

    Write one goal. The CEO splits it into tasks for the leads, who break them down again for staff and interns. Managers plan and delegate — they don’t touch files while delegating.

  2. Everyone plays to their strengths

    Work is assigned by who can edit files, who still has usage left right now and who is good at what. Light jobs like commit messages and summaries go to the intern on your PC, saving your subscriptions.

  3. Stuck? It goes up

    When tests keep failing or a member reports being blocked, the work goes to their direct manager, who picks the cheapest fix first: retry with a hint, reassign, take over — and only then escalate one level.

  4. Results come together

    A task is done only when its acceptance checks — tests, lint, build — pass. Results roll up to the CEO for review, and every run keeps a record of who did what.

Features

Everyone does what they do best

Every subscription and model has different strengths. MoaEditor picks an assignee per task by rank, strength tags, whether it can edit files, and remaining usage.

Subscription agents

Runs the official CLI you signed in to yourself

  • Claude Code
  • Codex CLI
  • GitHub Copilot
  • Cursor Agent
  • Gemini CLI
  • Kimi CLI
  • Qwen Code
  • OpenCode
  • Goose

API keys

Key values live in the OS keychain, not in config files

  • Anthropic
  • OpenAI
  • Gemini API
  • DeepSeek
  • Mistral
  • Moonshot Kimi
  • Upstage Solar
  • HyperCLOVA X
  • Azure OpenAI
  • AWS Bedrock

Your PC · on-prem

Works without the internet

  • Ollama
  • LM Studio
  • llama.cpp
  • Jan
  • vLLM
  • SGLang

Product names are trademarks of their owners and do not imply partnership or endorsement.

Per-task assignment

Scores candidates by rank, strength tags and current load, with a small bonus for recent success rate and speed.

Checked against done criteria

Each task carries test, lint or build commands. On failure, the output goes back with the retry.

Usage-aware routing

Keeps the per-minute and per-day limits you set in a ledger, and rests an assignee that has reached them.

A record of every run

Who decided what and which commands ran are saved in your project’s .moa/runs folder.

When work gets stuck, it climbs

A manager always tries the cheapest fix first.

  1. Retry with a hintThe same assignee tries again with advice attached.
  2. ReassignAnother report on the same team picks it up.
  3. Take overThe manager switches to hands-on mode and fixes it.
  4. EscalateStill blocked? It goes one level up. If the CEO is stuck, you get a report.

Editor

The org works inside an editor you already know

Built on Code - OSS, so your shortcuts and extensions carry over. Type a goal and a task tree appears while the org chart reacts in real time.

  • Extensions from Open VSX
  • Telemetry and experiments off by default
  • Korean or English UI

Typing “Fix the bug that charges customers twice” into the editor makes the CEO split it into three tasks for two leads. Staff and interns work in parallel; when the intern gets stuck on a test, the lead reassigns it to another staff member who finishes it. The org graph on the right moves in sync with the task list on the left.

Local interns

Routine work goes to an intern on your PC

Free open models pulled with Ollama write commit messages, summaries and small fixes. MoaEditor recommends a model that fits your memory, and inside the workspace the intern can read and edit files.

  • No usage feesIt runs on your PC, so there is no API bill.
  • Code stays localWhat a local intern processes never leaves your machine.
  • License-checkedOnly models licensed for unrestricted commercial use are auto-installed.
Your PC’s memory (RAM)

8GB Autocomplete and short text jobs

  • InternQwen3.5-2BFast

    Commit messages, task routing, short summaries

    qwen3.5:2b-q4_K_MMemory 2.6GB / 4.2GBcontext 16KApache-2.0

  • AutocompleteQwen2.5-Coder-1.5B-BaseFast

    Suggests the next lines as you type

    qwen2.5-coder:1.5b-baseMemory 1.6GB / 4.2GBcontext 4KApache-2.0

What it can take on

  • Tab autocompleteYes
  • Commit messages & task routingYes
  • Summaries, code explanations, Korean writingGoes to subscriptions
  • Read, search & edit files (sandboxed, short tasks)Goes to subscriptions
  • Write tests & multi-step work (staff level)Goes to subscriptions

File edits and multi-step work go to your connected subscriptions.

16GB An intern that edits files

  • InternQwen3.5-9BModerate

    Short read-search-edit tasks, commit messages, summaries

    qwen3.5:9bMemory 8.1GB / 10.7GBcontext 32KApache-2.0

  • AutocompleteQwen2.5-Coder-1.5B-BaseFast

    Suggests the next lines as you type

    qwen2.5-coder:1.5b-baseMemory 1.6GB / 10.7GBcontext 4KApache-2.0

What it can take on

  • Tab autocompleteYes
  • Commit messages & task routingYes
  • Summaries, code explanations, Korean writingYes
  • Read, search & edit files (sandboxed, short tasks)Yes
  • Write tests & multi-step work (staff level)Goes to subscriptions

The intern reads and edits files inside your workspace, and escalates bigger jobs.

Other options: Gemma4-E4B · Qwen3.5-4B · A.X-4.0-Light-7B

32GB Local staff, too

  • InternQwen3.5-9BModerate

    Short read-search-edit tasks, commit messages, summaries

    qwen3.5:9bMemory 8.1GB / 24.7GBcontext 32KApache-2.0

  • StaffQwen3.6-35B-A3BModerate

    File edits, tests and multi-step work (3B-active MoE, usable on CPU)

    qwen3.6:35b-a3bMemory 24.1GB / 24.7GBcontext 32KApache-2.0

  • AutocompleteQwen2.5-Coder-1.5B-BaseFast

    Suggests the next lines as you type

    qwen2.5-coder:1.5b-baseMemory 1.6GB / 24.7GBcontext 4KApache-2.0

What it can take on

  • Tab autocompleteYes
  • Commit messages & task routingYes
  • Summaries, code explanations, Korean writingYes
  • Read, search & edit files (sandboxed, short tasks)Yes
  • Write tests & multi-step work (staff level)Yes

Intern and staff take turns in memory; switching may take a few seconds to reload.

Other options: gpt-oss-20b · Qwen3.6-27B · Gemma4-26B-A4B · Devstral-Small-2-24B

64GB Room for bigger models

  • InternQwen3.5-9BModerate

    Short read-search-edit tasks, commit messages, summaries

    qwen3.5:9bMemory 8.1GB / 53.7GBcontext 32KApache-2.0

  • StaffQwen3.6-35B-A3BModerate

    File edits, tests and multi-step work (3B-active MoE, usable on CPU)

    qwen3.6:35b-a3bMemory 24.1GB / 53.7GBcontext 32KApache-2.0

  • AutocompleteQwen2.5-Coder-1.5B-BaseFast

    Suggests the next lines as you type

    qwen2.5-coder:1.5b-baseMemory 1.6GB / 53.7GBcontext 4KApache-2.0

What it can take on

  • Tab autocompleteYes
  • Commit messages & task routingYes
  • Summaries, code explanations, Korean writingYes
  • Read, search & edit files (sandboxed, short tasks)Yes
  • Write tests & multi-step work (staff level)Yes

Intern and staff stay loaded together, with options up to an 80B coding model.

Other options: Qwen3.6-27B-Q8 · Qwen3-Coder-Next-80B-A3B · Qwen3.6-27B · gpt-oss-20b

  • Runs on your PC through Ollama. Code and prompts never leave the machine.
  • Every recommended model is Apache-2.0, so it is fine for work. Non-commercial models such as EXAONE are never auto-installed.
  • Figures assume a PC without a discrete GPU. With a GPU or Apple unified memory, the editor detects your hardware after install and recommends a faster setup.

Measured A qwen3.5:9b intern passed a test-verified bug fix 3 out of 3 times. (Ollama 0.35, 2026-10-02, dev PC)

Safety

Hand work to AI without losing the guardrails

The more an AI can edit files and run commands, the clearer its limits need to be. The defaults are conservative.

Nothing outside the workspace

Parent paths, links pointing outside, and the .git folder are blocked.

Secret files stay closed

.env files, private keys, certificates and cloud credentials can’t be read, searched or written; key variables are stripped from command environments.

Commands need approval

Anything other than the task’s own done-criteria command asks you first.

Vendor logins stay with the vendor

Each CLI signs in by itself. MoaEditor never reads or stores those tokens.

Free tiers blocked in enterprise mode

Free tiers that may train on inputs and personal trial keys are rejected at config time in enterprise mode.

Telemetry off

Usage data, experiments and surveys are off from the start.

Enterprise & public sector

The same org, even on an air-gapped network

On closed networks in government and finance, build the whole org — CEO to intern — from on-prem inference servers and local models. Licenses are verified offline too.

  • Closed-network edition

    Connects on-prem servers like vLLM and SGLang plus local models — no external AI service required.

  • Offline license

    Register a single .moalicense file. Signature and validity are checked on the PC, no server needed.

  • KCMVP-listed signatures

    Licenses are signed with ECDSA P-256, an algorithm on Korea’s KCMVP validated list.

  • Enterprise mode

    Blocks connections that may train on your inputs and reminds you to use business plans for subscriptions.

  • Audit logPlanned

    Records who used which model for what, per organization.

  • Enterprise SSOPlanned

    Finds your SAML or OIDC sign-in from your company email domain.

Contact salesEnterprise details

Pricing

Free for individuals

You pay AI vendors directly for their services. MoaEditor helps you use those subscriptions well — it never buys or resells usage in between.

  • Personal

    Available now

    For individual developers and students

    ₩0No sign-in required

    • The full AI org chart editor
    • Subscriptions, API keys and local models
    • Local intern picks and file editing
    • Korean and English UI
    Download free

  • Pro

    For individuals who delegate more

    Coming soonPrice announced at launch

    • Everything in Personal
    • MoaEditor account sync
    • More features announced at launch
    Coming soon

  • Enterprise

    For companies, institutions and closed networks

    Contact usQuoted for your organization

    • Enterprise mode
    • Organizations and seats
    • Offline license for closed networks
    • Audit log and budgets (planned)
    • Onboarding and install support
    Contact sales

Plans and prices may change before general availability.Pricing details

FAQ

Questions

Is this a tool for sharing accounts or stretching usage limits?

No. You connect your own accounts through each vendor’s official CLI. There is no feature for rotating multiple accounts, and MoaEditor never reads other vendors’ login token files. If anything, it helps you stay within the limits you set.

Do I need a paid AI subscription?

No. You can start with local models from Ollama alone. That said, the CEO and team-lead seats — which plan and judge — do best with a stronger model.

Does my code go to MoaEditor’s servers?

No. Work runs on your PC and content goes only to the AI services you connect (under their terms). What local interns process stays on your machine. The MoaEditor account server only handles sign-in and plan checks.

Can I use VS Code extensions?

Extensions on Open VSX install as usual. Microsoft-only extensions such as C/C++, Pylance and Remote-SSH can’t be used for licensing reasons; MoaEditor suggests alternatives.

Do I have to sign in?

Personal mode works without signing in. Enterprise mode turns on with an enterprise-plan account or a valid license file.

Is there a macOS or Linux version?

Right now there is a Windows 10/11 (64-bit) build. macOS and Linux versions are in the works.

Windows says “Windows protected your PC” when I install.

The installer isn’t code-signed yet, so SmartScreen warns about it. Click “More info” → “Run anyway” to continue. Signing will be in place before general availability.

How do we deploy on a closed network?

Get in touch. We issue a license file with your organization, seats and term, and help design the connection to your on-prem inference servers.

Set up your first AI org today

Installs to your user folder without admin rights. Free for individuals.

Download free for WindowsContact sales