MoaEditor: A Windows Code Editor with an AI Team and Org Structure

⚡ The gist in 5 seconds - MoaEditor has launched — a Windows code editor where AI agents are organized into a hierarchy: director, team leads, employees, and interns. - The project is published on GitHub (repo moaeditor/moaeditor), runs on Windows 10/11, with an interface in English and Korean. - Caveat: the installer is not code-signed, so Windows may show a warning; the source code is not available in this repository. ### 🔍 What was found The moaeditor/moaeditor repository on GitHub now features an overview of MoaEditor — a Windows code editor built on Code - OSS (MIT) and not affiliated with Microsoft. It claims support for 13 subscription-based tools: Claude Code, Codex, GitHub Copilot, Cursor, Gemini CLI, OpenCode, Kimi Code, Qwen Code, goose, Augment, Mistral Vibe, Factory Droid, and Cline. It also connects to OpenAI, Anthropic, Google Gemini, Groq, and OpenRouter APIs, enterprise clouds, and Korean models, while local options include Ollama, LM Studio, llama.cpp, Jan, Foundry Local, Docker Model Runner, vLLM, and SGLang. The key feature is that agents are arranged into an org structure: the director plans and reviews, team leads distribute work, employees write code, and interns handle simple tasks on a local model. Blocked work escalates up the hierarchy. You can describe a project in one line, and the AI will pick a model for each role, creating AGENTS.md, CLAUDE.md, and.gitignore. Tasks can be run all at once (Approve and Run) or step by step (Approve Each Step), with modes including Plan only, Ask for everything, Ask only for important things, and Auto. According to the authors' calculations, splitting work between expensive and cheap models cuts the bill for expensive models by up to 85% (planning and review are estimated at 15% of the work). A separate Helper AI verifies claimed fixes, risky commands, and minor calls; it runs locally on an NVIDIA GPU with 8 GB or more, or via the MoaEditor cloud (2,000 checks per day) or a Cloudflare token. ### 💡 Why it matters This approach avoids burning expensive tokens on simple operations: instead of a single model, a pool of different AIs works together, with each task assigned to the agent best suited to it in capability and cost. It's practical for developers who want to use Claude Code, Copilot, or local models in one window under unified control. The project is free for individuals, students, schools, non-profits, and companies with fewer than 50 employees — you only pay the AI providers. That said, keep in mind the unsigned installer and the fact that the project's source code hasn't been published, so trust it with reservations. ### 🧩 Context In a footnote, the repository's authors cite the RouteLLM research: a similar approach cut costs by up to 85% while retaining 95% of GPT-4's performance on MT Bench. They base their own estimates on Claude Opus 5.5 at Anthropic's official prices as of 2026-10-06, warning that the total token count may increase due to multiple AIs running simultaneously. The project belongs to Horizon Co., Ltd., © 2026.

🔗 Read on github.com