NextCoWork

A desktop workspace for coding agents.

NextCoWork brings agent-driven development to the desktop: multiple workspaces side by side, a full workbench (editor, terminal, Git, browser), and an agent kernel with a permission model you can actually trust — all in one native app for macOS, Windows and Linux.

English | 简体中文

Features

• Workspaces & sessions — open multiple workspaces in tabs, run parallel agent sessions per workspace, and keep streaming state isolated per session.
• Agent kernel — streaming replies, tool calls with a per-call approval chain, plan mode, subagents, background tasks, todo reconciliation, and scheduled runs.
• Inline visualization — the agent can render live HTML/SVG widgets (charts, diagrams, interactive UI) directly inside the conversation, sandboxed in their own iframe and following the app theme.
• Built-in workbench — file explorer and editors (including plugin-provided custom editors), a real PTY terminal, and background command execution.
• Git integration — status, diff, log and branches; stage, commit, create and switch branches.
• Built-in browser — per-workspace browser partitions that share login state with your tabs.
• Skills & MCP — SKILL.md-based skill packages plus Model Context Protocol support.
• Plugins — a manifest-driven plugin system with per-call capability gates. Plugins can contribute commands, menus, keybindings, custom editors, views, web apps, agent tools and bundled skills.
• Native experience — light/dark themes, accent colors, motion-level control, auto-update, and full Simplified Chinese / English localization.

Model access

Sign in with your NextCoWork account to use the built-in model channel. Create an account at , then sign in from the first-run welcome screen or from Settings → Account — authorization completes in your browser. Once signed in, the NextCoWork provider is added for you (base URL https://nextco.work/v1, OpenAI Responses by default) and your model list syncs on first sign-in; refresh it any time with Fetch model list from provider on the Models page. Usage is metered against your wallet balance, shown on the same Account page. Prefer a provider of your own — OpenAI, Anthropic, DeepSeek, a local gateway? Add it in Settings → Models with your own key.

Download

Installers are published for:

• macOS (Apple Silicon & Intel, .dmg)
• Windows (.exe NSIS installer)
• Linux (.AppImage)

Visit for the latest stable release.

Run a device service

The agent does not have to run inside the app. A headless device service (no Electron, Node.js 24+) can hold the database, the workspaces, Git and the terminals on another machine and keep running while the desktop is closed; the desktop pairs with it over a certificate-pinned connection.

npm i -g @aidotnet/nextcowork-device-service
ncw-device-service diagnose
ncw-device-service --data-dir ~/ncw-device --host 0.0.0.0 --port 7801 --pairing-port 7800 --advertise nas.lan --invite

Paste the printed ncwpair1.… invitation into Settings → Devices → Add device, check the fingerprint, then pair. Installing it as a system service (systemd / LaunchDaemon / WinSW), taking over an existing desktop data root, and troubleshooting are covered in docs/device-service.md.

Development

Requirements: Node.js ^20.19.0 || >=22.12.0 and npm.

npm install        # also runs patch-package and electron-builder install-app-deps
npm run dev        # start the app with hot reload

Verification (the same suite CI runs):

npm run typecheck  # typecheck main + renderer + shared
npm test           # vitest
npm run lint       # eslint

Packaging:

npm run dist       # build installers for the current platform
npm run dist:dir   # unpacked build, for quick checks

Project structure

src/main        Electron main process — agent kernel, plugin host, IPC, storage, updater
src/preload     Context bridge between renderer and main
src/renderer    React UI — chat, workbench, terminal, Git, browser, settings
src/shared      Types and pure functions shared across processes
packages/       Plugin ecosystem — plugin-api (types), plugin-cli, create-nextcowork-plugin
examples/       Sample plugins (custom editors, web apps, …)
resources/      App icons, bundled skills, plugin runtime

Writing plugins

Scaffold a plugin that is ready to build and publish:

npm create nextcowork-plugin@latest

• Type definitions: @aidotnet/plugin-api
• Build / package / publish: @aidotnet/plugin-cli (binary nextcowork-plugin)
• Working examples live in examples/

A plugin is a package.json manifest plus optional code: declare what you contribute (commands, custom editors, views, agent tools, web apps, skills, …) and the capabilities you need; the host enforces per-call permission gates at runtime.

License

This project is licensed under the Apache License 2.0.

The plugin ecosystem packages under packages/ (@aidotnet/plugin-api, @aidotnet/plugin-cli, create-nextcowork-plugin) are distributed under the MIT license — see each package's package.json.