CoWork OS – An Open‑Source AI‑Powered Desktop Workspace

What it is

• A cross‑platform (macOS & Windows) desktop application that bundles code editing, browsers, terminals, document creation, and AI agents into a single, locally‑stored workspace.
• Users can plug in any LLM provider (OpenAI, Anthropic, Gemini, Azure, local models, etc.) or use API‑key‑based gateways, while the UI and automation logic stay the same.

Core ideas

• Local‑first: All workspace state, credentials and generated artifacts live on the user’s machine. Only the data required for a specific AI request is sent to the chosen model.
• Governed AI actions: Access profiles (Ask‑for‑approval, Approve‑for‑me, Full‑access, Custom) let teams enforce sandboxing, approvals and resource limits for every agent task.
• Unified UI + CLI: The graphical “Bots/ Mission Control” console is complemented by a cowork command‑line tool that can run interactive sessions or one‑off tasks, sharing the same configuration.

Key features (as described in the README)

• Multi‑provider model routing – configure API keys, gateways, cloud credentials or local model binaries; switch providers without changing the rest of the app.
• Agent runtime – long‑running “Ask/Do/Plan” agents, Mixture‑of‑Agents presets, multitask fan‑out, side‑chat, and visible routing/fallback state.
• Everything Workbench – generated Word/Excel/PowerPoint‑style files, PDFs, HTML pages, and code artifacts appear as clickable cards with in‑app preview and a follow‑up composer.
• Terminal Tabs – real PTY‑backed xterm.js terminals live inside the workspace, next to AI‑generated artifacts.
• Browser Workbench (V2) – an embedded, instrumented browser for web‑testing; can fall back to local Playwright, a user Chrome profile, or a cloud stealth browser.
• Automation Studio – visual, versioned flow builder with variables, branches, approvals and durable activity logs.
• Inbox & Channels – local‑first mail triage, searchable “Ask Inbox”, plus 17 messaging channel integrations (e.g., WhatsApp Business, Twilio SMS).
• Box Brain – optional background indexing of a Box folder for private, incremental recall while keeping Box as the source of truth.
• Skill & Plugin ecosystem – built‑in packs (e.g., resume builder kami, video generation, architecture design) and a store for community‑contributed skills.
• Security & privacy – encrypted local credential store, sandboxed cowork-preview:// web previews, session‑scoped location prompts, and a full automated test suite.

Typical use cases

• A developer who wants to code, run terminal commands, and ask an LLM to generate documentation or refactor code without leaving the desktop.
• A product team that routes tasks through an approval workflow, letting an AI draft emails, create slide decks, or run UI tests in the embedded browser.
• Knowledge‑workers who keep a local “LLM‑wiki” vault, generate PDFs, spreadsheets, or presentations, and have the AI retrieve and summarize that material on demand.

Technology stack

• Electron 44 for the cross‑platform desktop shell.
• Node.js runtime (published on npm as cowork-os).
• xterm.js + node‑pty for terminal tabs.
• React UI components (implied by the mention of React workbench and V2 browser).
• Various LLM provider SDKs (OpenAI, Anthropic, Gemini, Azure, OpenRouter, oMLX, etc.) accessed via API keys or local binaries.
• Playwright (optional) for browser automation; also supports custom Chrome instances.
• File‑type renderers (highlight.js, PDF.js, Office‑web viewers, etc.) for in‑app previews.

Maturity & community

• Actively versioned; latest stable release 0.5.60 (July 2024) with a detailed changelog and release notes.
• CI badge indicates automated testing on GitHub Actions.
• MIT‑licensed, with contribution guidelines and a security policy.
• Documentation is extensive (getting‑started guide, feature docs, CLI reference, architecture notes).

Getting started

1. Download the appropriate installer from the GitHub Releases page (macOS DMG or Windows EXE).
2. macOS users can install via a one‑line curl … | bash script that avoids quarantine dialogs.
3. Run the app, add a model provider (API key or local binary), and start using the built‑in bots or the cowork CLI.

License

• MIT © CoWork‑OS contributors.

---

CoWork OS is a genuine, open‑source AI‑agent desktop platform aimed at integrating everyday productivity tools with flexible, multi‑provider large‑language‑model capabilities.