What is Claude Cowork — and how to use it

Claude Cowork is Anthropic's agentic AI for knowledge work. It runs inside the Claude desktop app and works directly with your local files, folders, and applications. Where Claude chat is a conversation, Cowork is a working session: you describe a task, Claude plans and executes it across multiple files and steps, and you steer along the way. It uses the same agentic architecture that powers Claude Code, but without needing the terminal.

Cowork is included on every paid Claude plan — Pro, Max, Team, and Enterprise — but not on the Free plan. Higher tiers mainly raise how much you can use it before hitting limits, so if you run Cowork daily or on a schedule, expect to need a higher tier such as Max. Verify current plans and pricing at claude.com/pricing.

Your files stay on your machine. Cowork runs in an isolated virtual machine on your own computer with controlled file and network access, and it only touches the folders you explicitly grant it. Your documents and data are not uploaded for training or cloud storage. An internet connection is still required because the model reasoning runs on Anthropic's servers, but your local files are not sent there wholesale.

Both use the same underlying agent, but they target different work. Claude Code is for developers: it runs in the terminal, works inside a code repository, and its output is software. Claude Cowork is for knowledge workers: it runs in the desktop app with no terminal, works across documents, spreadsheets, and everyday apps, and its output is finished work — reports, organised files, extracted data. Reach for Code when the output is software; reach for Cowork when the output is knowledge work.

Cowork is built for tasks too big for a chat window. Typical jobs: reading dozens of files in a folder and producing a structured summary; extracting and reconciling data across many documents; preparing reports and briefs; sorting and renaming files by their content; standardising document sets; and running repeatable, multi-step workflows across your apps. It handles the whole task rather than answering one question at a time.

Cowork is single-user and desktop-only (no web or mobile), and it needs a paid plan plus an internet connection. Memory works within projects, but it does not retain memory across standalone sessions, and the desktop app must stay open — closing it stops a running task. Long, multi-step tasks consume plan usage much faster than ordinary chats. External connectors (such as Gmail or Google Drive) can be inconsistent, and — worth noting for regulated teams — Cowork activity is not captured in Anthropic's Compliance API at this time. Because it acts on real files, review its plan before letting it run on important folders.

Cowork runs through the Claude desktop app on macOS, Windows, and Linux (beta) — it is not available on the web or on mobile. Rather than publishing fixed hardware specs, Anthropic's app runs a quick readiness check to confirm your computer can run it. Because availability and system requirements change quickly for a new product, confirm the current list on Anthropic's Help Center before you plan a rollout.

This is Cowork's boundary, and it is an important one. Cowork is a personal desktop agent: it helps an individual finish their own knowledge work. It does not run team-wide automations, connect to your CRM or ERP, or operate as a shared always-on service. When you need automation that spans a whole team and plugs into your business systems, that is a custom Claude agent — built with the Claude Agent SDK and the Model Context Protocol (MCP). 安克极速, a member of the Anthropic Claude Partner Network, builds exactly that, and can also train your team to get more out of Cowork itself.