Claude Projects in Chat vs Cowork
Anthropic says projects work the same way in chat and in Cowork, and the same project can start either. What changes is what the work does: a chat answers, a Cowork task goes away and comes back with files. What each surface adds to a project, where memory is shared, and where it is not.
7 min read
A Claude project is the same project whether you open it in chat or in Cowork: Anthropic’s help centre says projects work the same way in both, and from one project you can start either a chat or a Cowork session, with the project’s knowledge as context. The difference is in the work, not the project. A chat answers you in the conversation. A Cowork task is handed off — Claude plans it, runs the steps, and comes back with finished files — and a Cowork project can also hold a local folder, scheduled tasks and its own memory. Below is what each surface adds, what is shared between them, and the limits the help centre spells out.
What Claude Cowork is
Anthropic’s guide to getting started with Claude Cowork (opens in a new tab) describes it as bringing Claude Code’s agentic capabilities to knowledge work beyond coding, with no terminal required. Instead of answering one prompt at a time, Claude takes on multi-step work: you describe an outcome, step away, and come back to formatted documents, organised files or a piece of research.
- Plans: Cowork is on paid plans — Pro, Max, Team and Enterprise — with Enterprise availability on some surfaces depending on the admin.
- Where it runs: sessions run in the cloud (in beta) on Anthropic’s servers, in an isolated environment, and follow your account across desktop, web and mobile. Work continues if you close your laptop.
- What needs your computer: reading your local files, using a browser, and computer use go through the Claude Desktop app, which has to be open.
- What it produces: spreadsheets with working formulas, presentations, documents, and organised folders, delivered to the session to preview and download.
- How much it asks: a permission setting controls whether Claude asks before each action or works on and checks actions itself.
One thing is moving under your feet. Anthropic announced on 16 September 2026 that Cowork and chat are becoming one Claude (opens in a new tab): you ask for what you need and Claude decides whether that is a quick answer or a task. It is rolling out gradually, starting with Pro and Max. If your message box still shows separate “Chat” and “Cowork” options, you do not have it yet; your chats, tasks and projects move over with you when you do.
What a project holds on each surface
In chat, a project has knowledge (files you upload), instructions that apply to every chat in it, its chats, and — where memory is on — its own memory. The help page on projects in Claude Cowork (opens in a new tab) lists what a Cowork project has of its own:
- Instructions: tone, formatting or rules that guide every task in the project.
- Scheduled tasks: recurring tasks that belong to that project.
- Context: a local folder, a linked chat project, or a URL for Claude to reference.
- Memory: Claude remembers context from tasks in the project and applies it to later tasks in the same project, and only that project.
You can create a Cowork project from scratch, from an existing folder on your computer, or by importing a chat project, which transfers its files and instructions (one project at a time). New projects made in Cowork are saved to your Claude account and follow you to other devices. Projects made from a folder on your computer stay on that computer.
Chat vs Cowork, side by side
- Shape of the work. Chat: a conversation, answer by answer. Cowork: a task Claude plans, breaks into steps and finishes, sometimes with sub-agents working in parallel.
- Where it runs. Chat: the conversation. Cowork: an isolated cloud environment, reaching your computer through the desktop app when a task needs it.
- What you get back. Chat: text and artifacts in the conversation. Cowork: files you download or keep refining.
- Unattended work. Chat: none. Cowork: tasks keep running after you close the laptop, and scheduled tasks run with no device online.
- Local folders. Chat: none; you upload files to knowledge. Cowork: a project can be tied to a folder, on desktop only.
- Sharing. Both: on Team and Enterprise, Cowork projects share the same way as chat projects, with Can view and Can edit. A Cowork session itself cannot be shared, though artifacts from it can.
One rule matters more than the rest. The page on using Cowork on web, desktop and mobile (opens in a new tab) says Cowork won’t change a project’s contents, so add anything you want to keep to the project yourself. A Cowork task can produce a polished report from the project’s knowledge; it does not file that report back into the project for the next chat.
Memory: shared in the cloud, not locally
Anthropic’s article on chat search and memory (opens in a new tab) says memory is shared between chat and Cowork when Cowork runs in the cloud: what Claude remembers from chats is available to a Cowork task, and what comes up in the task carries back to chat. Cowork sessions that run locally on your computer do not use memory.
Inside projects the walls stay up on both surfaces. A chat project has its own memory space, separate from other projects; a Cowork project’s memory is scoped to that project. The details — what is remembered, how to view or edit it, and incognito chats — are in Claude project memory.
Which surface for which job
- Use chat when you want to think with Claude: drafting, questioning a document, trying three phrasings of a paragraph. You stay in the loop on every reply.
- Use Cowork when you can describe the finished thing: “turn these five interview notes into a themed summary with a quote each”, “a spreadsheet with a tab per region”. Say the outcome, the format and the inputs.
- Use a Cowork project with a local folder for material that lives on your machine and changes often, such as a folder of contracts. Remember it only works on the desktop app.
- Use a scheduled task in a Cowork project for anything you would otherwise do every Monday.
- For work with real consequences — messages sent as you, important files — keep the permission setting on asking first and review what Claude does.
A third meaning of “project” is arriving too. Anthropic is rolling out a new version of projects in beta, starting in Claude Code, where a project is one conversation that splits work into parallel threads; chat and Cowork are due to follow. Claude Projects vs Claude Code explains that naming trap.
Keeping a record of Cowork work
Cowork is good at finishing things while you are away, which is exactly why you want a record of what it finished. A session cannot be shared, and Cowork does not write results back into the project, so the outcome of a task can end up visible only in one person’s session list.
A board fixes that. Custom connectors work in Cowork as they do in chat, so you can add fenbs with the Claude connector and put the rhythm in the project’s instructions: read the task before starting, comment what was produced and where, and leave the move to Completed to a person. Every change Claude makes appears in the board’s history as “Claude via” you, so a colleague can see what a Monday scheduled task did without opening your session.
## Board - At the start of a task, call fenbs_get_context, then find the task this work is for with fenbs_search. - When you finish, comment on that task: what you produced, the file names, and anything you could not do. - Never move a task to Completed. Say it is ready; a person will move it.
Related
Getting the most from a project on any surface: Claude Projects best practices. Whether a project can draw on another: can Claude projects reference each other?. Connecting any assistant to a board: how to connect Claude to your project board with MCP.