What are Claude Cowork Projects
On March 20, 2026, Anthropic released Claude Cowork Projects — a feature that turns the Claude desktop app into a structured AI workspace. Instead of starting every conversation from scratch, a Cowork Project bundles context, instructions, files, memory and scheduled tasks into a single persistent environment.
The concept is straightforward. You create a project, give it a name and a focus area, then add whatever Claude needs to work effectively: local folders, linked chat projects, URLs, and custom instructions. Every conversation inside that project inherits the full context automatically.
Cowork Projects are available on macOS and Windows (x64 only) for users on Pro, Max, Team and Enterprise plans. They run entirely on the desktop — files stay local, and there is no cloud sync at this stage. If you are already using Claude Cowork for businesses, Projects are the natural next step: a way to organize Cowork around specific roles, clients or workflows.
How to create a Cowork Project
There are three ways to start a Cowork Project. The first is Start from Scratch: you create a blank project, name it, add context sources and write instructions. This works well when you are building a new workflow from the ground up — for example, a project dedicated to weekly reporting or competitive analysis.
The second method is Import from Claude Chat. If you already have a Claude chat project with documents and system prompts, you can import it directly into Cowork. The context carries over, and you gain access to desktop-only features like local folder access and scheduled tasks.
The third option is Use Existing Folder. Point Cowork at a local directory — a codebase, a research folder, a set of financial documents — and the project indexes it as context. This is particularly useful for teams that already have well-organized file structures and want Claude to work within them.
Once the project is created, you can refine it at any time: add new context sources, update instructions, or enable and disable memory.
Memory: Claude remembers across conversations
Memory is enabled by default in every Cowork Project. This means Claude retains context from previous conversations within the same project — preferences, decisions, terminology, recurring patterns — without you having to repeat yourself.
In practice, this changes the dynamic significantly. If you spend the first session explaining your company's naming conventions, product tiers and reporting format, Claude carries that knowledge into every subsequent conversation in the project. The AI builds a working understanding of your domain over time.
Memory is scoped to the individual project. What Claude learns in your marketing project does not leak into your finance project. You can also review, edit or delete specific memories at any time, giving you full control over what persists.
For businesses, this is a meaningful shift. Instead of treating Claude as a stateless tool that needs re-prompting every session, Cowork Projects create a persistent working relationship — closer to how you would onboard a new team member who gradually learns how things are done.
Want to discuss with an expert?
30 minutes to discuss your specific case.
Instructions and scheduled tasks
Every Cowork Project supports custom instructions that apply to all conversations within it. These are not one-off system prompts — they are persistent rules that shape how Claude behaves across every task in the project. You can set tone of voice, output format, compliance requirements, terminology preferences or any other constraint that matters for the specific workflow.
For example, a legal review project might include instructions like: always cite the relevant article number, use formal language, flag ambiguous clauses explicitly. A content production project might specify: write in active voice, keep paragraphs under four sentences, follow our brand guidelines.
Scheduled tasks take this further. You can configure Claude to run specific tasks on a recurring basis using natural language — for example, "summarize all new emails in the shared inbox every morning at 8" or "run this analysis every 2 hours between 9 and 5." Claude executes the task within the project context, using whatever files, memory and instructions are already loaded.
This effectively turns Cowork Projects into lightweight automation hubs. Combined with Dispatch, which lets you trigger tasks remotely from your phone, you get a system where Claude works asynchronously on your behalf — checking, summarizing, drafting and flagging without waiting for a prompt.
Business use cases for Cowork Projects
The most immediate use case is role-based workspaces. A finance team can create a project loaded with quarterly reports, budget templates and forecasting models — with instructions to follow internal formatting standards. Every analyst on the team opens the same project and gets consistent, context-aware outputs.
Client-specific projects are another strong pattern. Consultancies, agencies and professional services firms can set up one project per client, with all relevant documents, past deliverables and communication history as context. Claude effectively becomes a client-aware assistant that remembers the engagement history.
Operational monitoring works well with scheduled tasks. A project can be configured to pull data from local files or connected sources at regular intervals, generate summaries and flag anomalies. This is particularly useful for teams that need to track KPIs, compliance metrics or market signals without dedicating staff to manual review.
Content and communications teams benefit from projects that encode brand voice, editorial guidelines and past campaign data. Instead of attaching a style guide to every conversation, the project holds it permanently — and Claude's memory ensures it learns from feedback over time.
For a deeper look at how Claude fits into enterprise workflows, see our guide on how to integrate Claude into your organization.
Current limitations and what is coming next
Cowork Projects are desktop-only. There is no cloud sync, which means a project created on your office Mac is not available on your laptop unless you manually recreate it. Files remain locally stored, which is good for security but creates friction for distributed teams.
Project sharing is not available yet. You cannot share a project with a colleague or export its configuration for someone else to import. Anthropic has indicated this is on the roadmap, but there is no confirmed timeline.
The feature is also not integrated with Claude Code yet. Developers using Claude Code for programming tasks cannot currently access Cowork Projects from the CLI — though Anthropic has confirmed that integration is planned.
Windows support is limited to x64 architectures. ARM-based Windows devices are not supported at launch.
Despite these constraints, the foundation is solid. Memory, instructions and scheduled tasks already cover the core needs for most structured workflows. As cloud sync, sharing and Claude Code integration arrive, Cowork Projects will become significantly more powerful for team-wide adoption.
What Cowork Projects mean for enterprise AI adoption
Cowork Projects represent a shift in how businesses can use Claude. Until now, even with Cowork, every interaction was essentially a standalone conversation — powerful, but disconnected. Projects add persistence, structure and automation to the equation.
For enterprises evaluating AI adoption, this changes the conversation. The question moves from "can Claude handle this task" to "can Claude own this workflow." A well-configured project with the right context, instructions and scheduled tasks can handle recurring workloads with minimal human intervention.
The local-first architecture also matters for regulated industries. Files never leave the device, memory is scoped and deletable, and instructions enforce compliance rules consistently. This makes Cowork Projects a viable option for sectors like finance, legal and healthcare where data governance is non-negotiable.
The lack of sharing and cloud sync will slow enterprise rollouts in the short term. But for individual power users and small teams that can work from a shared machine or standardize their setups, Cowork Projects already deliver meaningful productivity gains.
As Anthropic continues to build out the platform — with Dispatch for remote control, Marketplace for third-party integrations, and soon Claude Code for developer workflows — Cowork Projects are shaping up to be the central hub around which enterprise AI usage is organized.