Claude Projects: What They Are, How to Use Them, and Whether Claude Remembers (2026)

Anthropic shipped Projects on June 25, 2024, six months before ChatGPT had its own version, and gave each one a 200K context window, about a 500-page book's worth of room. Most guides still describe that 2024 version, where you upload some documents and every new chat inside the project starts cold. That part is now wrong. Claude's memory reached every plan in March 2026, and each project keeps its own.

This guide covers Projects in the Claude app, the workspaces you see at claude.ai/projects. It isn't Claude Artifacts, the side panel where Claude renders code and apps, and it isn't the project context Claude Code reads from a CLAUDE.md file in a repo. If you're weighing it against the OpenAI equivalent, our ChatGPT Projects guide covers that side, and there's a comparison below.

What are Claude Projects?

Claude Projects are self-contained workspaces, each with its own chat history and knowledge base, that keep the documents and instructions for one body of work together instead of scattered across chats. Rather than re-pasting the same brief, style guide, or codebase into every conversation, you load it into the project once and every chat inside that project can see it.

The knowledge base is the heart of it. You add documents, text, or code, and Claude treats that material as background for every chat in the project. Each project gets a 200K context window, which Anthropic pitched at launch as roughly a 500-page book. Project instructions sit alongside it and shape how Claude behaves inside that workspace, telling it to answer as a specific role, hold a particular tone, or follow your house conventions.

One detail that gets missed: project instructions aren't a tone setting. They work as a standing pre-prompt applied to every chat in the project, which makes them the highest-leverage field in the whole feature.

How do you create a project and add knowledge?

You create a project at claude.ai/projects by clicking New Project and giving it a name and description. From there you fill the project knowledge with the files that matter, set the instructions, and start chatting. Existing conversations can be moved into a project, and projects can be starred, archived, or deleted later.

Two practical numbers shape how you load it. A chat supports up to 20 files, at 30 MB each, so a single oversized PDF eats a lot of room. Cleaning documents before uploading them, stripping boilerplate and duplicate pages, buys you noticeably more useful context than dumping raw exports in.

Project content is cached, and it doesn't count against your per-message usage limits. In a year when Anthropic has changed Claude's usage limits three separate times since August 2025, that's a reason to put stable background into a project rather than pasting it into each new chat.

Does Claude remember past chats in a project?

Yes, and this is the part most guides still get wrong. Claude's memory became available on every plan, free included, in March 2026, and projects were given their own. In Anthropic's own wording, each project has a separate memory space and a dedicated project summary, so the context inside a project stays focused and separate from other projects or non-project chats.

The isolation is deliberate rather than a limitation. Your global memory profile doesn't apply inside a project chat, and what Claude learns in your client-work project doesn't leak into your personal one. If you've kept work and personal contexts apart on purpose, projects are how that separation actually holds.

What memory doesn't do is worth being precise about. It synthesises, roughly every 24 hours, rather than storing conversations word for word, so it holds patterns, preferences, and recurring decisions instead of a transcript. Ask it for the exact service fee you discussed last Tuesday and it may not have kept that. Retrieving a specific detail from an earlier chat is what chat search is for, and search is on paid plans only, also scoped to within the project you're in.

What do people actually use Claude Projects for?

The projects that earn their keep share a shape: one ongoing body of work with reference material that doesn't change much, where you'd otherwise re-explain the background every time. A client engagement, with the contract, the brief, and the notes loaded once, so you can ask "what's the agreed monthly fee?" and get an answer out of the actual document. A newsletter or blog, with your past performance numbers and style guide sitting in the knowledge base. A codebase or API you're learning, turned into something closer to a tutor that already knows your setup.

The cleverest pattern I came across while researching this is a project built to write prompts. Greg Olsen keeps a dedicated "Prompter Project" whose instructions turn Claude into a prompt engineer, then uses it to draft the detailed prompts he'd struggle to write from scratch for everything else. It costs one of your five free project slots and pays for itself quickly.

A rough threshold: if you're having more than three or four conversations on the same topic, that topic wants a project.

What Claude Projects can't do

Three gaps catch people out, and none of them are obvious until they bite.

There's no export-project button. Everything you build up inside a project, the knowledge, the instructions, the conversations, lives there with no one-click way out, so any backup is a manual job you have to design yourself from the start.

Projects also don't come with you when you upgrade from a personal plan to a Team plan, because that move means a new account. The projects, their conversations, and their settings stay behind. Anyone planning to graduate a personal workflow into a team one should know that before they build six months of context in the wrong place.

The free tier has a harder edge than the marketing suggests. You get projects and memory, but not retrieval and not chat search, which are the two things that make a large knowledge base usable. Sharing a project needs Team or Enterprise; there's no way to hand one to a colleague on a personal plan.

Is Claude Projects free, and what are the limits?

Projects work on the free plan, capped at five. Anthropic's help centre states plainly that Projects are available to all users including free accounts, with a maximum of five projects on free. Worth knowing if you go looking: the feature list on Anthropic's pricing page reads as though Projects start at Pro, so the help centre is the more specific source here.

Retrieval is the paid feature that matters most. As a project's files approach the model's context limit, Claude switches to retrieval on its own, pulling only the passages a question needs instead of loading everything, which lifts effective capacity by up to about 10x with no setup from you. Prices are Anthropic's listed USD rates and change often, so check the pricing page before you commit.

Claude Projects vs ChatGPT Projects

Both do the same core job: a container that holds files, instructions, and its own memory for one body of work. The differences are in what's capped and what costs money.

The honest read: Claude gives you more room per project and a cleaner memory separation, while ChatGPT Projects is more generous on the free tier and doesn't charge you to search your own history. Claude's cap counts projects; ChatGPT's counts files. If you run a handful of deep, document-heavy workspaces, Claude's model suits you. If you want many small ones, five projects gets tight fast. Our ChatGPT vs Claude comparison covers the wider picture, and other AI tools puts both in context.

What this post does not cover

This is a plain-English guide to Projects in the Claude app, not a manual for Claude Code's repo context, the Anthropic API, or enterprise admin controls. It isn't a security or compliance review, so check your own policy before loading company documents into a project. Plans, caps, and which features are free move often, and Anthropic's pricing page and help centre don't always agree, so confirm the current details in Claude before you rely on a number here. And Claude can be confidently wrong about your own documents, so check what it tells you against the source file.

Sources

1. Claude Help: What are projects?
2. Claude Help: How can I create and manage projects?
3. Claude Help: Use Claude's chat search and memory to build on previous context
4. Anthropic: Collaborate with Claude on Projects
5. Claude pricing