Claude Projects vs Memory: Which One Should You Use?
What claude.ai Projects hold, how project memory works next to account memory, and when to use each one.
Claude Projects and Claude memory both carry context into new chats, but they come from opposite directions. A Project holds what you put in it on purpose, while memory holds what Claude picks up on its own as you talk.
This post is about Projects in claude.ai and Claude Desktop, the chat workspaces with knowledge and instructions. Claude Code has its own memory and its own Projects, covered in a short section near the end.
What is a Claude Project?
A Claude Project is a self-contained workspace with its own chat history, its own knowledge base and its own instructions. Every chat you start inside it begins with that material loaded, so you set up a client, a product or a manuscript once instead of re-pasting it into each conversation.
A Project has three parts you control, per Anthropic's Projects guide:
- Project knowledge: documents, text, code and other files you upload. Claude uses them as background in every chat in the Project.
- Project instructions: standing rules for that Project only, such as "answer as a tax adviser for UK freelancers" or "use the client's house style".
- Project memory: what Claude learns from the Project's chats. More on that below.
Projects are available on every plan, and free accounts can create up to five (Anthropic, 2026). When Anthropic launched Projects on June 25, 2024, each one came with a 200K-token context window, "the equivalent of a 500-page book" (Anthropic, 2024). Paid plans now go further: when project knowledge approaches the context window limit, Claude switches to retrieval (RAG) and searches the files instead of loading them all, which expands capacity by up to 10x (Anthropic, 2026). On Team and Enterprise plans you can share a Project with colleagues as "Can view" or "Can edit".
To create one, open Projects in the sidebar (or go to claude.ai/projects), click + New Project, then add files with + and write your rules under Set project instructions. The name and description are for you; Anthropic notes Claude can't see them.
One detail trips people up. Anthropic says context "is not shared across chats within a project unless the information is added into the project knowledge base." A decision you reach in one project conversation doesn't automatically land in the next one, unless memory picks it up or you add it to the knowledge.
What is Claude memory?
Claude memory is a set of topics Claude saves about you as you chat and reads at the start of new conversations. It's on by default for Free, Pro and Max, and you can read, edit or delete every topic under Settings > Your Claude > Manage next to "Your memory" (Settings > Memory in the older settings layout) (Anthropic, 2026).
You don't write memory the way you write project instructions. Claude decides what's worth keeping, or you say "remember this". The full walkthrough, including import from ChatGPT, is in the Claude memory guide.
How does project memory relate to account memory?
Each Project has its own memory, and inside a Project Claude uses that Project's memory together with your account memory, but never memory from another Project. Your account memory is what Claude learned in chats outside Projects.
Here's how the pieces interact, per Anthropic's memory guide:
- Inside Project A: Claude reads Project A's memory plus your account memory. Project B's memory stays out.
- New memories from Project A are saved to Project A. If memory is on, Claude can also reference them in chats outside Projects.
- Full isolation: turn off Use account memory under Memory on the Project's page, and the Project's memory stays fully separate from the rest of your account.
- One chat with no memory: click + in the message box and turn off Memory before your first message. That chat uses neither your account memory nor the Project's.
- Chat search (paid plans) works the same way: inside a Project it only searches that Project's chats.
An example makes the walls clearer. You run a Project for a client, Acme, and mention in an Acme chat that their launch moved to March. Claude saves that to the Acme Project's memory. A chat in your separate Globex Project never sees it. A plain chat outside Projects can, unless you switched off "Use account memory" for Acme.
Deletion follows the same split. Deleting a Project deletes its memory, and the Reset memory option in settings permanently deletes all memories, Project memories included. Steps for both are in how to delete ChatGPT and Claude memory.
Claude Projects vs memory: how do they compare?
Projects are written by you, scoped to one body of work, and only change when you edit them. Memory is written by Claude, follows you everywhere, and updates as you talk.
| Layer | Who writes it | Where it applies | Stays current? | Plans |
|---|---|---|---|---|
| Project knowledge | You upload files | Chats in that Project | Only when you re-upload | All; RAG on paid plans |
| Project instructions | You | Chats in that Project | Only when you edit them | All; Free is capped at 5 Projects |
| Project memory | Claude, from that Project's chats | That Project (and outside Projects unless isolated) | Yes, as you chat | All plans with memory on |
| Account memory | Claude, from your chats | Every chat; inside Projects too | Yes, as you chat | On by default for Free, Pro, Max |
| Instructions for Claude | You | Every chat | Only when you edit them | All |
| Unabyss | Built from Gmail, Calendar, Notion, Slack, GitHub and more | Claude, ChatGPT, Cursor and any MCP client | Yes, from your apps | 7-day trial, then from $15/month |
Sources: Anthropic's guides to Projects, memory and personalization; Unabyss pricing checked October 2026.
The account-wide Instructions for Claude field is the third piece people mix up with these two. It's written by you and applies everywhere; how to fill it in is covered in Claude preferences.
When should you use a Project instead of memory?
Use a Project when the work has its own documents and rules; rely on memory for facts about you that are true in every chat. Most people need both.
A Project is the better choice when:
- The work comes with files. A contract, a style guide, a codebase export, a 40-page research brief. Memory doesn't store documents; project knowledge does.
- The rules only apply here. "Cite the client's 2025 annual report" belongs in one Project, not in every chat you start.
- It has to stay separated. Client work, a confidential deal, a side project you don't want colouring your day job. Turn off "Use account memory" and it's sealed.
- You share it. On Team and Enterprise, a shared Project gives colleagues the same files and instructions.
Memory is the better choice when:
- It's about you, not a task. Your role, your writing style, the tools you use, the people you work with.
- You don't want upkeep. Memory fills itself from your chats. A Project only knows what you last uploaded.
- The context is small and spread out. A deadline you mentioned, a preference you corrected. Nobody writes a document for that.
A setup that works: put who you are in Instructions for Claude and let memory fill in the details, then make one Project per client or long-running job with its files and rules. If you're re-explaining yourself inside every Project, that belongs in account memory. If Claude keeps applying one client's rules to another, that client needs its own isolated Project. More layouts in how to make Claude keep context between chats.
Are Claude Code projects the same thing?
No. Claude Code keeps memory in CLAUDE.md files you write and in auto memory, notes Claude writes about your repository, both loaded at the start of every session (Claude Code Docs). It also has a separate Projects feature in beta on Pro and Max, where one ongoing conversation runs work as parallel threads, mostly in the cloud (Claude Code Docs). Anthropic has started moving existing Projects for some Pro and Max users to that new version, says nothing is deleted in the move, and keeps Projects in chat and Cowork on the current version for now (Anthropic, 2026). For the file-based side, see what CLAUDE.md is.
What do Projects and memory both miss?
Both live only in Claude, and Projects only know what you last uploaded. Open ChatGPT or Cursor and none of it is there; change your pricing or switch roles, and the PDF in your project knowledge is still last quarter's version.
That's where personal context differs from memory. Unabyss builds it from the apps where your work already happens (Gmail, Google Calendar, Notion, Slack, GitHub, Linear), keeps it current as those apps change, and serves the same version to Claude, ChatGPT and Cursor over MCP. What that structured context did to cost, time and output quality in a benchmark is in What Structured AI Context Is Worth.
Keep using Projects for files and memory for continuity inside Claude. When the same project brief has to reach three different assistants, see how Unabyss works.
