Projects and Memory in Claude: Make It Remember Your Work
Every new chat with Claude starts almost from zero. Projects give it a standing workspace with documents and instructions, and memory keeps details about you between chats. Here is what each one does, when to use it, and how to avoid the traps.
01First: what is context?
Context is everything Claude can see when it answers: your message, the conversation so far, attached files and standing instructions. If it is not in context, Claude does not know it.
In a normal chat, that context is gone when you open a new one. So an engineer who keeps re-explaining the same project, the same standards and the same report format is wasting time.
Projects and memory are two ways to put fixed information into context automatically, without pasting it every time.
02A binder on the desk vs a colleague who knows you
A Project is like a project binder on the desk: drawings, specs and work instructions, and whoever sits at that desk works by it. Memory is like a long-time colleague who knows you: they know your company name, your units and the style you like, without opening any binder. The binder belongs to one job; the colleague is with you in every chat.
03How it works in every chat
- 1You ask a questionInside a Project or in a regular chat
- 2Claude gathers contextProject instructions, knowledge files and memory summary
- 3It all goes into contextTogether with the current conversation
- 4A tailored answerIn your language, units and terminology
04What a Project is made of
Project knowledge
Files you upload once: specs, drawing lists, price lists, PDFs of internal standards. Claude uses them in every chat in the project.
Project instructions
Standing text that sets how to answer, for example: reply in Hebrew, use Israeli standard terminology, end each answer with a list of assumptions.
Chats that share context
Every chat started inside the project sees the same files and instructions.
Separate memory per project
Each Project has its own memory and summary, kept apart from other projects and regular chats.
Team sharing
On Team and Enterprise plans you can share a project with colleagues with view or edit permission.
05Memory: what it keeps and how you control it
What is kept
Work context: your role, projects, working style and technical preferences. The summary usually updates within 24 hours of your chats.
View and edit
In Settings, under Capabilities, there is a View and edit memory option. You can also just tell Claude in a chat what to remember or change.
Pause or reset
Pause memory stops using and creating memories but keeps what exists. Reset memory deletes everything permanently, including project memories.
Incognito chat
An Incognito chat (ghost icon on a new chat) is not saved to history and is not added to memory.
Availability
According to the help center, memory is available on all plans, including free. Searching past chats is on paid plans.
06A project for each construction site
07A project for writing quotes
08Memory that saves explanations
09Which one for which need
| Need | Right tool | Why |
|---|---|---|
| Same documents in every chat about one site | Project | Files and instructions load automatically |
| Claude remembers your company name and units | Memory | Follows you across all chats |
| A repeatable process, like a site visit report in a fixed format | Skill | A reusable work recipe, see Procedures for AI |
| Access to email, Drive or an outside system | Connector | Brings in live data, see Claude meets your tools |
| A team working on the same client | Shared Project (Team or Enterprise) | Everyone works from the same knowledge and instructions |
| A sensitive question that should not be saved | Incognito chat | Not saved to history or memory |
| A handy tool to show or calculate something | Artifact | An interactive page or tool, see Outputs, not just answers |
10Create your first Project in 5 minutes
- 1
Open Projects
In the claude.ai sidebar choose Projects and click New Project.
- 2
Give it a name and description
For example: Herzl Street residential tower. Name and description are for you; Claude does not rely on them.
- 3
Upload knowledge
In the Project knowledge area click + and upload a spec, drawing list or price list.
- 4
Write instructions
Click to set project instructions and say how to answer: language, units, format, what to do when information is missing.
- 5
Start a chat and test
Ask something whose answer is in a file, and check that Claude actually uses it.
- 6
Share if needed
On Team or Enterprise plans you can share with colleagues with view or edit permission.
11Pitfalls and safety
Do
- Update and delete old files when a new drawing or price list revision comes out
- Put the revision and date in the file name
- Check now and then what memory has kept, and correct it
- Use Incognito for sensitive questions
- Check company policy before uploading client documents
Don't
- Upload confidential client data without permission
- Keep both the old and the new revision in the project
- Trust an answer without checking the source document
- Assume memory keeps every detail from every chat
- Share a project with the whole organization when it holds sensitive data
12FAQ
Are Projects free?
Yes. According to Claude's help center, free plan users can create up to five projects. On paid plans Claude can automatically expand knowledge capacity when it gets close to the context limit.
How do I turn memory off?
In Settings, under Capabilities. Choose Pause memory to stop it and keep what exists, or Reset memory to delete everything permanently. For a single chat, use Incognito.
What is the difference between project instructions and memory?
You write project instructions, and they apply only in that project. Memory is built from your chats and follows you everywhere, with a separate memory for each project.
Do ChatGPT and Gemini have something similar?
Yes, in general terms. ChatGPT has Projects and memory, and Gemini has Gems with standing instructions. Names, settings and limits differ, so check each tool's help center.
Should I upload full standards?
Only if your license allows it. It is better to upload internal summaries, working tables and checklists, and always check the source before an engineering decision.