Does Claude Have Memory? Chat History, Projects, and Limits

Short Answer

Claude has several ways to use context, but chat history, a saved memory feature, Projects and Claude Code project instructions are not the same thing. Which options you can use depends on the current product, account and rollout. Check the settings and official Anthropic help pages for the exact memory controls available to you.

Start with the official Claude website, Anthropic Support and the Claude Code documentation. Those pages are the appropriate places to confirm current behavior. They could not be reopened during this repository only run, so this article explains the concepts and safe checks rather than claiming a live feature audit.

What the Question Really Means

When someone asks whether Claude has memory, they may mean one of four things. They may want to reopen a previous chat. They may want Claude to remember a preference across separate conversations. They may want a Project to keep files and instructions together. Or they may mean Claude Code should follow repository rules while working on a codebase.

These needs have different solutions. Chat history helps you find previous conversations. A saved memory feature, when available, may carry selected context forward. A Project can group material for a continuing task. Claude Code can read project instructions such as CLAUDE.md to guide work in a repository. Treating these as identical can lead to missing context or accidental disclosure.

The first step is to name the job you want memory to do. If you need a record of a past decision, keep the original chat or a project note. If you need repeated preferences, review the memory controls in your account. If you need consistent code rules, use project instructions and version control. Each approach should have a clear owner and a way to correct stale information.

How It Works

A conversation has a context: the messages and materials available to the assistant while answering. A chat history view may let you return to a previous conversation, but that does not guarantee every old message is automatically loaded into a new one. A saved memory feature, where offered, is a separate mechanism that may select details for future use.

Projects are useful for ongoing work because they can keep related files, notes and instructions in one place. A project does not mean the assistant has unlimited recall or that every file is always used in every answer. Large projects still need clear organization, and users should point the assistant to the relevant source when accuracy matters.

Claude Code is different again. Project instruction files can describe how a coding assistant should work in a repository. For example, a file can tell the assistant which tests to run, which folders are sensitive and what coding style to follow. These instructions help shape work inside that project. They should not be used as a place to store credentials or private personal details.

Practical Steps

1. Check the product you are using. Claude on the web, mobile, and Claude Code may have different context and memory controls.
2. Review settings. Open the current account settings and read the available controls before changing anything. If memory is listed, inspect what it saves and how it can be reviewed or disabled.
3. Separate chat history from saved context. If you only need an old answer, search or reopen the original conversation. Do not assume it has been saved as a reusable memory.
4. Use Projects for continuing work. Keep the most relevant documents and project instructions together. Name files clearly and remove outdated guidance so the assistant does not follow conflicting instructions.
5. For coding, review CLAUDE.md. Keep repository instructions concise, scoped and version controlled. Add test commands and safety boundaries, not credentials.
6. Test what carries forward. In a non sensitive chat, ask Claude to restate the preference or project detail you expect it to know. Compare the response with the source of truth.
7. Correct stale information. If the account offers memory editing or deletion controls, use the documented method. If not, update the project note or provide fresh context explicitly.
8. Verify important decisions. Memory is a convenience, not proof. Check the original source before acting on a remembered price, policy, deadline or technical requirement.

Common Problems

Claude may appear not to remember something because the information was in another conversation, the project did not include the relevant file, a memory control is unavailable, or the detail was never saved. It may also summarize a detail incorrectly. A confident answer does not prove the memory is complete.

The opposite problem is stale context. A preference, product plan or technical instruction can change. If old information continues to appear, inspect the saved source and project instructions rather than repeating the correction only in chat. For code, check whether the repository has several instruction files that conflict or a project note that no longer matches the current code.

If a Project contains many files, use clear names and a short index. Tell Claude which document is authoritative for the task. For code repositories, use version control so changes to project instructions can be reviewed just like code. This makes it easier to identify when a bad answer came from outdated instructions rather than a model limitation.

What Does Not Work

Do not assume that opening an old conversation means Claude will remember it in a new chat. Do not assume a Project, chat history and saved memory share the same controls. Do not expect a coding instruction file to carry personal preferences across unrelated projects.

Avoid copying large amounts of private information into a memory note just because it might be convenient. Keep only what is needed for the task. Avoid passwords, access tokens, API keys, customer records and confidential data. If the task requires sensitive material, use an approved secure system and follow the organization's data policy.

Do not treat remembered content as a source citation. If Claude recalls a contract clause, a number or a product feature, verify it against the original document. A useful memory system reduces repeated work, but it cannot replace a source of truth.

Safety and Limitations

Memory can affect what an assistant emphasizes, so users should know what information may carry into future conversations. Review current settings and privacy guidance before using memory for personal or workplace data. If a shared account or organization workspace is involved, confirm who controls the settings and whether administrators apply additional rules.

This is a concept and safety guide, not a live test of every Claude account. Current features may differ by product, region or rollout. The official Anthropic pages above are primary starting points, but external pages were not reopened during this GitHub only run.

Key Takeaways

• Identify whether you need chat history, saved memory, Project context or Claude Code instructions.
• Keep memory short, useful and free of secrets.
• Verify important remembered facts against the original source.

Related Reading

• What is CLAUDE.md? Claude Code project memory explained
• How to use Claude Code in VS Code

FAQ

Does Claude remember all my old conversations?

Do not assume that it does. Chat history and any saved memory features have different purposes. Check current product settings and provide the relevant source when a new conversation needs a specific fact.

Can Claude remember writing preferences?

A memory feature may support preferences where it is available, but the exact behavior and controls can change. Review the settings and test with a non sensitive preference.

Are Claude Projects a memory system?

Projects can organize ongoing work and relevant context, but they should not be treated as unlimited recall. Keep the source documents clear and identify which one is authoritative.

What is Claude Code memory?

Claude Code can use project instruction files and other documented context mechanisms to guide work in a repository. These are different from a consumer chat memory feature.

Can memory contain secrets?

Avoid putting secrets into any memory or project file. Use environment variables or a secret manager for credentials, and rotate a key if it has been exposed.

What should I do when Claude remembers something wrong?

Check the original source, correct or update the stored note if the product provides a control, and explicitly provide current context. Do not rely on a remembered detail for a high impact decision.