---
title: Claude memory not working? A checklist for the apps and Claude Code
description: Claude memory not working? Check pause, incognito, Projects, plan and admin settings, then CLAUDE.md and auto memory in Claude Code, with the fix for each.
date: 2026-11-18
updated: 2026-11-18
author: The Esment Team
category: Memory fixes
tags: claude, claude-code, memory, troubleshooting
canonical: https://esment.notas.ai/blog/claude-memory-not-working
---

# Claude memory not working? A checklist for the apps and Claude Code

*The Esment Team · 2026-11-18 · [HTML version](https://esment.notas.ai/blog/claude-memory-not-working)*

## In short

- Most "Claude memory not working" cases are a switch, a chat type or a Project boundary. Check Settings › Your Claude first, then ask whether the chat was incognito or memory-off.
- Memory and chat search are separate features. Anthropic lists chat search for paid plans, not Free, so a Free account can have one without the other.
- In Claude Code, run /context all and look under "Memory files" to see whether CLAUDE.md loaded. Auto memory loads only the first 200 lines or 25KB of MEMORY.md.
- Some gaps are by design: incognito chats, sensitive topics and financial account numbers are not saved.

If Claude memory is not working, the cause is usually a switch, a chat type or a boundary you didn't know was there, not a bug. **Work through the checks below in order.** Each one says how to recognize the problem and what to change, first in the Claude apps and then in Claude Code.

This is a troubleshooting list, not an explanation of how memory works. For the concepts, read [how Claude's memory and chat search work](https://esment.notas.ai/learn/does-claude-remember-across-chats) and [CLAUDE.md versus a memory server](https://esment.notas.ai/learn/claude-md-vs-memory-mcp). Product details are as of November 2026, from Anthropic's [memory help article](https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context) and [Claude Code memory docs](https://code.claude.com/docs/en/memory). Neither page shows an absolute "last updated" date, and menu labels shift, so match them against your own screen.

## Why Claude memory is not working: which memory is failing?

"Claude memory" can mean four different things, and each fails in its own way. Name yours before you change anything.

| Mechanism | Who writes it | Where to look | Typical failure |
|---|---|---|---|
| Memory in the apps | Claude, as topics while you chat | Settings › Your Claude › Manage | Paused, off, wrong chat type, wrong Project |
| Chat search | Nobody; Claude searches old chats | Settings › Your Claude | Off, or not on your plan |
| CLAUDE.md in Claude Code | You | `/context all`, `/memory` | Not loaded, too long, conflicting |
| Auto memory in Claude Code | Claude, as files on your machine | `/memory` | Over the load cap, other repo or machine |

**Figure.** In the apps, memory has to be allowed at every layer. The checks below go roughly from the bottom layer up.

- Org admin (Team, Enterprise)
- Your plan and app version
- Your two settings
- Chat type: incognito, memory off
- Project scope
## Check the chat you are in

Start with the cheapest explanation: this particular chat was never going to remember anything.

- **Incognito.** Anthropic's help article says incognito chats (the ghost icon at the upper right of a new chat outside a project) are not saved to history or memory, and Claude does not search them. Incognito is available on all plans, Free included. Fix: start a normal chat.
- **Memory switched off for one chat.** On web and desktop you can turn memory off from the "+" menu before the first message, and a crossed-out memory icon appears next to the title. On mobile the switch locks after you send. Anthropic says this cannot be changed later. These chats stay in your history and can still be searched from other chats. Fix: start a new chat with memory on.
- **Voice mode.** Anthropic says to turn memory off before tapping voice, which implies the setting is read at the start. If a voice chat remembered or forgot unexpectedly, check how the chat began.

## Check your two settings

Memory and chat search are separate features with separate switches. Open **Settings › Your Claude**. The older layout used **Settings › Memory**, so look for these labels either way.

> [!STEPS]
> 1. Open **Settings › Your Claude**.
> 2. Check that **Generate memory from chats** is on, not paused.
> 3. Check that **Search and reference chats** is on.
> 4. Select **Manage** next to "Your memory" and look for the topic you expect.

Paused is not the same as off. Per Anthropic, pausing keeps existing memories but Claude neither uses them nor creates new ones, and conversations held while paused are not added later when you unpause. If you want health, religion, politics or similar topics saved, turn on **Include sensitive topics in memory** as well. Anthropic says government ID numbers, criminal history, financial account numbers and immigration status are never saved, even on request.

## Check your plan and app version

If a switch is missing, the cause may be availability. As of November 2026, Anthropic's article says memory is on by default for Free, Pro, Max and Ultra on web, desktop and mobile, and that mobile needs the latest app version. It lists chat search for Pro, Max, Ultra, Team and Enterprise on those same surfaces. Free is not on that list, so we treat chat search on Free as not documented rather than confirmed missing.

Two more limits from the same page. Enterprise organizations that use customer-managed encryption keys cannot use past-chat search. And Cowork shares memory only when it runs in the cloud, so local Cowork sessions do not use it. Fix: confirm your plan, update the app, and look again.

## Look at the Project boundary

Projects are the most common reason Claude knows you in one place and not another. Per Anthropic, each project has its own memory. Inside a project, Claude uses that project's memory plus your account memory from chats outside any project, but never another project's memory. Chat search follows the same rule: it covers chats outside projects, plus the conversations inside each individual project, and does not cross between projects.

Recognize it by facts that appear in ordinary chats but not inside one project, or the reverse. To isolate a project on purpose, the project page has a "Use account memory" switch. Check that it is set the way you intended. A memory-off chat inside a project ignores both account and project memory. For the same idea in another assistant, see [ChatGPT memory not working](https://esment.notas.ai/blog/chatgpt-memory-not-working), and for how projects handle memory in both products, [how memory works in ChatGPT Projects and Claude Projects](https://esment.notas.ai/blog/chatgpt-claude-projects-memory).

## If you are on Team or Enterprise, ask your admin

On an organization plan, owners control memory under **Organization settings › Capabilities**, and one default changed this year. Anthropic's article says organizations created on or after October 31, 2026 (Pacific Time) start with memory on for each member, unless a member had turned it off earlier with the same account. Older organizations start with it off. The creation date counts, not the date of an upgrade. Memory also does not start on in HIPAA-enabled organizations, government-platform organizations or Claude for Teachers organizations.

Owners can turn memory off for everyone, but cannot turn it on for individual users and cannot view or edit members' memories. While it is off, members cannot see or delete their entries.

> [!NOTE]
> Sources disagree on what turning org memory off does to existing entries. The current help article says it stops use and saving but does not delete entries. A secondary search summary said it deletes them. We follow Anthropic's page, but ask before anyone flips the switch.

## Check what is actually saved

Before assuming nothing was saved, look. In **Manage**, entries are listed under Topics, and each one links back to the chat it came from. You can edit or delete a topic there, or tell Claude in chat to remember, change or forget something.

One trap: Anthropic says deleting a conversation does not delete the memory entries generated from it. If Claude keeps repeating something you thought you erased, delete the entry itself. **Reset** deletes all memories, project memories included, and cannot be undone, so use it last.

If you imported memory from another assistant and it looks thin, that is expected. Anthropic describes import as experimental and work-focused, so unrelated personal details may be dropped. Review the result under Manage and add what is missing by hand. In the legacy layout, imports appear as "memory edits" and updates can take up to 24 hours.

## Claude Code: CLAUDE.md not loading or ignored

Claude Code has no switch like the apps do. It reads files. Per Anthropic's docs, CLAUDE.md is loaded at the start of each conversation and treated as context, not enforced configuration. The first question is whether it loaded at all.

**Figure.** Confirm the file is in the Memory files list before you rewrite it.

CLAUDE.md on disk → /context all → Instruction followed?
> [!STEPS]
> 1. Run `/context all` and look under **Memory files** for your CLAUDE.md.
> 2. If it is missing, check the location: `./CLAUDE.md` or `./.claude/CLAUDE.md` for the project, `./CLAUDE.local.md` for personal notes, `~/.claude/CLAUDE.md` for yourself.
> 3. If it is listed but ignored, look for vague or conflicting instructions, and for a file over 200 lines, which Anthropic says may reduce adherence.
> 4. Run `/memory` to open and edit the files Claude sees.

Some details that catch people. Files in directories above your working directory load at launch, while a CLAUDE.md in a subdirectory loads on demand when Claude reads a file there, so it will not appear in the launch list. Files over 4 MiB are skipped. After `/compact`, the project-root CLAUDE.md is re-read from disk, but instructions you only gave in conversation are lost. If a rule must hold every time, Anthropic points to a PreToolUse hook rather than a CLAUDE.md line. For what fills the context and what survives compaction, see [Claude Code's context explained](https://esment.notas.ai/learn/claude-code-context).

## Claude Code: auto memory and AGENTS.md

Auto memory is the other half: notes Claude writes for itself. Anthropic says it is on by default in local sessions, off by default in self-hosted environment sessions, and can be toggled in `/memory`. Setting `CLAUDE_CODE_DISABLE_AUTO_MEMORY=1` disables it, and it cannot be turned on from background sessions.

When it seems to forget, three limits explain most cases:

- **Load cap.** Only the first 200 lines or 25KB of MEMORY.md (whichever comes first) loads at session start. Topic files are read on demand, not at startup.
- **Per repo, per machine.** Memory lives in `~/.claude/projects/<project>/memory/`, keyed to the git repository, and is not shared across machines. A different repo or a laptop has a different directory.
- **Subagents.** The main conversation's auto memory is not loaded into subagents, except forks.

For AGENTS.md, Claude Code reads it directly only when no CLAUDE.md or CLAUDE.local.md exists in the working directory or above. Anthropic's fix is a CLAUDE.md that contains `@AGENTS.md`, or setting "Project instructions" in `/config`. It needs version 2.1.277 or later, and some sessions before 2.1.281 (Bedrock, telemetry disabled) could not read it. More on keeping one file for several agents in [AGENTS.md vs CLAUDE.md](https://esment.notas.ai/blog/agents-md), and on adding your own memory with hooks in [Claude Code hooks for memory](https://esment.notas.ai/blog/claude-code-hooks-memory).

## The whole list in one table

| Symptom | Likely cause | Where to fix it |
|---|---|---|
| Claude forgets everything | Memory paused or off | Settings › Your Claude › Generate memory from chats |
| One chat remembers nothing | Incognito or memory-off chat | Start a new chat |
| Can't find an old chat | Chat search off, or not on your plan | Search and reference chats; check plan |
| Works outside a Project, not inside | Project-scoped memory | Project page › Use account memory |
| Health or politics not saved | Sensitive topics off | Include sensitive topics in memory |
| Org members see no memory | Org setting, or org created before Oct 31, 2026 | Organization settings › Capabilities |
| Imported memory partly missing | Import is experimental and work-focused | Manage; add by hand |
| CLAUDE.md ignored | Not loaded, too long, conflicting | `/context all`, `/memory` |
| Auto memory not recalled | 200 lines or 25KB cap; other repo or machine | `/memory`, the project memory directory |
| AGENTS.md ignored | A CLAUDE.md exists, or version below 2.1.277 | `/config` › Project instructions |

## When it is by design

Some gaps are not faults. Incognito and memory-off chats leave no trace on purpose. Sensitive categories are opt-in. A few identifiers are never stored. Claude's memory is also per vendor: what it holds is not visible to ChatGPT or Cursor, and Claude Code's auto memory stays on one machine.

> [!ESMENT]
> If you use Claude next to ChatGPT or Cursor and want one memory you can edit that all of them read, that is a different job from fixing these switches. Esment is a memory kept outside any one assistant. It can [import your Claude memory](https://esment.notas.ai/docs/memories/capture) and be [connected to Claude](https://esment.notas.ai/docs/assistants/claude). See [how Esment works](https://esment.notas.ai/learn/how-esment-works).

For why one assistant's memory stays this fragile, read [why AI assistants need a memory they can't lose](https://esment.notas.ai/blog/why-ai-assistants-need-memory) and [what local-first memory means](https://esment.notas.ai/blog/local-first-ai-memory). To move what you know between assistants, see [sharing memory between ChatGPT and Claude](https://esment.notas.ai/learn/share-memory-chatgpt-claude).

## Questions, answered

### Why does Claude not remember my previous chats?

Usually one of four things. Memory is paused or off, the chat was incognito or had memory switched off, you are inside a Project (memory is scoped per project), or chat search is not available on your plan. Check Settings › Your Claude for "Generate memory from chats" and "Search and reference chats".

### Does Claude memory work on the free plan?

Anthropic's help center says memory is on by default for Free, Pro, Max and Ultra on web, desktop and mobile. Chat search is a separate feature that it lists for Pro, Max, Ultra, Team and Enterprise, not Free. So a Free account may get memory without searching past chats.

### Where can I see what Claude remembers about me?

Open Settings › Your Claude and select Manage next to "Your memory". Entries are listed under Topics. Select one to read it, use the edit icon to change it, or delete it. You can also tell Claude in a chat to remember, change or forget something, and it will do so.

### Why does Claude Code ignore my CLAUDE.md?

Claude Code treats CLAUDE.md as context, not enforced configuration. Run /context all and check "Memory files" to confirm it loaded. If it did, look for vague or conflicting instructions and files over 200 lines. If something must always happen, use a hook instead of an instruction.

### Why is my AGENTS.md not read by Claude Code?

By default Claude Code reads AGENTS.md only when no CLAUDE.md or CLAUDE.local.md exists in the working directory or above it. Add a CLAUDE.md that contains @AGENTS.md, or set "Project instructions" in /config. Anthropic's docs say this needs version 2.1.277 or later.

### Why were only some of my imported memories added?

Anthropic calls memory import experimental. Claude may not incorporate everything, and it focuses on work-related context, so unrelated personal details can be dropped. Check the result under Manage next to "Your memory" and add anything missing by hand. In the legacy layout, updates can take up to 24 hours.
