AI assistants with memory compared

An AI assistant with memory keeps useful facts between conversations, such as your role, how you like answers written, or the project you are on, so you do not retype them. That saves time and it also creates a new thing to manage: a growing file about you. This guide explains the two kinds of memory, compares how ChatGPT, Claude, Gemini and nerix handle it, and lists the controls worth checking before you rely on any of them for work.

Menus and plan rules change often. We read the vendors' help pages on 1 November 2026 where they were available and mark where we could not confirm a detail.

Two kinds of memory

Saved memories. The assistant keeps short facts, either because you asked ("remember I prefer short answers") or because it decided a fact was useful. You can usually list them and delete them.

Chat history reference. The assistant searches or draws on your earlier conversations when answering. This is broader and less visible than a list of facts.

Both can be helpful. The difference matters for privacy: a list of ten lines is easy to review, while a reference to everything you ever typed is not.

What to look for

Whatever product you use, these controls decide whether memory is safe at work.

1. You can see it. A full list, not a hint.
2. You can edit or delete one line. And delete all.
3. You can pause it. Existing memory stays, nothing new is saved.
4. There is a no memory mode. Incognito or temporary chats for sensitive questions.
5. You know where it came from. Learned in chat, learned from a file, or written by you.
6. Workspace owners can set a policy. Important for teams.
7. Sensitive topics are not saved by default.

The options at a glance

ChatGPT

OpenAI's memory help page describes two controls: saved memories, and reference to chat history. Memories can be managed under Settings, then Personalization, then Manage memories. Turning off saved memories also turns off chat history reference. OpenAI notes that deleting a saved memory does not remove mentions from past chats, so you may need to delete the source chat too. Temporary chats do not use or create memories. Which features your plan includes has varied, so check the live page.

Claude

Anthropic's help page says memory is on by default for Free, Pro, Max and Ultra plans, while on Team and Enterprise plans owners decide whether it is available and it is off for each member until they enable it. Claude saves memory as topics; you can read, edit or delete them under Settings, Your Claude, or ask in chat. Pause memory keeps existing memory but stops Claude using or adding to it, and Reset deletes everything. Each project has its own memory. Chat search covers past conversations on paid plans. Incognito chats are not saved. Sensitive topics such as health are not saved by default unless you opt in.

Gemini

Google describes past chats as personal context, which you can turn off in settings. The sources we found describe a temporary chat option that does not affect personalization, and noted that turning the setting off does not delete your history, which is managed separately in your activity settings. Menu names have changed more than once, such as Keep Activity, so look at the current app rather than older articles.

nerix

nerix keeps short lines about you and your work, such as your role or how you like answers written. Each line says where it came from: learned in a conversation, learned from a document, or written by you. Memory in Settings lists all of it. You can add a line, mark one Keep this so it survives a clear out, forget one, or forget everything except the kept lines (this cannot be undone). Turning off Remember things about me pauses it: nothing new is saved and what is saved stays. If the workspace owner set it for the whole workspace, only an owner or admin can pause it.

What it is not: memory is about you, not a record of the workspace. What nerix knows about your company comes from the documents you add to the library, which follow roles and access rules. And it is early in its life, so expect fewer options than the large assistants.

Memory and company data

Memory raises a business question the consumer help pages skip: whose data is it? Three habits keep you safe.

• Do not use a personal assistant account for client work. Memory then mixes your private and professional life, and a leaver takes it with them.
• Check training and retention terms. Whether your chats train a model depends on the product and plan. Read the privacy page for your plan. Our note on keeping your data in the EU covers location questions.
• Keep facts about people out of memory. Customer health, finances and employee issues belong in controlled documents, not in a memory line.

Also remember that an assistant with memory can be wrong about you. A stale line ("I work in sales") can colour answers for months. Review the list every few months and delete what no longer holds.

A ten minute check

Do this on whichever assistant you use:

1. Open the memory list and read every line aloud.
2. Delete anything confidential, outdated or about another person.
3. Add two or three lines that really help, such as your role and preferred answer length.
4. Find the switch that pauses memory, and find the temporary or incognito mode.
5. Ask the assistant "what do you remember about me" and compare its answer with the list.

The short version

Memory is useful when it is short, visible and under your control. Check that you can see it, delete it, pause it and talk without it. Use temporary or incognito chats for sensitive questions, keep other people's details out, and review the list every few months. Menus change often, so rely on the settings screen in front of you.