# AI agent memory: how to see and correct what your agents remember (2026)

> Source: https://myturnstone.ai/blog/ai-agent-memory
> By Aryan Vij, October 28, 2026

**AI agent memory is the notes an AI keeps about you between chats, and in most big apps you can now open it and fix it. ChatGPT and Claude let you see and edit it in Settings, and so does Gemini for some accounts. Some agents, like ChatGPT dots and Grok Bot, give you no memory screen at all.**

Memory is what lets an agent draft the right email without you explaining who the person is. It's also how an agent ends up repeating a price you dropped months ago. This guide explains how agent memory works, why seeing it matters, and how to view, edit and delete it in the main apps. Checked October 28, 2026.

## How does memory work in AI agents?

Most AI apps mix four kinds of memory.

1. **Chat history.** The app looks back at past conversations for context. ChatGPT calls this "Reference chat history." You don't see a list of facts.
2. **Saved memories.** Short facts the AI keeps, like "prefers short emails" or "works at Acme." This is the part most apps now show you in Settings.
3. **Project memory.** Memory tied to one project, so client work doesn't leak into personal chats. ChatGPT and Claude both offer this.
4. **Shared memory across agents.** Once you have several agents, do they share what they know? Some apps give each agent its own memory. Some share one. Some do both.

Agents also learn from your email, calendar and documents. OpenAI says a dot forms memories from connected apps "even when you haven't asked a specific question." So an agent can learn things you never typed.

## Why should you be able to see what your AI remembers?

Because memory goes stale, and stale memory sounds confident. Plans change on calls and prices get rethought. If the AI only learned the old version, it will repeat it in your voice. We wrote about this in [what happens when your AI remembers an old decision](https://myturnstone.ai/blog/what-happens-when-your-ai-remembers-an-old-decision).

Being able to see memory fixes three problems:

- **Stale facts.** You spot the old phone number, the old job title or the old price before an agent sends it to someone.
- **Wrong decisions repeated.** If a note says a project is on hold when it isn't, every agent that reads it acts on that.
- **Things it shouldn't know.** Agents that read your inbox can pick up details from the wrong account. You can only remove what you can see.

People also hand bigger jobs to an agent once they can check what it believes. That's the argument in [why we want you to see what your AI knows](https://myturnstone.ai/blog/why-we-want-you-to-see-what-your-ai-knows).

## ChatGPT memory vs Claude memory vs other AI agents

Here's how the main apps handle memory, from each company's own help pages.

| App | Can you see it? | Can you edit it? | Shared across agents? | Where it's stored |
|---|---|---|---|---|
| **Turnstone** (ours) | Yes. Plain files you can open | Yes. Open the file and correct it | Yes. One Brain every Agent uses, plus each Agent's own notes | Your Mac |
| ChatGPT and ChatGPT Work | Yes. Memory summary or saved memories in Settings | Yes. Delete saved memories, or correct one from Sources under a reply | Work uses the same ChatGPT memory | OpenAI's cloud |
| ChatGPT dots | No | No. Only Reset, which deletes the dot | Gets ChatGPT memory one way, keeps its own on top | OpenAI's cloud |
| Claude, Cowork and Duet | Yes. Topics in Settings > Memory | Yes. Read, edit or delete each topic | Chat and Cowork share one memory. Duet uses it and keeps its own | Anthropic's cloud |
| Gemini and Tasks in Gemini | Yes, on some accounts. Memory summary | By asking in chat. There's a memory history log | Tasks need Personal Intelligence, which includes memory | Google's cloud |
| Grok Bot | No memory screen. You ask in chat | Tell the Bot in chat | Each Bot has its own, except a few facts like name and time zone | The cloud |
| Muse | Partly. Files like MEMORY.md | Ask it to forget, "to the best of its ability" | One agent per person | A Meta-hosted cloud computer |
| Lindy | Yes. A versioned file system | Yes. Edit, see history, restore | Personal and team scopes | Lindy's cloud |
| Perplexity Computer | Yes. Brain, under Customize | You can remove memories | Across your sessions | Perplexity's cloud |
| OpenClaw | Yes. Plain Markdown files | Yes. Edit the files | One assistant | Your computer or server |
| Hermes Agent | Yes. Two small files | Yes, but capped at a few thousand characters | Docs warn against pointing two agents at one memory | Your computer or Hermes Cloud |
| Manus | No documented memory screen | Not documented | Its privacy policy mentions memory for "other Agents acting for you" | Manus's cloud |

Most apps let you see memory in some form. ChatGPT dots and Grok Bot don't. Only OpenClaw, Hermes and Turnstone keep it on your own machine.

## How do I see and edit ChatGPT's memory?

1. Go to **Settings > Personalization > Memory**.
2. Depending on your account, you'll see a **Memory summary** or a list of **Saved memories**. Choose **Manage** to review them.
3. Delete any saved memory you don't want. You can also just tell ChatGPT to forget something.
4. When a reply uses something personal, **Sources** may appear under it. From there you can correct the memory, delete it or mark it as not relevant.
5. To stop it learning from old chats, turn off **Reference chat history**.

OpenAI says the summary "may omit details." Deleted memories may stay in logs for up to 30 days. To fully remove something, delete the memory and the chat where you first shared it. Temporary chats don't create new memories, and projects can use project-only memory.

ChatGPT Work uses this same memory. [Turnstone vs ChatGPT Work](https://myturnstone.ai/comparisons/chatgpt-work)

ChatGPT dots are different. OpenAI says you "cannot view, delete, or directly modify individual dot memories." The only option is Reset, which deletes the dot, its chats, its memories and its scheduled tasks. Disconnecting an app doesn't remove what the dot already learned. [Turnstone vs ChatGPT dots](https://myturnstone.ai/comparisons/chatgpt-dots)

## How do I see and edit Claude's memory?

1. Go to **Settings > Memory**.
2. Your memory is a list of files under **Topics**. Open any one to read, edit or delete it.
3. You can pause memory or reset it at any time.

A correction in one topic applies to all later conversations. Since August 25, 2026, chat and Cowork share the same memory. Each Claude project has its own memory. To keep a chat out of memory, use an incognito chat. Memory from Cowork tasks that ran only on your computer stays with those tasks. Claude can also import memory from other AI apps. [Turnstone vs Claude Cowork](https://myturnstone.ai/comparisons/claude-cowork)

Claude Duet uses your account memory and keeps its own history and memory too. Anthropic says what Duet learned from an app "stays until you reset its memory," even after you disconnect the app. It doesn't document a way to edit Duet's own memory item by item. [Turnstone vs Claude Duet](https://myturnstone.ai/comparisons/claude-duet)

## How do I see and edit Gemini's memory?

1. On gemini.google.com, go to **Settings > Personal Intelligence > Memory summary**.
2. To change it, type a request. Google's examples are "Add to my memory that I speak Japanese" and "Remove my old phone number from your memory."
3. Click **Memory history** to see a dated record of changes.
4. Turn Memory off in the same Personal Intelligence menu to stop Gemini learning about you.

The summary isn't on every account yet. Google limits it to adults with a Google AI plan, using Gemini in English, so check your own settings. There's no delete button. Edits go through chat. Google also says asking Gemini to forget "doesn't always work perfectly." [Turnstone vs Gemini Spark](https://myturnstone.ai/comparisons/gemini-spark)

## How do I see what Grok Bot and other agents remember?

**Grok Bot.** xAI's docs say "There is no memory screen for a personal Bot. Ask in chat instead." A Bot decides on its own what to keep and "doesn't ask first." To change something, you tell it in chat. [Turnstone vs Grok Bot](https://myturnstone.ai/comparisons/grok-bot)

**Muse.** You can ask it directly or browse files like MEMORY.md under System Files. Removal goes through asking it to forget. Meta's privacy policy says Muse "may still 'remember' information it learned" after you delete something. [Turnstone vs Muse](https://myturnstone.ai/comparisons/muse)

**Lindy.** Its memory is "a versioned file system you can see into, edit, and share with your team," with history and restore. [Turnstone vs Lindy](https://myturnstone.ai/comparisons/lindy)

**OpenClaw and Hermes Agent.** Both keep memory as Markdown files on your computer. OpenClaw says "there is no hidden state." Hermes keeps two small files with hard size caps. [Turnstone vs OpenClaw](https://myturnstone.ai/comparisons/openclaw) and [Turnstone vs Hermes Agent](https://myturnstone.ai/comparisons/hermes-agent)

## What makes a good AI assistant with memory?

Look for memory you can see in full, not a summary that may leave things out. You should be able to fix one fact without wiping everything. If you use several agents, a correction should reach all of them. And it should be clear where the memory lives and where each fact came from.

## How Turnstone handles memory

We make Turnstone, so take this section with that in mind. Turnstone is a free, model-agnostic Grok Bot alternative that runs on the AI subscriptions you already have, including ChatGPT, Claude, Gemini and Grok.

Turnstone keeps one shared memory, called the Brain, of your people, projects and decisions. Every Agent uses it. It's stored as plain files on your Mac that you can open, read and correct. Each Agent also keeps its own notes on what it's doing. When you correct something in the Brain, every Agent uses the corrected version.

The catch: it's Mac only today, with no Windows, Linux or Android app. Turnstone Cloud, the iPhone app and the web app are coming soon. If you're comparing agents more broadly, see [the best AI agent apps](https://myturnstone.ai/blog/best-ai-agent-apps) and our [Grok Bot alternatives guide](https://myturnstone.ai/comparisons/best-grok-bot-alternatives).

## How to keep AI memory accurate

Whatever app you use, read your memory once a month, put dates on decisions, and ask the AI where it got a price or a date. After you disconnect an app, check what it learned, because in ChatGPT dots, Claude Duet and Muse that stays behind. Our post on [old decisions](https://myturnstone.ai/blog/what-happens-when-your-ai-remembers-an-old-decision) goes through these habits in more detail.

## Questions

**Can I see what ChatGPT remembers about me?**
Yes. Go to Settings > Personalization > Memory. You'll see a memory summary or a list of saved memories. OpenAI says the summary may omit details. ChatGPT dots are the exception, with no way to view their memories.

**What's the difference between ChatGPT memory and Claude memory?**
ChatGPT shows a summary or a list of saved memories, and can also learn from your chat history. Claude lists memory as topics you can read, edit or delete one by one. Both keep memory in the company's cloud, and both offer separate project memory.

**How do I delete AI memory?**
In ChatGPT, delete saved memories in Settings and delete the chat where you shared the fact. In Claude, delete the topic in Settings > Memory. In Gemini, ask it to remove the fact. Grok Bot has no memory screen, so you ask in chat.

**Do AI agents share memory with each other?**
It depends on the app. Claude chat and Cowork share one memory. Grok Bots each keep their own. A dot gets ChatGPT memory but keeps its own on top. In Turnstone, every Agent uses one shared Brain, and each Agent keeps its own notes too.

**Does disconnecting an app delete what the AI learned from it?**
Often not. OpenAI says this for dots, Anthropic for Duet, and Meta for Muse. Check the memory itself after you disconnect.

**Which AI assistant with memory keeps it on my computer?**
OpenClaw and Hermes Agent keep memory as files on your computer or server. Turnstone keeps the Brain as plain files on your Mac. ChatGPT, Claude, Gemini, Grok Bot, Muse, Lindy and Manus keep it in the cloud.

## Sources

- [Memory in ChatGPT, OpenAI Help](https://help.openai.com/en/articles/8590148-memory-faq) (checked October 28, 2026)
- [ChatGPT Work memories, OpenAI](https://learn.chatgpt.com/docs/customization/memories) (checked October 27, 2026)
- [Getting started with your dot, OpenAI Help](https://help.openai.com/en/articles/20001530-getting-started-with-your-dot) (checked October 27, 2026)
- [Dots privacy, security and safety FAQs, OpenAI Help](https://help.openai.com/en/articles/20001529-dots-privacy-security-and-safety-faqs) (checked October 27, 2026)
- [Claude's memory works everywhere, and you decide what's in it, Anthropic](https://claude.com/blog/claudes-memory-works-everywhere-and-you-decide-whats-in-it) (checked October 28, 2026)
- [Use Claude's chat search and memory, Claude Help](https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context) (checked October 28, 2026)
- [Use incognito chats, Claude Help](https://support.claude.com/en/articles/12260368-use-incognito-chats) (checked October 28, 2026)
- [Claude Cowork and chat are one Claude, Claude Help](https://support.claude.com/en/articles/16761823-claude-cowork-and-chat-are-one-claude) (checked October 27, 2026)
- [Import and export your memory from Claude, Claude Help](https://support.claude.com/en/articles/12123587-import-and-export-your-memory-from-claude) (checked October 27, 2026)
- [Claude Duet safety, privacy and security, Anthropic](https://claude.com/resources/articles/claude-duet-safety-privacy-and-security) (checked October 27, 2026)
- [Memory summary in Gemini Apps, Google Help](https://support.google.com/gemini/answer/18617127?hl=en&co=GENIE.Platform%3DDesktop) (checked October 28, 2026)
- [Instructions for Gemini, Google Help](https://support.google.com/gemini/answer/16598625?hl=en) (checked October 27, 2026)
- [Manage your tasks and workflows in Gemini Apps, Google Help](https://support.google.com/gemini/answer/17094507?hl=en&co=GENIE.Platform%3DAndroid) (checked October 27, 2026)
- [Bots and what a Bot remembers, xAI docs](https://docs.x.ai/grok-bot/bots) (checked October 27, 2026)
- [Muse privacy policy](https://muse.ai/privacy) (checked October 27, 2026)
- [Meta Help: Muse memory](https://www.meta.com/help/artificial-intelligence/1385290430137537/) (checked October 27, 2026)
- [Lindy docs: Memory](https://docs.lindy.ai/teammate/memory.md) (checked October 27, 2026)
- [Perplexity changelog: Brain](https://www.perplexity.ai/changelog/brain-faster-computer-models-website-publishing) (checked October 27, 2026)
- [OpenClaw memory docs (GitHub)](https://github.com/openclaw/openclaw/blob/main/docs/concepts/memory.md) (checked October 27, 2026)
- [Hermes docs: Persistent memory](https://hermes-agent.nousresearch.com/docs/user-guide/features/memory) (checked October 27, 2026)
- [Manus privacy policy](https://manus.im/privacy) (checked October 27, 2026)