· Updated · 9 min read · Omkar Satpute and Milind Soni

AI agent memory: who keeps it, and can you delete it?

Where OpenAI dots, Meta Muse, Grok Bot, Cue, Instinct, OpenClaw, Hermes, and Claude keep memory, whether you can delete one entry, and what trains the model.

Every personal AI agent now remembers you. Three questions decide whether that is fine: where the memory lives, whether you can read and delete a single entry, and whether it trains the model. The answers range from "a Markdown file you can open" to "delete the whole agent". OpenAI's dot has no per-memory view or delete at all, which is fewer controls than the ChatGPT it lives inside. Meta Muse gives you the file but warns it may still remember what you removed. Local agents such as OpenClaw, Hermes, and MausBot keep memory on your disk, where you can open it. The table below puts them side by side.

Three questions to ask any agent

An agent that reads your inbox, your calendar, and your files will build a picture of you fast. That picture is the memory. Before you hand one the keys, the help pages answer three things, or do not.

Where does it live? On a server the company runs, or in a file on your machine. This decides who can read it, who holds a copy after you leave, and whether a subpoena or a breach reaches it.

Can you read and delete one entry? Not the whole thing, one line: the home address it picked up from a delivery receipt, the salary figure from a PDF. Some products show you a list. Some show you a file. Some only offer a reset.

Does it train the model? Memory that trains is memory you cannot take back. Look for the toggle, the default, and whether the opt-out applies to what was already collected.

OpenAI dots

A dot receives your ChatGPT memories and then builds its own. The dots privacy FAQ is direct about the controls: "Your dot can retain context from your conversations and plugins for as long as you keep your dot." And: "You currently cannot view, delete, or directly modify individual dot memories, including specific details that enter the dot's context from plugins." The remedy is the big one: "Deleting your dot starts deletion of its own context." Disconnecting a plugin "stops new access through that connection. It does not delete information your dot has already built into its context." Turning Memory off in ChatGPT "does not delete information that your dot has already received."

Compare that with ChatGPT itself, which lets you view and delete each saved memory in Memory settings, edit one where those controls are available, or ask it to forget, with the note that OpenAI may retain logs of deleted saved memories for up to 30 days. The dot lives inside ChatGPT and has fewer memory controls than ChatGPT. Training on personal plans follows the "Improve the model for everyone" toggle and can include dot actions, delegated work, automations, and connected-app data; business plans are not trained on by default, and OpenAI says it does not train directly on proactive background research or its notes.

Meta Muse

Muse keeps memory in the per-user cloud VM that Meta's security post calls "the system of record", as a MEMORY.md file. The same post says: "You can inspect, edit and download these files freely, including Muse's memory about you." The help page says you can also "ask Muse to forget a topic or set of interactions", which Meta describes as best-effort, with this caveat verbatim: "After you've deleted something from Muse, Muse may still remember information that it learnt from what you deleted." Reset "deletes all your Muse data." Training is on by default, with a toggle under Settings, Data controls, and Meta says the data is not shared with its ad systems. Meta's access to the VM is limited by policy today, and Meta says users will be able to choose a Muse Confidential VM in the future, with no date given. Retention after you delete the account is not stated.

xAI Grok Bot

The Grok Bot FAQ says "A Bot can retain stable preferences, role context, and summaries of prior work," and adds that "Memory is not a substitute for an authoritative source." The bots page is plain about the controls: "There is no memory screen for a personal Bot. Ask in chat instead." You remove a memory the same way, as in "Forget that I'm on the launch team." Deleting a bot erases its profile, conversations, memory, routines, and attachments, and resetting a conversation keeps memory. The FAQ says deletion "removes the Bot's active profile, conversation, and routines" and that "files and logins on that computer may remain. Backend retention follows the applicable Cursor terms." Training opt-out "follows the applicable Cursor account and privacy settings." Each routine keeps its 20 most recent run records. A Cursor staff reply on the forum, August 26, reported there is no manual compaction and that bots auto-summarize near the context limit.

Cue by Manus

Cue is the thinnest documented. The Play listing says "Memory and Dream help Cue remember your food preferences, travel habits, learning progress, and the plans you want to stick with." Its privacy policy adds only that, where memory is enabled, it may keep what you share to support later tasks and other agents. No page explains where memory lives, how to view it, or how to delete one entry. The Manus account deletion page says "we will take steps to delete your content", with retention "as long as necessary", and deleting your account now deletes Cue and Manus together; the privacy policy says deleted data goes within 30 days unless it must be kept longer. On training, the policy says de-identified data may be used "to develop and improve our models", and Manus lists a data training opt-out as a Team plan feature.

Instinct

Most of this is reported; Instinct's terms and privacy policy, both revised August 26, cover the rest. TechCrunch (August 24) and Claire Vo on X (August 21) described revoking Instinct's Google access at 11 AM and receiving an inbox summary at 2 PM. Asked, the bot reportedly said emails were "stored in plain text for later searches". An early user said Instinct at first refused to delete his Gmail records and later added a deletion tool in settings. Its terms now say disconnecting does not delete indexed data; you request deletion at app.instinct.com/workspace, and deleting your account deletes the Google data it collected. The terms let Instinct train AI models on your materials by default, with an opt-out that applies going forward, and its privacy policy says it never trains on data received from Google Workspace. There is no retention schedule. Our Instinct comparison has the rest.

OpenClaw

OpenClaw 2026.8.1 shipped "visible workflows" under Settings, Memory that let you search indexed sources, inspect how memory is working, import supported history, and remove attributable derived memory. The CLI adds openclaw memory forget, a provenance-based purge by session, hook, or participant, with the caveat that "Source session transcripts are retained"; memory reset clears the index only. The stores are local files: an SQLite database per agent, plus MEMORY.md, USER.md, dated notes under memory/, and DREAMS.md in the workspace. Memory is local; your model provider's own setting covers what you send it.

Hermes Agent

Hermes keeps two files in ~/.hermes/memories/: MEMORY.md, capped at 2,200 characters, and USER.md, capped at 1,375, with sessions in an SQLite store at ~/.hermes/state.db. The caps are hard: "Memory does not auto-compact" and the agent "returns an error instead of silently dropping entries." To delete something, edit the file, remove one entry with hermes journey delete, or turn on write approval and use /memory pending, approve, and reject on new entries. Local files; your model provider's own setting covers what you send it.

Claude and Cowork

Anthropic's memory help page, as read on November 15, 2026, is the most complete of the cloud agents: Settings, Your Claude, then Manage next to Your memory (Settings, Memory in the older layout) lists each memory, and you can read, edit, or delete any one of them. Pause keeps everything in place; Reset is irreversible and deletes everything including project memories. Memory is per project where you want it. One caveat: "When a conversation is deleted, including under a retention period, memory entries generated from it aren't removed, but you can delete individual memories while memory is on." It is on by default on Free, Pro, Max, and Ultra. On Team and Enterprise, owners decide, and in most organizations created since October 31, 2026 it starts on. Cowork sessions that run locally on your computer do not use memory.

Side by side

Every cell comes from the company's own help page or docs as of November 15, 2026, except parts of the Instinct row, which are reported. "Not documented" means the company has not said.

ProductWhere it livesView a single memoryDelete a single memoryWhat reset doesTrains the model
OpenAI dotsOpenAI's servers, for as long as you keep your dotNoNoDeleting the dot starts deletion of its context; your files, Codex threads, ChatGPT conversations, and ChatGPT memories stayPersonal plans follow the Improve the model for everyone toggle; business plans not by default
ChatGPT itselfOpenAI's serversYes, in Memory settingsYes; logs of deleted memories may be kept up to 30 daysNot covered in this reviewFollows the same toggle
Meta MuseA per-user cloud VM, as MEMORY.mdYes: inspect, edit, downloadAsk Muse to forget, best-effort; it may still remember what it learned from the deleted itemDeletes all your Muse dataOn by default; toggle in Settings, Data controls
Grok BotGrok Bot's servers, under Cursor's account terms; stable preferences, important facts, and summariesNo memory screen; ask in chatAsk in chat to forget itDeleting a bot erases its profile, chats, memory, and routines; files and sign-ins on the shared computer can remain. Resetting a chat keeps memoryFollows your Cursor account and privacy settings
Cue by ManusNot documented; the privacy policy says memory may be kept for later tasksNot documentedNot documentedDeleting your account deletes Cue and Manus together, within 30 days unless the data must be kept longerDe-identified data may be used to improve models; a training opt-out is a Manus Team plan feature
InstinctA cloud computer; emails reportedly stored in plain text for later searchesNot documentedA deletion tool was added after an early user's request was refused, reported; deletion requests go through app.instinct.com/workspaceDeleting your account deletes the Google data it collectedTrains by default; the opt-out applies going forward; Google Workspace data is never used for training
OpenClawLocal files: an SQLite index plus MEMORY.md, USER.md, dated notes, DREAMS.mdYes: search and inspectmemory forget purges by session, hook, or participant; source transcripts are keptmemory reset clears the index onlyNo; memory stays local, and your model provider's own setting applies to what you send it
Hermes AgentLocal files: ~/.hermes/memories/MEMORY.md and USER.md, plus an SQLite session storeYes: open the fileEdit the file, remove one entry with hermes journey delete, or turn on write approval and reject pending entriesDelete the files yourselfNo; memory stays local, and your model provider's own setting applies to what you send it
Claude and CoworkAnthropic's servers, as topics; Cowork sessions that run locally do not use memoryYes, per topicYes, per topic; entries from a deleted conversation are not removed with itPause keeps it; Reset deletes everything including project memories, irreversiblyNot covered in this review
MausBot~/.openmausbot on the computer that runs it: Markdown notes per bot, an About me every bot reads, and the conversation recordYes: Bot Settings, Memory shows every note and how much loads each turnYes: edit or delete one note in the panel, or say "forget that" in chat; every change can be undoneSwitch memory off per bot, or empty the notes yourselfNo; memory stays local, and your model provider's own setting applies to what you send it

The pattern

Cloud agents give you a reset button. Local agents give you a file. That is the whole table in two sentences. Dots have no way to remove one entry, Grok Bot does it only by asking in chat, and Cue documents none. Muse and Claude are the exceptions among the cloud products, and Muse's own help page says the forget is best-effort. Hermes keeps memory as a file in a folder you already own, OpenClaw's delete is a command over local files, and MausBot puts its files behind a panel where you can edit or delete one note and undo the change.

The oddest row is the dot. It lives inside ChatGPT, inherits ChatGPT's memories, and then offers fewer controls than ChatGPT does: no list, no edit, no single delete, and no undo when you disconnect an app. If per-memory control matters to you, the product that has it today is the chat app, not the agent built on top of it.

How MausBot handles it

MausBot, published on GitHub as OpenMausBot, keeps memory as files in ~/.openmausbot on the computer that runs it. When this post first ran, that meant the conversation record alone, and we said there was no per-memory list to edit. That changed. Each bot now keeps Markdown notes, a MEMORY.md plus topic files; the first 200 lines or 24 KB of MEMORY.md load each turn, and older notes move to an archive. In Bot Settings, Memory you can read every note, edit or delete a single one, and undo any change from a history of who changed what. Saying "forget that" in chat removes every note that mentions it. An About me holds lasting facts about you, written by you or learned by bots from your own words, and every bot reads it, so a preference you tell one bot reaches all of them; each fact a bot added there has a Remove button. Teams also get shared team memory. Prompts go only to the model provider each bot uses, and that provider's own data setting applies to what you send it, the same as it would in its own app. How the conversation record is kept, and what each model is shown from it, is in what your bot remembers.

The MausBot desktop app with a sidebar of bots and an open conversation; every chat is stored in ~/.openmausbot
Every chat in the sidebar, and every bot's notes, are files on the computer that runs MausBot. Prompts go only to the model provider you choose.

Frequently asked questions

Can I delete a single memory from my OpenAI dot?
No. OpenAI's FAQ says you currently cannot view, delete, or directly modify individual dot memories, including details that enter the dot's context from plugins. The only delete is deleting the dot, which starts deletion of its context and leaves your files, Codex threads, ChatGPT conversations, and ChatGPT memories in place.
Does Meta Muse forget when I ask?
Mostly. You can ask Muse to forget a topic or a set of interactions, and Meta describes that as best-effort. Its help page adds that after you have deleted something, Muse may still remember information it learned from what you deleted. You can also open, edit, and download the MEMORY.md file directly, and a full reset deletes all your Muse data.
Does deleting my ChatGPT memory delete my dot's memory?
No. The dot receives your ChatGPT memories and then builds its own. Turning Memory off in ChatGPT does not delete information the dot has already received, and disconnecting a plugin does not delete what the dot already built into its context. Deleting the dot is the only way to clear it.
Which AI agents store memory locally?
OpenClaw, Hermes Agent, and MausBot. OpenClaw keeps an SQLite index and Markdown files in the agent workspace. Hermes keeps MEMORY.md and USER.md under ~/.hermes/memories. MausBot keeps Markdown notes per bot, an About me every bot reads, and the conversation record in ~/.openmausbot. In each case the memory is a file or folder on your disk, and what reaches a model provider is what goes into your prompts. Dots, Muse, Grok Bot, Cue, Instinct, and Claude keep memory on their makers' servers.
Do AI agents train on my memory?
It depends on the product and the plan. Dots on personal plans follow the Improve the model for everyone toggle, and business plans are not trained on by default. Muse trains by default with an opt-out in Data controls. Grok Bot follows your Cursor account settings. Cue's privacy policy says de-identified data may be used to improve its models, and Manus lists a training opt-out as a Team plan feature. Instinct trains by default, with an opt-out that applies going forward, and never trains on Google Workspace data. OpenClaw, Hermes, and MausBot keep memory on your disk; the provider you send messages to applies its own setting.

Get started

If you want the file instead of the reset button, download MausBot for macOS, Windows, or Ubuntu, sign in with the Claude, ChatGPT, or Grok account you already have (MausBot installs Claude Code or Codex for you), and open ~/.openmausbot whenever you like. The app is free and Apache 2.0 licensed, apart from an enterprise/ folder; the source is on GitHub.

This post is based on the products as they were on . Products change. If something here is out of date, tell us and we will correct it.