Most personal AI agents keep your data until you take action to delete it. Only a few vendors publish a timeline. Anthropic says Claude Duet keeps your history until you delete or reset it, and deleting Grok Bot removes your data within 30 days. Meta has not published a retention period for Muse, and OpenAI gives no period for dots. This guide sets out what each vendor documents as of November 2026, based on the official docs and launch coverage.

What data does a personal AI agent actually keep?

An agent holds far more than a chatbot does. A typical setup keeps four kinds of data, and each one may follow its own retention rule:

  • Chat history: what you typed and what the agent replied.
  • Connector data: email, calendar entries, documents and messages the agent read while doing a task.
  • Memory: facts the agent decided to remember, such as your address or your manager's name.
  • Workspace and backups: files on the agent's cloud computer or VM, plus vendor backups.

When you delete one of these, the others are usually left in place. Disconnecting Gmail stops new reads, but it does not erase the memories the agent formed from your inbox. The table below reflects that.

Retention and deletion by vendor

Agent Stated retention Delete single items? Full deletion Disconnecting an app
Claude Duet Kept "until you delete your history or reset it" Delete history, reset memory Deleting the account wipes everything Keeps what Duet learned until memory reset
OpenAI dots No period published No, individual memories can't be viewed, edited or deleted Reset or delete the dot; Privacy Portal or dsar@openai.com Doesn't remove what the dot learned
Meta Muse Meta has not published a period Ask Muse to "forget" items; edit VM files and memory No published full deletion steps; cancelling is not deleting Not documented
Grok Bot Follows Cursor's terms Deleting one Bot is a soft delete Deleting Grok Bot deletes the Cursor account, data removed within 30 days Not documented
OpenClaw (self-hosted) Whatever you configure Yes, files are on your machine Delete the data directory You control it

As of November 2026. Sources: Anthropic's Duet safety article, OpenAI's dots FAQ, Meta's Muse help center, xAI's Grok Bot FAQ.

Grok Bot data removal after account deletion
30 days
Major hosted agents with no published retention period (Muse, dots)
2 of 4
Maximum retention Anthropic states for consumer feedback you submit
5 years

Claude Duet: kept until you delete it

According to Anthropic's safety article, Duet keeps what it has learned until you delete your history or reset memory. Duet draws on Claude's memory and the "Search and reference chats" setting, so turning those settings off limits what it can recall later. Deleting your Claude account wipes everything. Disconnecting a connector does not: what Duet learned from it stays until you reset memory.

Anthropic's general consumer data policy says that feedback you submit, such as a thumbs-down with the conversation attached, can be kept for up to five years. Anthropic has not published a Duet-specific retention schedule, so treat the consumer policy as the closest guide rather than a confirmed Duet rule.

OpenAI dots: reset is the only real delete

OpenAI's dots privacy FAQ is unusually blunt. You cannot view, edit or delete individual memories a dot holds. The options are resetting the dot or deleting it. Turning off ChatGPT Memory stops new information flowing to the dot but does not remove what the dot already has. Disconnecting an app also leaves its learned information in place.

For account-level requests, OpenAI points to its Privacy Portal or dsar@openai.com. It does not state how long dots data is kept after deletion.

Meta Muse: editable, but no published clock

Meta lets you inspect, edit and download the files on Muse's VM and its memory, and you can tell Muse to forget specific things. But Meta has not published a retention period or a full deletion procedure. VM data is backed up continuously, so a file you delete may survive in a backup for some unknown period.

Meta also states that the Muse architecture "does not prevent Meta from accessing data when necessary to support, secure or operate the service". The Confidential VM option that would limit this is not generally available; trusted testers have it, and Meta has said it is due by the end of 2026. Muse signs in with a Meta account, and cancelling Muse does not delete it.

Grok Bot: the clearest timeline, with a catch

xAI's FAQ and Cursor's documentation say Grok Bot follows Cursor's account and retention terms. Deleting Grok Bot deletes the Cursor account, and data is removed within 30 days.

The catch is per-Bot deletion. Each named Bot has its own memory, but all of a user's Bots share one cloud computer. Deleting a single Bot is a soft delete, and files and saved logins on the shared computer may remain. If you set up a Bot with access to a sensitive account and later delete it, check the shared computer and sign out of that account as well.

How to minimise what is retained

  1. Audit connectorsRemove apps the agent no longer needs, then reset memory, because disconnecting alone keeps learned data.
  2. Clear memoryUse the agent's forget, reset or delete option on a schedule, such as monthly.
  3. Check the workspaceOn Muse and Grok Bot, look at files and saved logins on the agent's computer.
  4. Delete the accountFor a full exit, delete the account itself, not just the subscription.
  5. File a requestUse the vendor's privacy request route if you need confirmation of deletion.

Our step-by-step guide to deleting an AI agent's memory covers the exact menus for each agent. If you are leaving an agent entirely, also revoke its access at the source, such as your Google or Microsoft account permissions page, so tokens stop working even if the vendor keeps a copy.

Why retention matters for safety

Retained memory affects safety as well as privacy. Researchers have shown that planted text can become a lasting memory that steers later actions, a technique covered in our explainer on AI agent memory poisoning. Long retention also means a compromised account exposes months of email-derived facts, not just today's chat. And training is a separate question from retention: an agent can delete your chat while a model trained on it keeps what it learned, which is why the training opt-out settings matter.

UK readers have extra rights under data protection law; see UK data rights with AI agents. For a side-by-side of memory features, see AI agent memory privacy compared.

Bottom line

Assume your agent keeps everything until you delete it, and that disconnecting an app deletes nothing. Claude Duet and Grok Bot give you the clearest controls. With Meta Muse and OpenAI dots, the vendor publishes no clock, and on dots resetting the whole dot is the only way to clear memory. Build a routine: prune connectors, reset memory and delete accounts you have stopped using.