Referee · 21 October 2026

Who owns your agent's memory?

thecontinuitylayer.com/field-watch/who-owns-your-agents-memory

Three always-on AI agents shipped in the last three weeks. Each one remembers you. We compared what their makers say about that memory: where it lives, and whether you can see it, edit it, delete it, export it, or take it to another provider.

This is read from published documentation and reporting as of 21 October 2026. We did not test these agents by hand. Where a source does not answer a question, the table says “Not stated”. Every answer links its source.

01The three agents

What shipped

ChatGPT dots (OpenAI). OpenAI began rolling out dots on 29 September 2026 to Pro and Business Premium subscribers (OpenAI, Releasebot). A dot keeps working after you log off and starts from your ChatGPT memories (OpenAI help).

Gemini agent (Google). Google announced the Gemini agent at Gemini at Work on 8 October 2026, for businesses first (Google, TechCrunch). It keeps one set of memories across the places it works, in four kinds: session, semantic, procedural and episodic (9to5Google). It can also use Anthropic's Claude models (TechCrunch).

Claude (Anthropic). On 9 October 2026 Anthropic added dynamic workflows to Claude Managed Agents, in beta (Releasebot, Anthropic release notes). Managed Agents keep memory in memory stores, as files (Anthropic). The Claude app has its own memory, which you can view and edit in Settings (Claude help).

02The comparison

What you can do with the memory

Agent	Where the memory lives	Can you see it?	Edit it?	Delete it?	Export it?	Take it to another provider?
ChatGPT dotsOpenAI · Rolling out from 29 September 2026	Inside the dot, on OpenAI's side. A dot starts from your ChatGPT memories and adds its own.source	No. Individual dot memories cannot currently be viewed.source 1 source 2	No. Individual dot memories cannot currently be edited.source 1 source 2	Only all at once, by deleting the whole dot.source	Not statedsource	Not statedsource
Gemini agentGoogle · Announced 8 October 2026 (enterprise)	One set of memories in Google's cloud, shared across the places the agent works. Four kinds: session, semantic, procedural, episodic.source 1 source 2	Not statedsource	Not statedsource	Not statedsource	Not statedsource	Not stated. The agent can use Claude models, but the sources do not say the memory can leave Google.source
Claude Managed Agents (and Claude app memory)Anthropic · Dynamic workflows beta 9 October 2026	In memory stores on Anthropic's platform, kept as files. Stores can also attach to a sandbox you host.source 1 source 2	Yes. Memories can be listed and read through the API or the Claude Console. In the Claude app, memory is listed in Settings.source 1 source 2	Yes, through the API. The Claude app also lets you edit memory.source 1 source 2	Yes, one memory or a whole store, through the API. The Claude app lets you delete topics or reset memory.source 1 source 2	Yes, through the API. The Claude app has no export file; you copy memory out by viewing it or asking Claude to write it out.source 1 source 2	Partly. You can read it out and carry it yourself. No tool for moving it into another provider is stated.source

03Why this matters

Your agent now remembers you

These agents run on their own and build up memory about you as they work. With dots, you cannot currently see or edit that memory; your only option is to delete the whole dot (OpenAI help, NeuralTrust). Google has not said whether you can see, change or take out the Gemini agent's memory (Google). Anthropic documents reading, editing, deleting and exporting Managed Agents memory through its API (Anthropic docs).

None of the three says you can move that memory straight into another provider's agent. So the agent may know you well, and you may not be able to see, move or own what it knows.

For every provider's memory controls in one table, see the Memory Check.