Skip to main content
Knowledge mapWho owns an agent’s memory?Industry concepts and standards

You are here. See how this question connects to other ideas.

Select a node to open its page · Expand to read within the map

Knowledge mapFollow a connection. Understand a question.
← AI and agents

One question

Who owns an agent’s memory?

Distinguish personal records, task context and transient execution state.

Original notes, selected task passages and temporary tool results are different data. Calling them all “memory” hides who holds them, can delete them or permits disclosure.

A reading assistant might keep originals in a personal data space and read only authorized material for a task. What reaches a model service depends on the actual request flow. Personal storage alone does not prevent disclosure to a model.

AFS organizes accessible context; Small World limits the visible working surface. Arc Space addresses personal data and application access. Storage ownership, access permission and inference traffic still require separate checks.

Trace one note

Where is its original? Who can read it? What is sent to the model now? What remains after the task? These questions are more revealing than a claim of long-term memory.

Sources and further reading

Check your understanding

Must changing models mean surrendering all personal records again?

These need not be coupled; portability depends on storage, interfaces and authorization.

Continue along a learning path