Memory Management

Create, retrieve, update, and delete memory entries programmatically using the Memory Management APIs. These APIs supplement the automatic extraction pipeline by enabling you to curate knowledge manually. They also serve as the underlying mechanism for UI-based memory curation where users view, edit, and delete facts and preferences.

Note: Memory Management APIs operate alongside the automatic extraction pipeline. Both can write to the same scope. For endpoint details, request schemas, and complete field reference, see Agentic Memory & Context API Reference.

When both the Memory Management API and the automatic extraction pipeline write to the same scope, these behaviors apply.

  • API writes take priority: The system cancels any active extraction batch for that scope when an API write occurs.
  • Pipeline re-runs: The extraction pipeline re-runs with the updated state on its next cycle.
  • Entry tagging: API-created entries are tagged with author type USER_API to distinguish them from system-generated entries.

When your data space uses Unified Individual, retrieve memories across all member individuals in a single call using the unified flag. Data 360 performs these operations.

  • Resolves the unified identity.
  • Fans out reads across all member individuals.
  • Returns a merged result set with the source individual ID on each entry.

This enables agents to access the complete memory profile across all identities that comprise a unified individual.

  • Data 360 scopes each entry to one individual and one memory type.
  • You can’t create conversation-scoped memories using the API. The conversationId scope is reserved for system-generated summaries.
  • Expired entries aren’t physically deleted. You can recover them by updating the validTo field.
  • Data 360 stores freeform fields with no type validation.
  • Deletion is permanent and you can’t undo a deletion.