GUM

GUM (General User Model) is the shared memory layer for agents. It is not just a database or a chat history archive. It is governed user context infrastructure.

Core problem

Without persistent memory, agents can only perform stateless prompt execution. Every task has to re-explain preferences, decisions, and user context.

GUM solves context continuity across sessions, tasks, devices, and products.

Memory types

Memory	Purpose
Chat Memory	Long-term preferences, commitments, facts, and task clues from conversations
Profile Memory	Dynamic user profiles, background, interests, and decision styles
Action Memory	Tool calls, web operations, and behavior patterns
Search Memory	Search intent, retrieved results, sources, and later usage

Governance

GUM is not about remembering more. It is about remembering in a controlled way:

• Memory needs lifecycle management.
• Memory updates need versions and provenance.
• Users need erasure and export capabilities.
• Agent answers need to explain which memories influenced them.

Product direction

GUM combines lightweight SDK access with automatic context assembly, while emphasizing white-box memory, explainability, and user control.

Developers should be able to call the SDK explicitly or derive user context from conversation, search, and action events with low integration overhead.