Shared Selective Persistent Memory for Agentic LLM Systems
Apple Machine Learning Research
Shared selective persistent memory was introduced for agentic LLM coding systems that otherwise start each session from zero and lose prior task context, while full history persistence can hurt quality. In enterprise deployments it reached 96% task completion versus 79% without memory and 71% with full history. The approach preserves reusable task specifications, data schemas, tool settings, and output constraints across users with access control, while discarding session-specific reasoning so agents need fewer tokens and can reuse artifacts without re-invoking the model for recurring data refreshes.
Why it matters
Agentic LLM systems that generate code through multi-turn tool use face a fundamental context problem: each session starts from zero, discarding the configuration choices, domain constraints, data schemas, and tool-use patterns that made previous sessions productive. Naively persisting entire conversation histories is both token-inefficient and counterproductive—irrelevant context degrades generation quality. We introduce shared selective persistent memory, a memory architecture for agentic systems that identifies and retains four categories of reusable context—task specifications, data…