Apple Machine Learning Research1 min readpaperadvanced
Shared Selective Persistent Memory for Agentic LLM Systems
Summary
Apple proposes a memory architecture for agentic LLMs that selectively persists reusable context (specs, schemas, configs, constraints) across sessions and users. Shared workspaces with role‑based access and a zero‑token data‑refresh mechanism cut token usage by 97×, reduce task time by 14×, and raise task‑completion rates to 96% versus 71%‑79% for baselines.
- Selective persistent memory stores only four reusable context categories—task specifications, data schemas, tool configurations, and output constraints—while discarding session‑specific reasoning traces.
- Memory is shared across users via role‑based access, enabling collaborative reuse of context without re‑specifying it for each session.
- A zero‑token data‑refresh mechanism decouples generated code from runtime data, allowing artifacts to be updated without re‑invoking the LLM.
- In three enterprise deployments, the approach achieved 96% task completion (vs. 79% without memory, 71% with full‑history persistence).
Agentic LLM workflows suffer from context bloat and repeated specification overhead. By persisting only high‑value context and sharing it across users, the system dramatically improves efficiency and reliability, making LLM‑driven code generation viable for larger, collaborative, production‑grade e…
8/10
