Loading
Life sciences · Preprint
arXiv · September 9, 2026
Early or partial results. Treat as a signal, not a conclusion.
This is an unreviewed systems engineering study proposing kernel-managed shared memory for multi-agent AI personalization. The authors report that their design improves personalization scores by 2.4–4.0 points (5-point scale) over an unmanaged memory backend and standard retrieval-augmented injection across three language models, with 15–61% latency reduction compared to full unfiltered context, all at reported p < 10^-18. The work is technically novel but lacks peer review, real-world validation, and user studies.
Comparative systems evaluation across three models and four conditions. Three language models in a multi-agent system framework; no human participants or real-world user study.. Intervention: Kernel-managed shared memory system with structured tagged memories, centralized retrieval, privacy enforcement, and prompt injection governance.. Compared with: Unmanaged external memory backend (Mem0 using identical storage), standard retrieval-augmented injection, and full unfiltered context concatenation..
Personalization scores improved 2.4–4.0 points on 5-point scale versus Mem0 backend; GPT-4o profile usage rose from 1.05 to 4.69, p < 10^-18 End-to-end latency 15–61% lower than full context concatenation across all three models Kernel-managed injection statistically matched full context on two of three models; small model-specific deficit on third
Safety was not reported in the material analysed. Check the source before drawing any conclusion about harm.
The source did not state who this applies to in practice.
A systems engineering study implementing and evaluating a novel kernel-managed memory abstraction across three models in a controlled experimental setting, but without peer review, clinical or real-world validation, or comparison to established baselines beyond custom alternatives.
As stated by the source record.
Quoted from the source exactly as published.
Graded across the dimensions that decide whether you should act, each from what the source actually supports. There is no single score, and where a dimension was not assessed it says so.
AI systems become more useful when they can adapt to the people using them, but in multi-agent systems, useful context learned by one agent often remains unavailable to others. We present kernel-managed shared memory, a system-level abstraction in which specialized agents write structured, tagged memories while the agent-system kernel, not individual agents, governs retrieval, privacy enforcement, and prompt injection. We implement and evaluate this design on AIOS and compare it against three alternatives across three assistant models (GPT-4o, Llama-3.1:8B, Qwen-2.5:7B) and 1,800 total trials. Against an unmanaged external memory backend (Mem0) using identical underlying storage, kernel-managed retrieval and injection improve personalization scores by 2.4-4.0 points on a 5-point scale (e.g., 1.05 to 4.69 profile usage on GPT-4o), with every comparison significant at p < 10^-18. Against standard retrieval-augmented injection, gains are similarly large and consistent across all three models. Against full, unfiltered context concatenation, a soft ceiling on available context rather than on response quality, kernel-managed injection statistically matches performance on two of three models and shows a small, model-specific deficit on the third, while using substantially shorter prompts: end-to-end latency is 15-61% lower across all three models, with corresponding reductions in per-call token usage and inference cost. These results indicate that centralizing memory management in the agent-system kernel, rather than leaving retrieval and privacy enforcement to individual agents, delivers most of the personalization benefit of unconstrained context at a fraction of its cost.
Taken from the source record, never inferred. Follow any of these and new work involving them reaches your briefing.