Prism: An Evolutionary Memory Substrate for Multi-Agent Open-Ended Discovery

Suyash Mishra

cs.AIApr 8, 2026

Prism: An Evolutionary Memory Substrate for Multi-Agent Open-Ended Discovery

Authors: Suyash Mishra

Organizations: AI Researcher · Basel, Switzerland

Abstract

We introduce \prism{} (Probabilistic Retrieval with Information-Stratified Memory), an evolutionary memory substrate for multi-agent AI systems engaged in open-ended discovery. \prism{} unifies four independently developed paradigms -- layered file-based persistence, vector-augmented semantic memory, graph-structured relational memory, and multi-agent evolutionary search -- under a single decision-theoretic framework with eight interconnected subsystems. We make five contributions: (1)~an entropy-gated stratification mechanism that assigns memories to a tri-partite hub (skills/notes/attempts) based on Shannon information content, with formal context-window utilization bounds; (2)~a causal memory graph $\mathcal{G} = (V, E_r, E_c)$ with interventional edges and agent-attributed provenance; (3)~a Value-of-Information retrieval policy with self-evolving strategy selection; (4)~a heartbeat-driven consolidation controller with stagnation detection via optimal stopping theory; and (5)~a replicator-decay dynamics framework that interprets memory confidence as evolutionary fitness, proving convergence to an Evolutionary Stable Memory Set (ESMS). On the LOCOMO benchmark, \prism{} achieves 88.1 LLM-as-a-Judge score (31.2% over Mem0). On CORAL-style evolutionary optimization tasks, 4-agent \prism{} achieves 2.8 $\times$ higher improvement rate than single-agent baselines.%

Figures & tables

Panel of Figure 1 — Figure 1: Prism architecture with eight subsystems. Multiple agents feed into a shared extraction engine. Memories are stratified into a tri-partite hub (skills/notes/attempts). The heartbeat controller triggers reflection, consolidation, and redirection. Replicator-decay dynamics govern confidence evolution. VoI retrieval assembles context-enriched prompts.

Method	$J$ ( $\uparrow$ )	F1 ( $\uparrow$ )	p95 (s)	Savings
Full Context	71.4	58.2	17.12	—
RAG-512	59.8	47.1	2.31	82%
DeerFlow	62.5	49.8	0.42	88%
CC Memory	65.1	52.4	0.38	90%
Mem0	67.2	53.9	1.44	90%
Mem0 g	68.4	55.1	2.59	88%

Table 1: LOCOMO benchmark.

J

= LLM-as-a-Judge (higher better).

Method	Single	Temporal	Multi-hop	Open
Mem0 g	73.8	58.1	70.2	73.6
Prism	91.2	84.7	88.3	87.1

Table 2: Per-category LOCOMO scores (

J

Method	Agents	IR ( $\uparrow$ )	KR	Best Score	Evals
100-City TSP (minimize distance)
Random Evo	1	0.08	—	42,180	500
OpenEvolve	1	0.15	—	38,920	500
Prism (1)	1	0.31	0.42	35,640	380
Prism (4)	4	0.87	0.71	33,210	420
Circle Packing (maximize density)

Table 3: Evolutionary optimization tasks. IR = improvement rate (higher = better). KR = knowledge reuse rate.

Configuration	$J$ (LOCOMO)	IR (TSP)
Full Prism (4 agents)	88.1	0.87
$-$ Tri-partite hub (flat tiers)	82.4	0.68
$-$ Causal edges ( $E_{c}$ )	76.8	0.59
$-$ Evolutionary VoI	79.3	0.44
$-$ Heartbeat controller	84.6	0.51
$-$ Replicator dynamics	85.2	0.62

Table 4: Ablation: removing individual Prism components.

Panel of Figure 2 — Figure 2: Knowledge reuse rate over time on TSP. Multi-agent Prism accumulates shared knowledge at nearly double the rate of single-agent, with the gap widening after turn 200 as agents build on each other’s discovered skills.

Panel of Figure 3 — Figure 3: Exploration divergence vs. improvement rate across agent pairs and tasks. Higher divergence correlates with higher improvement ( $r=0.91$ ), confirming Theorem 3.2 : diverse multi-agent exploration yields better coverage of the solution space.

Appendix figures & tables2 assets

Supplementary material from the paper’s appendix.

Appendix

$\tau_{1}$	$\tau_{2}$	$J$	Skills	p95 (s)
1.5	4.0	85.3	42	0.94
2.0	4.5	86.8	28	0.91
2.5	5.0	88.1	18	0.89
3.0	5.5	87.4	12	0.93
3.5	6.0	84.9	8	1.02

Appendix

Table 5: Entropy threshold sensitivity (LOCOMO

J

score).

$\lambda$ (decay)	$\mu$ (mutation)	IR	$\|\mathcal{M}^{*}\|$
0.005	0.001	0.72	340
0.01	0.005	0.87	185
0.02	0.005	0.81	120
0.01	0.01	0.83	210
0.05	0.005	0.65	62

Appendix

Table 6: Replicator dynamics parameters on TSP improvement rate.

Agent Memory Management

Agent memory management studies how AI agents store, update, retrieve, compress, and forget information across long-running tasks.

See all 446 papers

MEME: Multi-entity & Evolving Memory Evaluation
May 12, 2026Seokwon Jung, Alexander Rubinstein, Arnas Uselis +2Agent Memory Persistent Memory for Language Models
1KAIST AI · 2Tübingen AI Center, University of Tübingen · 3NAVER AI Lab
Omni-SimpleMem: Autoresearch-Guided Discovery of Lifelong Multimodal Agent Memory
Apr 1, 2026Jiaqi Liu, Zipeng Ling, Shi Qiu +9Agent Memory Multimodal Memory
UNC-Chapel Hill · University of Pennsylvania · University of California, Santa Cruz +2
Remembering Is Not Enough: Refining Memories and Learning When to Use Them in Multi-Agent Systems
Oct 27, 2026Yipeng Hao, Yuxuan Cai, Wenkai Ji +6LLM Agent Memory Memory Management for LLM Agents
School of Computer Science and Technology, East China Normal University, Shanghai · Ocean University of China · Tongji University +1

May 12, 2026J/K move · Enter open · S save

MEME: Multi-entity & Evolving Memory Evaluation

Seokwon Jung, Alexander Rubinstein, Arnas Uselis, Sangdoo Yun+1

1KAIST AI · 2Tübingen AI Center, University of Tübingen · 3NAVER AI Lab

Agent Memory Persistent Memory for Language Models LLM Agent Evaluation Memory Agent Benchmarks Memory Management for LLM Agents

Open paper AskPDF arXiv

LLM-based agents increasingly operate in persistent environments where they must store, update, and reason over information across many sessions. While prior benchmarks evaluate only single-entity updates, MEME defines six tasks spanning the full space defined by the multi-entity and evolving axes, including three not scored by prior work: Cascade and Absence (dependency reasoning) and Deletion (post-removal state). Evaluating six memory systems spanning three memory paradigms on 100 controlled episodes, we find that all systems collapse on dependency reasoning under the default configuration (Cascade: 3%, Absence: 1% in average accuracy) despite adequate static retrieval performance. Prompt optimization, deeper retrieval, reduced filler noise, and most stronger LLMs fail to close this gap. Only a file-based agent paired with Claude Opus 4.7 as its internal LLM partially closes the gap, but at ~70x the baseline cost, indicating closure currently depends on configurations that are not practical at scale. Code and data are available on the project page: https://seokwonjung-jay.github.io/meme-eval/.

Abstract

Agent Memory Management

Explore similar work

MEME: Multi-entity & Evolving Memory Evaluation

Omni-SimpleMem: Autoresearch-Guided Discovery of Lifelong Multimodal Agent Memory

Remembering Is Not Enough: Refining Memories and Learning When to Use Them in Multi-Agent Systems