MUSE: Multi-Level Memory with Selective Runtime Injection for Scientific Coding Agents

Haoran Jin; Defu Lian; Xiting Wang; Hong Xie

cs.LGOct 28, 2026

MUSE: Multi-Level Memory with Selective Runtime Injection for Scientific Coding Agents

Authors: Haoran Jin, Defu Lian, Xiting Wang, Hong Xie

Organizations: University of Science and Technology of China · State Key Laboratory of Cognitive Intelligence, Hefei, Anhui, China · Gaoling School of Artificial Intelligence, Renmin University of China

Abstract

Historical experience can guide scientific coding at multiple levels, but task-level retrieval may miss implementation lessons that become relevant during execution. We introduce MUSE, a memory management framework combining multi-level memory with selective runtime injection for scientific coding agents. MUSE builds global, approach-level, and implementation-level memory from historical trajectories. At initialization, the agent receives global memory and problem-matched approach-level memory. During execution, current interactions are used to retrieve implementation-level memory. An LLM gate selects relevant experience and generates specific review advice, or skips injection. Across four benchmarks spanning numerical problem solving, kernel optimization, and scientific software repair, MUSE outperforms retrieval and memory baselines on all evaluation metrics. Ablations support the multi-level design and gated injection, while transfer experiments show that memory remains useful across kernel benchmarks. These results show that multi-level memory and selective runtime injection enable effective reuse of experience in scientific coding. Code is available at https://github.com/hr-jin/MUSE.

Figures & tables

Method	Success & Failure Exp.	Multi-Level Memory	Selective Runtime Injection
Trajectory RAG (control)	$\checkmark$	–	–
ReasoningBank ( Ouyang et al., 2026 )	$\checkmark$	–	–
ReMe ( Cao et al., 2026 )	$\checkmark$	$\bigcirc$	$\bigcirc$
AutoSci–SciMem ( Qian et al., 2026 )	$\checkmark$	$\bigcirc$	–
AdaMEM ( Zhang et al., 2026c )	–	$\bigcirc$	$\checkmark$
EvoScientist ( Lyu et al., 2026 )	$\checkmark$	$\bigcirc$	$\bigcirc$

Table 1: Comparison of memory mechanisms. Symbols denote full (

\checkmark

), partial or project-local (

\circ

), and no explicit (–) support in the compared configurations. Multi-level memory separates approaches from implementation details. Selective runtime injection can skip injection within a trajectory.

Figure 1: Memory construction and injection in MUSE . (a) Approach-level (L1) and implementation-level (L2) lessons are extracted from historical trajectories. Consolidation and review build global memory (L0). (b) The agent receives L0 and problem-matched L1 references at initialization. During scientific coding, the LLM gate selects a retrieved source record and provides specific review advice, or skips injection.

Benchmark	Extract	Test
PDEAgent-Bench	194	451
KernelBenchX	53	122
KernelGenBench-MS	63	140
SWE-bench Science	35	84

Table 2: Task counts for memory extraction and testing.

	PDE	KBX		KGB		SWE Science
Method	Pass %	Correct %	Accel. %	Correct %	Accel. %	Pass %
Vanilla agent	44.8	57.4	31.1	44.3	8.6	29.8
RAG	72.3	53.3	29.5	57.9	12.1	28.6
ReasoningBank	63.6	59.8	28.7	47.9	15.7	31.0
ReMe	50.1	50.0	29.5	40.0	20.0	35.7
TRAD	75.8	60.7	29.5	48.6	9.3	31.0

Table 3: Results on four scientific coding benchmarks. Bold and underlined values mark the best and second-best scores in each column. Accel. denotes the percentage of test tasks passing correctness checks with a speedup of at least

1.01\times

over the benchmark reference.

Method	Pass %
Vanilla agent	44.8
Agent KB	69.6
AutoSci–SciMem	78.5
ACE	83.8
EvoScientist	87.8
MUSE L0 only	82.7

Table 4: Cumulative memory-level ablation on PDE. The strongest baseline is underlined.

	Pass/Correct %		Accel. %
Benchmark	MUSE	w/o gate	MUSE	w/o gate
PDE	92.2	90.7	–	–
KBX	79.5	66.4	44.3	37.7
KGB	66.4	62.1	24.3	20.0
SWE Science	40.5	32.1	–	–

Table 6: Gate ablation across four benchmarks. The no-gate variant retains all three memory levels and retrieval rules, but directly injects the highest-similarity eligible source record. Pass/Correct denotes pass rate for PDE and SWE Science and correctness rate for KBX and KGB.

Figure 2: Memory use in three kernel tasks. Global memory (L0) supplies a precision rule, approach-level memory (L1) an FP8 encoding approach, and implementation-level memory (L2) a two-pass variance implementation. Speedups are relative to the corresponding benchmark reference implementations.

LLM Agent Memory

Large language model agent memory lets agents retain, organize, retrieve, and update information across interactions and long-running tasks.

See all 376 papers

REM: What to Store, When to Recall, What to Forget in Experience Memory for Coding Agents
Oct 19, 2026Xinyi Lisa Chen, Pedro Herrero-Vidal, Xiaohan Chen +6Memory Management for LLM Agents Agent Memory
Yale University · Amazon
ScienceIDE: Turning World's Scientific Codebase into Agent Learnable Environments
Sep 16, 2026Hejia Geng, Zesen Huang, Haoyang Li +42AI for Science
VibeMemBench: Evaluating Memory Systems for Coding Agents on Real Repository Coding Tasks
Sep 20, 2026Liyang Fan, Yingcheng Shi, Yongbin Li +7Benchmark Design Software Engineering Benchmarks
Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences · SUAT · Alibaba Token Hub, Alibaba Group +1

Oct 19, 2026J/K move · Enter open · S save

REM: What to Store, When to Recall, What to Forget in Experience Memory for Coding Agents

Xinyi Lisa Chen, Pedro Herrero-Vidal, Xiaohan Chen, Xiaojing Yu+5

Yale University · Amazon

Memory Management for LLM Agents Agent Memory Software Engineering

Open paper AskPDF arXiv

Large language model (LLM) agents now autonomously complete long-horizon tasks, such as repository-level coding, through hundreds of think-act steps, yet the experience each run produces is rarely reused on the next. Prior memory systems concentrate on retrieval, finding the most relevant past trajectory. Retrieval cannot observe the running agent, so it cannot decide whether the agent needs an experience, at which point, or for how long. We introduce REM (Reusable Experience Memory), a framework that manages the memory lifecycle from inside the agent's think-act loop through three explicit decisions. What to store: each completed trajectory is consolidated into a phase-structured document, two orders of magnitude smaller than the trajectory. When to recall: the document is injected at the start of the task; the costlier implementation record is withheld until an in-loop gate detects the agent struggling. What to forget: the record is evicted after a single exposure. On repository-level bug fixing (SWE-ContextBench Lite), REM retrieves the related trajectory more accurately than every state-of-the-art memory system (68.7% vs. 55.6% match@1) and improves, in every replicate, end-task resolution (27.6% vs. 24.9%) at 17% lower cost. Injecting the same experience as a summary or raw trajectory, even when oracle-retrieved, does not. Ablating any of the three decisions reduces resolution. Altogether, these results demonstrate that the bottleneck in experience reuse is not retrieval fidelity but how experience is stored and governed across its lifecycle, opening a control-theoretic perspective on agent memory.

Abstract

LLM Agent Memory

Explore similar work

REM: What to Store, When to Recall, What to Forget in Experience Memory for Coding Agents

ScienceIDE: Turning World's Scientific Codebase into Agent Learnable Environments

VibeMemBench: Evaluating Memory Systems for Coding Agents on Real Repository Coding Tasks