From Memory Hits to Answerable Evidence: Query-Conditioned Memory Compilation for Long-Term Agent Memory

Shangze Li; Yanyu Yao; Derong Xu; Zhi Zheng; Tong Xu

cs.AIOct 16, 2026

From Memory Hits to Answerable Evidence: Query-Conditioned Memory Compilation for Long-Term Agent Memory

Authors: Shangze Li, Yanyu Yao, Derong Xu, Zhi Zheng, Tong Xu

Organizations: State Key Laboratory of Cognitive Intelligence, University of Science and Technology of China, Hefei, China

Abstract

Long-term agent memory requires recovering answer-relevant evidence dispersed across turns and sessions. Existing systems increasingly use structured memories and adaptive retrieval, yet often pass retrieved or expanded units directly to the reader, coupling memory addressing with evidence presentation and leaving the final context redundant or incomplete. To address this gap, we propose MemCompile, a query-conditioned evidence compilation framework for long-term conversational memory. Given a query, the Address Plane locates relevant support through compact source-linked descriptors and capacity-aware Cells, while a normalization boundary resolves heterogeneous retrieval hits to stable SourcePacket identities. Its Evidence Plane then allocates a bounded evidence budget across complementary packets, admits limited additional support, and renders the retained claims associated with each selected packet into an answer-ready context. MemCompile achieves 80.52% Judge Accuracy on LoCoMo and extends the empirical quality-cost Pareto frontier to a higher-quality regime. Its strong performance on LongMemEval-S further suggests that the benefits of query-conditioned compilation generalize across long-term conversational-memory benchmarks.

Figures & tables

Figure 1: Motivation and overview of MemCompile . Existing memory systems face different trade-offs in retrieval units and evidence representation. MemCompile addresses these challenges through source-normalized retrieval, complementary evidence selection, and packet-scoped reader-view compilation.

Figure 2: The overall architecture of MemCompile

Method	F1 $\uparrow$	BLEU-1 $\uparrow$	J $\uparrow$	F1 $\uparrow$	BLEU-1 $\uparrow$	J $\uparrow$	F1 $\uparrow$	BLEU-1 $\uparrow$	J $\uparrow$	J $\uparrow$	F1 $\uparrow$	BLEU-1 $\uparrow$	J $\uparrow$	Tokens (K)
	Single-Hop			Multi-Hop			Temporal			Open-Domain	Overall
LoCoMo	37.81	26.23	57.07	20.97	11.09	48.58	34.79	15.46	38.01	41.67	33.85	20.88	50.58	827.20
MemoryBank	17.85	12.13	46.61	11.25	8.20	28.01	14.26	9.60	34.27	23.96	15.76	10.80	39.22	926.98
A-Mem	33.87	27.64	54.10	21.17	15.40	43.26	34.13	29.98	34.89	32.29	31.45	25.77	46.75	11,687.58
Mem0	54.95	44.71	82.40	36.02	25.25	60.99	30.36	29.69	33.02	50.00	45.79	37.57	66.17	55,300.30
MemoryOS	49.72	43.58	66.47	37.15	26.87	60.99	41.99	36.32	34.58	51.04	45.55	38.70	57.86	19,207.67

Table 1: LoCoMo category and overall results . Best results are bold and second-best results are underlined. Tokens exclude judging and embedding.

	Overall
Setting	F1 $\uparrow$	BLEU-1 $\uparrow$	Recall $\uparrow$	Judge $\uparrow$
A0 Full system	55.94	48.65	85.50	80.52
A1 w/o frontier	51.78	44.87	80.84	74.94
A2 w/o scheduler and frontier	42.00	36.01	51.99	60.91
A3 w/o scheduler, with frontier	51.91	44.75	76.55	75.26
A4 Cell-only route	50.64	43.85	69.85	71.88

Table 2: Controlled pipeline-path ablations on LoCoMo . A1 retains the packet anchors returned by Complementary Packet Scheduling but disables subsequent semantic-frontier expansion. A2 and A3 replace the scheduled anchors with deterministic dialogue-level MMR Top-10 anchors, the scoring function and weights are detailed in Appendix B.3 . A7 directly selects the dense cosine Top-10 SourcePackets and presents their claim bundles to the Reader.

Figure 3: RQ1 quality-cost comparison , blue squares form the empirical Pareto frontier.

Controlled stage ( $K=10$ )	Unique $\uparrow$	DBR $\downarrow$	Gold $\uparrow$	$\Delta$ Gold vs. previous (95% CI)	Robust
Unnormalized claim top- $K$	7.690	23.08	48.89	—	—
Exact-text deduplication	7.742	22.56	48.56	$-0.33$ [ $-0.87$ , 0.21]	—
Source-normalized top- $K$	9.988	0.00	51.91	+3.35 [2.45, 4.26]	$\dagger$
$+$ Complementary source scheduler	9.988	0.00	80.84	+28.93 [26.81, 31.14]	$\dagger$

Table 3: Controlled source selection at

K=10

. Gold Recall and DBR are percentages. Gold differences are percentage points with paired question-bootstrap 95% CIs;

\dagger

means that both question-level and conversation-cluster CIs exclude zero.The source-normalized Top-

K

ranks packets directly by the maximum claim score.

Fixed-packet view	F1 $\uparrow$	BLEU-1 $\uparrow$	Judge $\uparrow$
Representative claim	40.68	35.13	57.01
Raw SourcePacket	53.00	46.01	75.32
Full claim bundle	55.94	48.65	80.52

Table 4: Reader-view ablation at identical packet Recall (85.50%).

Appendix figures & tables7 assets

Supplementary material from the paper’s appendix.

Appendix

Depth	Eligible $N$	Unique	DBR $\downarrow$	Affected	Max-source $\mu$ /P95/max
Top-10	1,539	7.69	23.08%	84.21%	2.42/4/6
Top-20	1,537	14.60	27.02%	97.27%	3.18/5/7
Top-40	1,533	26.94	32.66%	100.00%	4.12/6/8

Appendix

Table 5: Pre-normalization aliasing at three retrieval depths. DBR and Affected are percentages. Max-source reports the mean, 95th percentile, and global maximum number of slots occupied by one source.

Figure 4: Category-level effects of source normalization, complementary scheduling, and source-scoped claim-bundle compilation. (a) Gold Source Recall after exact-text deduplication, source-ID normalization, and complementary source scheduling. (b) Judge Accuracy across the controlled stages and the fixed-source claim-bundle reader view.

Figure 5: Cell-placement contribution to recall. Bars are gains over Flat construction; translucent bands and dashed endpoints show conversation-bootstrap 95% intervals.

Component	Setting	Value
Models	construction, intent, scheduling, answer	gpt-4o-mini
	dense retrieval	text-embedding-3-small
	temperature / maximum completion	$0$ / 2,048 tokens
Cell construction	layout / token cap / item cap	bounded lexical Cell / 360 / 12
	$\beta,\gamma,\eta$ / new-Cell penalty	1.4, 1.0, 0.45 / 0.65
	entropy weight / entropy threshold	0.4 / 1.8

Appendix

Table 6: Configuration of the reported MemCompile experiment.

Figure 6: Post-hoc frontier sensitivity around the fixed reported configuration.

Category	MemCompile	LightMem Fang et al. (2026)	Mem0 Chhikara et al. (2025)
	J $\uparrow$	J $\uparrow$	J $\uparrow$
SSU	91.43	84.29	90.00
MS	54.14	57.14	65.41
TR	72.18	51.88	48.12
KU	78.21	76.92	74.36
SSA	94.64	33.93	44.64

Appendix

Table 7: Judge accuracy (%) on LongMemEval-S. Best category-level results are bold. SSU, MS, TR, KU, SSA, and SSP denote single-session user, multi-session, temporal reasoning, knowledge update, single-session assistant, and single-session preference, respectively.

Stage	Total tokens	Tokens/question	Share of non-Judge total
Fact extraction	1,578,484	1,025	5.9%
Metadata extraction	12,492,607	8,112	46.8%
Cold-build subtotal	14,071,091	9,137	52.7%
Query intent	313,178	203	1.2%
packet scheduler	6,170,000	4,006	23.1%
frontier reranker	4,380,000	2,844	16.4%

Appendix

Table 8: Generation-token accounting for MemCompile on the 1,540 LoCoMo questions. Totals include prompt and completion tokens from unique successful LLM calls; Judge and embedding usage are excluded. Per-question costs amortize each total over all 1,540 questions, and shares are rounded independently.

LLM Agent Memory

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

See all 373 papers

Recall What Is Missing: Composing Evidence from Long-Term Agent Memory
Oct 11, 2026Qi Feng, Jicong FanLong-Term Conversational Memory Information Retrieval
School of Data Science, The Chinese University of Hong Kong, Shenzhen
JustMem: Just-Enough Memory Access for Long-Term Conversations
Sep 17, 2026Guanhua Chen, Yanting Wang, Wenjing Zhi +1Conversational Memory Conversational QA
Beihang University
CueMem: Cue-Guided Context Reconstruction for Long-Term Conversational Memory
Sep 11, 2026Changjian Wang, Rongzhen Li, Weili Guan +3Conversational Memory Long-Term Conversational Memory
Mashang Consumer Finance Co., Ltd. · Harbin Institute of Technology, Shenzhen

Oct 11, 2026J/K move · Enter open · S save

Recall What Is Missing: Composing Evidence from Long-Term Agent Memory

Qi Feng, Jicong Fan

School of Data Science, The Chinese University of Hong Kong, Shenzhen

Long-Term Conversational Memory Information Retrieval

Open paper AskPDF arXiv

Under a limited context budget, a long-term agent must recall not merely memories that are individually relevant, but evidence that is collectively sufficient for the current query. Conventional memory access retrieves, scores, and truncates memories, often spending multiple context slots on redundant facts while leaving complementary evidence behind. We cast the read phase of long-term memory as query-conditioned evidence composition: each memory is valued by what it adds beyond the evidence already selected. We instantiate this operation with MemPursuit, a training-free mechanism that uses the residual of a sparse convex query reconstruction as a representation-space proxy for what the current memory set is missing. By reusing the query and candidate embeddings already available from retrieval, MemPursuit progressively composes nested evidence contexts without invoking a language model or cross-encoder to score each query--memory pair. The effectiveness of our method is verified on 3 popular benchmarks in comparison to strong baselines. In fixed-budget LoCoMo experiments, MemPursuit improves end-to-end accuracy by 7.2 percentage points at Top-10 and 5.5 percentage points at Top-20, with consistent gains across multiple memory frameworks, while achieving a 32.9

\times

speedup in median reranking latency over fully batched BGE-v2-m3. Our implementation is available at https://github.com/FengQi-HITSZ/MemPursuit.

Abstract

LLM Agent Memory

Explore similar work

Recall What Is Missing: Composing Evidence from Long-Term Agent Memory

JustMem: Just-Enough Memory Access for Long-Term Conversations

CueMem: Cue-Guided Context Reconstruction for Long-Term Conversational Memory