Why Git Is the Memory Solution for the Agentic Development Lifecycle

Frank Guo

REVIEW 4 major objections 3 minor 17 references

This paper argues that the memory system for coding agents should be git itself — bound to version control, not bolted on — and that answering from it is a routed three-mode problem, not a single retrieval problem.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-02 02:12 UTC pith:CCQ2Z3C3

load-bearing objection The architecture is a real contribution, but the paper's headline token-economics claim is not measured for the full three-mode router — the routed condition omits decision synthesis. the 4 major comments →

arxiv 2607.14390 v1 pith:CCQ2Z3C3 submitted 2026-07-15 cs.SE cs.AIcs.IR

Why Git Is the Memory Solution for the Agentic Development Lifecycle

Frank Guo This is my paper

classification cs.SE cs.AIcs.IR

keywords agentic development lifecyclegit-bound memoryretrievalanswer sufficiencydecision synthesisepisodic memoryself-labeling benchmarkmemory architecture

verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper argues that memory for coding agents should be bound to git rather than built as separate machinery: the commit supplies ground truth, rebuild supplies freshness, the merge supplies verification, and code review supplies containment. It then shows that answering from that ledger is not one retrieval problem but three modes — a structural map for breadth, confidence-gated episodes for pointed questions, and decision synthesis for why-questions — dispatched by a router. On a young production system of roughly 50k lines, the routed system reaches 0.83 answer-sufficiency on decision rationale and 0.60 overall, at 382–980 tokens per question, with ground truth mined from commit–session links at zero labeling cost. A sympathetic reader would care because this reframes agent memory as a version-control problem with guarantees inherited by construction, and because every result is replicable on any team's own git history.

Core claim

The central claim is that version control already supplies the four things memory machinery struggles to build: annotation (the commit links a session to the verified change it produced), freshness (the index is rebuilt from the append-only ledger and compiled structure is a function of the tree at a SHA), self-confirmation (only checkpoints landing on the default branch are exportable), and containment (code review is the only audited egress). On that ledger the paper separates seed supply from answer assembly: retrieval is closed as a study where five imported ranking mechanisms are rejected and two levers kept, while answer assembly requires a router that sends breadth to a structural map

What carries the argument

The load-bearing object is a git-bound ledger: an append-only store of parsed assistant session turns, tool calls, and commit SHAs, kept deliberately thin (no diff content; code is reconstructed from the SHA on demand), from which labels are mined by SQL over commit–session links. Standing on it are three memory modes — a structural map (condensed, regenerable subsystem description authored from the repository), confidence-gated episodic recall (seeds injected only when the retriever's own scores clear a calibrated bar), and decision synthesis (a decision-scoped gather of turns plus one synthesis call that reconstructs an evolved why-arc with per-claim commit pointers) — dispatched by an unt

Load-bearing premise

The whole evaluation rests on the post-commit hook accurately and completely linking sessions to commits; if sessions are missed or linked to the wrong commit, the self-mined gold labels and every retrieval and sufficiency score are computed on a biased sample.

What would settle it

Manually audit a sample of the post-commit hook's session-to-commit links on a fresh corpus and compare the mined gold labels against human-verified labels; if a substantial share of links are wrong or missing, or if recomputing MRR and sufficiency on corrected labels moves the headline numbers materially, the central claim is falsified. Alternatively, run the same routed system on a corpus where the full reasoning was separately written down and check whether decision synthesis recovers what was recorded.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

A memory system for coding agents should be built into version control, inheriting annotation, freshness, verification, and containment from commits, rebuilds, merges, and review rather than constructing them separately.
Once transcripts are parsed into attributed turns, retrieval ranking is a solved-enough seed supplier: five imported re-ranking mechanisms failed, and the one gain came from an orthogonal facet evidence layer, not rank polish.
Answer assembly is not one retrieval problem: single-shot retrieval scores only 0.07–0.20 sufficiency on real questions, while routing by question kind reaches the per-kind floor at 382–980 tokens per question.
Ungated episode injection measurably degrades good answers, so confidence gating that can choose silence is a required component of the system.
Because ground truth is mined from commit–session links, the full benchmark and all scores are reproducible on any user's own history at zero labeling cost.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The paper specifies but does not run a single expected-cost figure from the real question-kind distribution; instrumenting the router's own query log would turn the last cost table into E[tokens] = sum over kinds of p(kind) times cost(mode), a concrete next step.
Because the paper frames capture — how much reasoning agents verbalize — as the binding constraint, a natural test is to measure per-corpus verbalization rate against sufficiency; if raising verbalization raises sufficiency more than better retrieval does, the bottleneck claim is confirmed.
The confidence-gate signal (top-1 score and top-1–top-2 gap) is calibrated from the retriever's own scores, so the same trick could serve other agent-memory systems as a label-free injection gate.
The git-bound design could be stress-tested on older, larger histories to see whether the young-corpus advantage persists or whether stale episodes begin to poison synthesis over time.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Desk Editor's Note

The architecture is a real contribution, but the paper's headline token-economics claim is not measured for the full three-mode router — the routed condition omits decision synthesis.

read the letter

The paper's central claim — a git-bound, routed memory answering real developer questions at 382–980 tokens with sufficient context — is not actually measured for the full system. The "routed" condition in Table 2 is map + gated episodes only; decision synthesis, the third mode and the one that hits 0.83 on Corpus B, is kept separate and costs ~2.8–3.1k tokens. So the abstract's token figure is for a two-mode system, and the router's dispatch decisions are never evaluated end-to-end. That's a load-bearing gap, and the paper's own limitations section half-admits it by deferring the expected-cost figure to a future run.

That said, the paper earns a serious look. The git-bound thesis is a genuine design position: using the commit as ground truth, the merge as verification, and the ledger as an append-only store gives the memory system properties that bolted-on machinery has to fake. The self-labeling benchmark via commit-session links is clever and cheap, and the retrieval study is more disciplined than most of what crosses my desk — honest floors, held-out splits, paired bootstrap CIs, five imported mechanisms rejected. The facet term ported from SPM is a nice contribution, and the three-mode routing (breadth/pointed/rationale) is a useful way to decompose the problem. The paper also does the rare thing of stating its own weaknesses clearly — small n, single judge, self-labeled gold, directional point estimates.

The soft spots beyond the router gap: the headline MRR of ~0.31 is derived from the tuned baseline plus a facet marginal rather than measured directly; the "label noise biases scores down" claim is asserted, not proven, and false-positive labels could just as easily inflate scores; and the answer-sufficiency evaluation uses 15 self-authored questions per corpus with one automated judge. These are not fatal — they match the paper's own framing as a within-system characterization — but they mean the empirical support is thinner than the prose suggests.

I'd bring this to a reading group: the architecture is worth arguing about, the evaluation is a good template for how to be transparent about limits, and the router gap is a useful teaching example of the difference between a designed system and a measured one. The flaws are addressable, and the direction is worth engaging with. Send it to peer review; a good referee could push the authors to run the full router end-to-end and validate the mining labels against human judgment.

Referee Report

4 major / 3 minor

Summary. The paper argues that the memory solution for the agentic development lifecycle is git itself: a post-commit hook links assistant sessions to the commits they produced, creating a self-labeling ledger (RekalBench), and a skill-layer router sends each question to one of three modes — a structural map, confidence-gated episodic recall, or decision synthesis. The seed-supply stage closes with a retrieval study on eight corpora (two with clean splits) under an incumbent-versus-candidate discipline, reporting that a tuned hybrid with a facet term reaches ≈0.31 pooled MRR versus grep floors, while five imported mechanisms are rejected. The answer-assembly stage reports per-kind answer-sufficiency results on two corpora (n=15 each), with decision synthesis reaching 0.83 on a young production corpus, and a two-mode 'routed' condition costing 382–980 tokens per question. The authors state that ground-truth labels are mined at zero annotation cost and that all artifacts are public.

Significance. If the results hold, the paper makes a useful conceptual contribution: treating git as the source of inherited guarantees (annotation, staleness, self-confirmation, containment) and distinguishing three question kinds with different answer-assembly mechanisms. The study is more methodologically disciplined than typical memory-system papers: pre-registered ship discipline, held-out splits, paired bootstrap CIs, explicit rejection of weak mechanisms, and publication of run records. The self-labeling benchmark, if its linking accuracy is validated, would be a substantial reproducibility contribution. However, the empirical support for the headline numbers is partly derived (best-config MRR), partially unmeasured (full-router token cost and sufficiency), and underpowered by the authors' own admission (n=15, one judge). The paper's significance is therefore conditional on closing these gaps; in its current form the abstract and conclusion overstate what is demonstrated.

major comments (4)

[§7.4, §8.3, Tables 2 and 4] The claimed token economics 'Routed, the system answers at 382–980 tokens per question' is not supported for the full three-mode system described in §7.4. The row labeled 'routed' in Table 2 is 'map + gated episodes'; it excludes decision synthesis, which is the mode the router dispatches for 'why' questions and which costs 2,762–3,135 tokens per question (Table 2, overall avg tok). The router's triage policy is never executed end-to-end, and the expected-cost formula is deferred to a 'future run' (§11). The abstract thus presents a two-mode cost figure as applying to the designed three-mode router. A concrete remedy is to run the full router on a held-out question set and report measured costs and sufficiency per routed question.
[§6 with §5 and §8] The gold labels for both studies are self-mined from commit–session links produced by the system's own post-commit hook. No independent validation of link accuracy or capture completeness is reported. The claim that label noise 'biases scores down, not up' (§6) is not proven; systematic misspecification (e.g., wrong-commit links, or unverbalized reasoning) could bias metrics in either direction. Since every MRR and sufficiency number inherits this gold, the linking precision is load-bearing. A concrete test is needed: human-validate a subsample of mined gold pairs and measure the hook's precision/recall against a manual session-to-commit mapping, then report the effect on the headline metrics.
[§8.1–8.3, §11, §12] The answer-sufficiency evaluation has n=15 per corpus, one blind judge, and one execution model; the authors themselves state the pooled point estimates are 'not statistically separable' (§8.1) and that the honest summary is per-kind patterns, not separable pooled rankings. In this context the abstract's claim of 0.60–0.83 sufficiency and the conclusion's claim that the combination 'outperforms every single-mechanism alternative on coverage' (§1, §12) are not supported at the stated confidence level. The same limitation applies to the synthesis result on Corpus B (0.83 CI overlaps others). A second judge/agreement, larger n, or consistently hedged wording is required.
[§5.1–5.2, Table 1] The headline seed-stage number '≈0.31 pooled MRR' is not a directly measured held-out value. Table 1's footnote says 'best-config cells are derived from the tuned baseline plus the facet marginal,' and §5.2 says the marginal takes the hybrid 'to ≈0.31.' This assumes additivity of the facet marginal on the tuned baseline, which is not demonstrated; if the run record contains the actual end-to-end held-out MRR, it should be reported in the table. As published, the abstract's 'best configuration ≈0.31' and the derived '≈60×/≈15×' comparisons rest on an estimate, not a directly measured observation.

minor comments (3)

[§1, §12 vs §11] The Conclusion's 'combination outperforms every single-mechanism alternative on coverage' is in tension with the Limitations section's 'per-kind patterns, not separable pooled rankings.' Recommend aligning the summary claims with the self-assessed strength of the evidence.
[Tables 2 and 4] The label 'routed' for 'map + gated episodes' is confusing given the three-mode router of §7.4; consider 'two-mode routed' or 'routed (no synthesis).'
[References [9], [10]] References [9] and [10] use placeholder author strings ('Survey authors', 'State-contamination authors'); these should be replaced or corrected before publication.

Circularity Check

1 steps flagged

Headline token-economics claim is defined over a two-mode subset, not the full router

specific steps

self definitional [Abstract; §7.4; Table 2; Table 4; §11 Limitations]
"A router dispatches breadth to a git-anchored structural map, pointed lookups to confidence-gated episodes, and rationale to decision synthesis... Routed, the system answers at 382–980 tokens per question. [Table 4:] routed: map + gated episodes ... 382 / 980. [§11:] Specified but not run here: ... wild-question kind-distribution (and with it the expected-cost figure)."

The full router in §7.4 is defined as three modes, and 'why' questions are dispatched to decision synthesis, which Table 2 prices at 2762–3135 tokens and which supplies the 0.83 sufficiency number. The rows labeled 'routed' in Tables 2 and 4 are explicitly 'map + gated episodes' only, excluding synthesis. The abstract's 'Routed, the system answers at 382–980 tokens per question' is therefore not a measurement of the defined system; it is true only for a two-mode subset. The paper itself defers the expected-cost calculation over the real question-kind distribution to a 'future run,' so the headline token economics is made true by definitional subset choice rather than by evaluating the router as designed.

full rationale

The only concrete definitional reduction I can exhibit is the meaning of 'routed.' The architecture in §7.4 has three modes, and the router must send 'why' questions to decision synthesis, but the measured 'routed' row includes only map + gated episodes. The token-economics headline (382–980 tokens) is therefore not the cost of the full system; it is a subset result relabeled as the routed system. The rest of the empirical work is more self-contained: the seed-stage MRR study is run against internal floors (raw-transcript grep, parsed-turn grep) and held-out splits with bootstrap CIs; the facet-term mechanism is tested additively and is byte-identical at boost=0; the sufficiency ablations are transparently small and the paper itself warns that pooled contrasts are not statistically separable. The self-mined commit–session labels are a dependent source of supervision, but they are not a function of retrieval scores, so I do not count that as a circular derivation. The acknowledged limitations (n=15, single judge, 'within-system characterization') are validity caveats, not circularity. Score 4 reflects one partial circularity in a central claim while the core retrieval and per-mode results retain independent content.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 0 invented entities

No new physical entities are introduced. The ledger, structural map, and decision-synthesis mode are software components, not postulated phenomena. The free parameters are tuning choices evaluated on held-out data, though the derived best-config MRR is a concern. The axioms are the underlying domain assumptions about capture quality, evaluation validity, and label bias that, if false, would undermine the measured results.

free parameters (4)

facet_boost = 0.3
Tuned on dev split via joint re-tune; the marginal is reported on the held-out split. This is a fitted parameter affecting retrieval scores.
layer mix weights (BM25/LSA/neural) = 0.35/0.10/0.55 (ship default)
Shipped defaults, but per-corpus weight tuning is a validated lever in the mechanism study, adding +0.032 MRR on Corpus B; values are data-dependent.
confidence gate threshold for episodic injection = Not explicitly numeric; calibrated from retriever scores
Calibrated from top-1 score and top-1–top-2 gap; the gate suppresses 11/12 bad injections, so it is a tuned decision boundary.
4-gram Jaccard leakage ceiling = 0.30
Choice for label paraphrase generation; not a physical constant, but a methodological threshold.

axioms (4)

domain assumption A post-commit hook reliably captures and links every relevant agent session to the commit it produced.
This underlies the self-labeling benchmark and the entire retrieval evaluation; if the link is incomplete or wrong, the gold pairs are biased. The paper itself flags capture as the bottleneck (§9) but does not validate link accuracy.
domain assumption Answer-sufficiency as judged by a single automated model is a valid proxy for actual task help.
The sufficiency metric is the main outcome in §8; the paper acknowledges it is a proxy, not task completion, and the judge is a single model without agreement checks.
domain assumption Self-mined gold labels are biased toward lower scores, not higher.
The paper states 'label noise biases scores down, not up,' which is an assertion about the noise mechanism; this justifies using the mined labels as conservative ground truth.
domain assumption The LLM-generated structural map accurately represents the repository architecture from the directory skeleton and READMEs.
The map answers breadth questions; its quality depends on the repository's own documentation and structure, which the paper lists as a limitation (§11).

pith-pipeline@v1.3.0-alltime-deepseek · 154 in / 8012 out tokens · 142445 ms · 2026-08-02T02:12:30.301527+00:00 · methodology

0 comments

read the original abstract

Coding agents now produce a growing share of a team's code, while the reasoning behind each change -- the alternatives weighed, the constraints discovered, the approaches rejected -- is trapped in assistant transcripts that vanish with the session. Memory for this setting, the agentic development lifecycle (ADLC), is usually posed as one retrieval problem and built as machinery: tiered stores, memory graphs, compiled wikis, model-judged admission. We argue memory should instead be git-bound -- built into the repository's version control, inheriting the guarantees the machinery struggles to construct: ground truth from commits, freshness from rebuild, verification from the merge, containment from review. On this ledger we solve two problems separately, then combine them. Seed supply is closed as an eight-corpus retrieval study under a pre-registered ship discipline: five imported ranking mechanisms rejected, two kept, and a best configuration of ~0.31 pooled MRR -- ~60x the raw-transcript grep floor, ~15x an honest parsed-turn floor. Answer assembly is where ranking stops helping: single-shot retrieval scores only 0.07-0.20 answer-sufficiency on real developer questions, and ungated episode injection measurably degrades good answers. A router dispatches breadth to a git-anchored structural map, pointed lookups to confidence-gated episodes, and rationale to decision synthesis, which reconstructs why-arcs no single session contains (0.83 sufficiency on a young ~50k-LOC production system). Routed, the system answers at 382-980 tokens per question -- three orders of magnitude below the recorded history. Because ground truth is mined from commit-session links rather than annotated, every result is replicable on any user's own history at zero labeling cost. The remaining constraint is capture. Code, benchmark, and paper source: github.com/rekal-dev/rekal-cli.

Figures

Figures reproduced from arXiv: 2607.14390 by Frank Guo.

discussion (0)

Reference graph

Works this paper leans on

17 extracted references · 12 linked inside Pith

[1]

Towards retrieving interaction spaces for agentic search, 2026

Shengyao Zhuang et al. Towards retrieving interaction spaces for agentic search, 2026. URL https://arxiv. org/abs/2606.06880

Pith/arXiv arXiv 2026
[2]

Evolving agents in the dark: Retro- spective harness optimization via self-preference, 2026

Wenbo Pan et al. Evolving agents in the dark: Retro- spective harness optimization via self-preference, 2026. URLhttps://arxiv.org/abs/2606.05922

Pith/arXiv arXiv 2026
[3]

Shared selective persistent memory, 2026

Apple Machine Learning Research. Shared selective persistent memory, 2026. URL https://arxiv.org/ abs/2607.09493

Pith/arXiv arXiv 2026
[4]

Adamem: Adaptive user-centric memory for long-horizon dia- logue agents, 2026

Shannan Yan, Jingchen Ni, Leqi Zheng, et al. Adamem: Adaptive user-centric memory for long-horizon dia- logue agents, 2026. URLhttps://arxiv.org/abs/2603. 16496

2026
[5]

Memory is recon- structed, not retrieved: Graph memory for llm agents,

Shuo Ji, Yibo Li, and Bryan Hooi. Memory is recon- structed, not retrieved: Graph memory for llm agents,
[6]

Retrieval as reasoning: Self-evolving agent-native retrieval via llm-wiki, 2026

Haoliang Ming, Feifei Li, Xiaoqing Wu, Wenhui Que, et al. Retrieval as reasoning: Self-evolving agent-native retrieval via llm-wiki, 2026. URLhttps://arxiv.org/ abs/2605.25480

Pith/arXiv arXiv 2026
[7]

Escap- ing the self-confirmation trap: An execute-distill-verify paradigm for agentic experience learning, 2026

Shiding Zhu, Yudi Qi, Yajie Wang, et al. Escap- ing the self-confirmation trap: An execute-distill-verify paradigm for agentic experience learning, 2026. URL https://arxiv.org/abs/2606.24428

Pith/arXiv arXiv 2026
[8]

Evaluating very long-term conversational memory of llm agents, 2024

Adyasha Maharana et al. Evaluating very long-term conversational memory of llm agents, 2024. URLhttps: //arxiv.org/abs/2402.17753

Pith/arXiv arXiv 2024
[9]

A survey on long-term memory security in llm agents: Attacks, defenses, and governance across the memory lifecycle, 2026

Survey authors. A survey on long-term memory security in llm agents: Attacks, defenses, and governance across the memory lifecycle, 2026. URLhttps://arxiv.org/ abs/2604.16548

Pith/arXiv arXiv 2026
[10]

State contamination in memory-augmented llm agents, 2026

State-contamination authors. State contamination in memory-augmented llm agents, 2026. URL https:// arxiv.org/abs/2605.16746

Pith/arXiv arXiv 2026
[11]

Are we ready for an agent-native mem- ory system?, 2026

Wei Zhou et al. Are we ready for an agent-native mem- ory system?, 2026. URLhttps://arxiv.org/abs/2606. 24775

2026
[12]

Automem: Automated learning of memory as a cognitive skill, 2026

Shengguang Wu, Hao Zhu, Yuhui Zhang, Xiaohan Wang, and Serena Yeung-Levy. Automem: Automated learning of memory as a cognitive skill, 2026. URL https:// arxiv.org/abs/2607.01224

Pith/arXiv arXiv 2026
[13]

A simple yet strong baseline for long-term conversational memory of llm agents, 2025

Sizhe Zhou and Jiawei Han. A simple yet strong baseline for long-term conversational memory of llm agents, 2025. URLhttps://arxiv.org/abs/2511.17208

arXiv 2025
[14]

Sag: Sql- retrieval augmented generation with query-time dynamic hyperedges, 2026

Yuchao Wu, Junqin Li, XingCheng Liang, et al. Sag: Sql- retrieval augmented generation with query-time dynamic hyperedges, 2026. URL https://arxiv.org/abs/2606. 15971

2026
[15]

Beyond semantic similarity: Rethinking retrieval for agentic search via direct corpus interaction, 2026

Zhuofeng Li, Haoxiang Zhang, et al. Beyond semantic similarity: Rethinking retrieval for agentic search via direct corpus interaction, 2026. URL https://arxiv. org/abs/2605.05242

Pith/arXiv arXiv 2026
[16]

Training search agents for direct corpus interaction, 2026

GrepSeek authors. Training search agents for direct corpus interaction, 2026. URLhttps://arxiv.org/abs/ 2605.29307. 8

Pith/arXiv arXiv 2026
[2026]

URLhttps://arxiv.org/abs/2606.06036

Pith/arXiv arXiv

[1] [1]

Towards retrieving interaction spaces for agentic search, 2026

Shengyao Zhuang et al. Towards retrieving interaction spaces for agentic search, 2026. URL https://arxiv. org/abs/2606.06880

Pith/arXiv arXiv 2026

[2] [2]

Evolving agents in the dark: Retro- spective harness optimization via self-preference, 2026

Wenbo Pan et al. Evolving agents in the dark: Retro- spective harness optimization via self-preference, 2026. URLhttps://arxiv.org/abs/2606.05922

Pith/arXiv arXiv 2026

[3] [3]

Shared selective persistent memory, 2026

Apple Machine Learning Research. Shared selective persistent memory, 2026. URL https://arxiv.org/ abs/2607.09493

Pith/arXiv arXiv 2026

[4] [4]

Adamem: Adaptive user-centric memory for long-horizon dia- logue agents, 2026

Shannan Yan, Jingchen Ni, Leqi Zheng, et al. Adamem: Adaptive user-centric memory for long-horizon dia- logue agents, 2026. URLhttps://arxiv.org/abs/2603. 16496

2026

[5] [5]

Memory is recon- structed, not retrieved: Graph memory for llm agents,

Shuo Ji, Yibo Li, and Bryan Hooi. Memory is recon- structed, not retrieved: Graph memory for llm agents,

[6] [6]

Retrieval as reasoning: Self-evolving agent-native retrieval via llm-wiki, 2026

Haoliang Ming, Feifei Li, Xiaoqing Wu, Wenhui Que, et al. Retrieval as reasoning: Self-evolving agent-native retrieval via llm-wiki, 2026. URLhttps://arxiv.org/ abs/2605.25480

Pith/arXiv arXiv 2026

[7] [7]

Escap- ing the self-confirmation trap: An execute-distill-verify paradigm for agentic experience learning, 2026

Shiding Zhu, Yudi Qi, Yajie Wang, et al. Escap- ing the self-confirmation trap: An execute-distill-verify paradigm for agentic experience learning, 2026. URL https://arxiv.org/abs/2606.24428

Pith/arXiv arXiv 2026

[8] [8]

Evaluating very long-term conversational memory of llm agents, 2024

Adyasha Maharana et al. Evaluating very long-term conversational memory of llm agents, 2024. URLhttps: //arxiv.org/abs/2402.17753

Pith/arXiv arXiv 2024

[9] [9]

A survey on long-term memory security in llm agents: Attacks, defenses, and governance across the memory lifecycle, 2026

Survey authors. A survey on long-term memory security in llm agents: Attacks, defenses, and governance across the memory lifecycle, 2026. URLhttps://arxiv.org/ abs/2604.16548

Pith/arXiv arXiv 2026

[10] [10]

State contamination in memory-augmented llm agents, 2026

State-contamination authors. State contamination in memory-augmented llm agents, 2026. URL https:// arxiv.org/abs/2605.16746

Pith/arXiv arXiv 2026

[11] [11]

Are we ready for an agent-native mem- ory system?, 2026

Wei Zhou et al. Are we ready for an agent-native mem- ory system?, 2026. URLhttps://arxiv.org/abs/2606. 24775

2026

[12] [12]

Automem: Automated learning of memory as a cognitive skill, 2026

Shengguang Wu, Hao Zhu, Yuhui Zhang, Xiaohan Wang, and Serena Yeung-Levy. Automem: Automated learning of memory as a cognitive skill, 2026. URL https:// arxiv.org/abs/2607.01224

Pith/arXiv arXiv 2026

[13] [13]

A simple yet strong baseline for long-term conversational memory of llm agents, 2025

Sizhe Zhou and Jiawei Han. A simple yet strong baseline for long-term conversational memory of llm agents, 2025. URLhttps://arxiv.org/abs/2511.17208

arXiv 2025

[14] [14]

Sag: Sql- retrieval augmented generation with query-time dynamic hyperedges, 2026

Yuchao Wu, Junqin Li, XingCheng Liang, et al. Sag: Sql- retrieval augmented generation with query-time dynamic hyperedges, 2026. URL https://arxiv.org/abs/2606. 15971

2026

[15] [15]

Beyond semantic similarity: Rethinking retrieval for agentic search via direct corpus interaction, 2026

Zhuofeng Li, Haoxiang Zhang, et al. Beyond semantic similarity: Rethinking retrieval for agentic search via direct corpus interaction, 2026. URL https://arxiv. org/abs/2605.05242

Pith/arXiv arXiv 2026

[16] [16]

Training search agents for direct corpus interaction, 2026

GrepSeek authors. Training search agents for direct corpus interaction, 2026. URLhttps://arxiv.org/abs/ 2605.29307. 8

Pith/arXiv arXiv 2026

[17] [2026]

URLhttps://arxiv.org/abs/2606.06036

Pith/arXiv arXiv