row_id
string
section
string
section_slug
string
resource_type
string
marker
string
title
string
url
string
url_kind
string
domain
string
annotation
string
description
string
key_contribution
string
novelty
string
impact
string
signal
string
signal_strength
string
source_readme
string
source_line
int64
source_url
string
date_added
string
collection
string
collection_slug
string
user_goal
string
lifecycle_stages
string
audience
string
loop_layer
string
scope_fit
string
evidence_class
string
evidence_tier
string
source_status
string
canonical_url
string
source_title
string
source_description
string
authors
string
publication_date
string
publication_year
string
publication_venue
string
publisher
string
doi
string
publication_note
string
primary_category
string
metadata_source
string
github_repo
string
github_stars
string
github_forks
string
github_license
string
github_created_at
string
github_updated_at
string
arxiv_id
string
audited_at
timestamp[ms]
ale-0601
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Memory for Autonomous LLM Agents: Mechanisms, Evaluation, and Emerging Frontiers
https://arxiv.org/abs/2603.07670
external
arxiv.org
Formalizes agent memory as a write-manage-read loop and surveys compression, retrieval, reflective self-improvement, and policy-learned management across recurring runs.
Formalizes agent memory as a write-manage-read loop and surveys compression, retrieval, reflective self-improvement, and policy-learned management across recurring runs.
Formalizes agent memory as a write-manage-read loop and surveys compression, retrieval, reflective self-improvement, and policy-learned management across recurring runs.
Evaluation data is used as the feedback signal for improving loop behavior. Formalizes agent memory as a write-manage-read loop and surveys compression, retrieval, reflective self-improvement, and policy-learned management across recurring runs.
Use Memory for Autonomous LLM Agents: Mechanisms, Evaluation, and Emerging Frontiers to carry context, state, and receipts across runs and failures.
Research source arXiv:2603.07670; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,288
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1288
Persist
persist
Carry context, state, and receipts across runs.
context;verification
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2603.07670
[2603.07670] Memory for Autonomous LLM Agents:Mechanisms, Evaluation, and Emerging Frontiers
Large language model (LLM) agents increasingly operate in settings where a single context window is far too small to capture what has happened, what was learned, and what should not be repeated. Memory -- the ability to persist, organize, and selectively recall information across interactions -- is what turns a statele...
Pengfei Du
2026-03-08
2026
arXiv
arXiv
cs.AI
arxiv-api
2603.07670
2026-09-08T16:27:48
ale-0602
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Externalization in LLM Agents: A Unified Review of Memory, Skills, Protocols and Harness Engineering
https://arxiv.org/abs/2604.08224
external
arxiv.org
Reviews how durable state, reusable skills, protocols, and the harness move out of model weights into external infrastructure, the substrate that lets loops persist progress and reuse capability across runs.
Reviews how durable state, reusable skills, protocols, and the harness move out of model weights into external infrastructure, the substrate that lets loops persist progress and reuse capability across runs.
Reviews how durable state, reusable skills, protocols, and the harness move out of model weights into external infrastructure, the substrate that lets loops persist progress and reuse capability across runs.
Durable execution and replay are treated as first-class loop infrastructure. Reviews how durable state, reusable skills, protocols, and the harness move out of model weights into external infrastructure, the substrate that lets loops persist progress and reuse capability across runs.
Use Externalization in LLM Agents: A Unified Review of Memory, Skills, Protocols and Harness Engineering to carry context, state, and receipts across runs and failures.
Research source arXiv:2604.08224; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,289
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1289
Persist
persist
Carry context, state, and receipts across runs.
context;state
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2604.08224
[2604.08224] Externalization in LLM Agents: A Unified Review of Memory, Skills, Protocols and Harness Engineering
Large language model (LLM) agents are increasingly built less by changing model weights than by reorganizing the runtime around them. Capabilities that earlier systems expected the model to recover internally are now externalized into memory stores, reusable skills, interaction protocols, and the surrounding harness th...
Chenyu Zhou; Huacan Chai; Wenteng Chen; Zihan Guo; Rong Shan; Yuanyi Song; Tianyi Xu; Yingxuan Yang; Aofan Yu; Weiming Zhang; Congming Zheng; Jiachen Zhu; Zeyu Zheng; Zhuosheng Zhang; Xingyu Lou; Changwang Zhang; Zhihui Fu; Jun Wang; Weiwen Liu; Jianghao Lin; Weinan Zhang
2026-04-09
2026
arXiv
arXiv
54 pages, tech report on Externalization in LLM Agents
cs.SE
arxiv-api
2604.08224
2026-09-08T16:27:48
ale-0603
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Meta Context Engineering via Agentic Skill Evolution
https://arxiv.org/abs/2601.21557
external
arxiv.org
A bi-level loop where a meta-agent evolves reusable skills while a base-agent optimizes context, co-evolving the harness and context artifacts across runs (ICML 2026).
A bi-level loop where a meta-agent evolves reusable skills while a base-agent optimizes context, co-evolving the harness and context artifacts across runs (ICML 2026).
A bi-level loop where a meta-agent evolves reusable skills while a base-agent optimizes context, co-evolving the harness and context artifacts across runs (ICML 2026).
Context is managed as durable loop state rather than a single prompt payload. A bi-level loop where a meta-agent evolves reusable skills while a base-agent optimizes context, co-evolving the harness and context artifacts across runs (ICML 2026).
Use Meta Context Engineering via Agentic Skill Evolution to carry context, state, and receipts across runs and failures.
Research source arXiv:2601.21557; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,290
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1290
Persist
persist
Carry context, state, and receipts across runs.
context
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2601.21557
[2601.21557] Meta Context Engineering via Agentic Skill Evolution
The operational efficacy of large language models relies heavily on their inference-time context. This has established Context Engineering (CE) as a formal discipline for optimizing these inputs. Current CE methods rely on manually crafted harnesses, such as rigid generation-reflection workflows and predefined context ...
Haoran Ye; Xuning He; Vincent Arak; Haonan Dong; Guojie Song
2026-01-29
2026
arXiv
arXiv
46 pages, 4 figures
cs.AI
arxiv-api
2601.21557
2026-09-08T16:27:48
ale-0604
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Are We Ready for an Agent-Native Memory System?
https://arxiv.org/abs/2606.24775
external
arxiv.org
Evaluates twelve agent memory systems across five workloads from a data-management perspective, decomposing memory into representation, extraction, retrieval, and maintenance modules and finding localized maintenance more cost-efficient than global reorganization.
Evaluates twelve agent memory systems across five workloads from a data-management perspective, decomposing memory into representation, extraction, retrieval, and maintenance modules and finding localized maintenance more cost-efficient than global reorganization.
Evaluates twelve agent memory systems across five workloads from a data-management perspective, decomposing memory into representation, extraction, retrieval, and maintenance modules and finding localized maintenance more cost-efficient than global reorganization.
Persistent memory is treated as an external runtime artifact. Evaluates twelve agent memory systems across five workloads from a data-management perspective, decomposing memory into representation, extraction, retrieval, and maintenance modules and finding localized maintenance more cost-efficient than global reorganiz...
Use Are We Ready for an Agent-Native Memory System? to carry context, state, and receipts across runs and failures.
Research source arXiv:2606.24775; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,291
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1291
Persist
persist
Carry context, state, and receipts across runs.
context;budget
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2606.24775
[2606.24775] Are We Ready For An Agent-Native Memory System?
Memory for large language model (LLM) agents has rapidly evolved from simple retrieval-augmented mechanisms into a data management system that supports persistent information storage, retrieval, update, consolidation, and dynamic lifecycle governance throughout agent execution. Despite this evolution, existing evaluati...
Wei Zhou; Xuanhe Zhou; Shaokun Han; Hongming Xu; Guoliang Li; Zhiyu Li; Feiyu Xiong; Fan Wu
2026-06-23
2026
arXiv
arXiv
Paper list available at: https://github.com/OpenDataBox/awesome-agent-memory. Source code available at: https://github.com/OpenDataBox/MemoryData
cs.CL
arxiv-api
2606.24775
2026-09-08T16:27:48
ale-0605
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Self-Evolving World Models for LLM Agent Planning
https://arxiv.org/abs/2606.30639
external
arxiv.org
Evolves a deployment-time world model while the agent and model weights stay frozen, retrieving observed transitions, distilling rules from prediction-observation mismatches, and filtering low-confidence forecasts so each run's errors improve later planning.
Evolves a deployment-time world model while the agent and model weights stay frozen, retrieving observed transitions, distilling rules from prediction-observation mismatches, and filtering low-confidence forecasts so each run's errors improve later planning.
Evolves a deployment-time world model while the agent and model weights stay frozen, retrieving observed transitions, distilling rules from prediction-observation mismatches, and filtering low-confidence forecasts so each run's errors improve later planning.
Makes persistence and context management visible as runtime design choices. Evolves a deployment-time world model while the agent and model weights stay frozen, retrieving observed transitions, distilling rules from prediction-observation mismatches, and filtering low-confidence forecasts so each run's errors improve l...
Use Self-Evolving World Models for LLM Agent Planning to carry context, state, and receipts across runs and failures.
Research source arXiv:2606.30639; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,292
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1292
Persist
persist
Carry context, state, and receipts across runs.
context;state
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2606.30639
[2606.30639] Self-Evolving World Models for LLM Agent Planning
World models offer a principled way to equip long-horizon LLM agents with foresight: predictions of action consequences before execution. However, unreliable foresight can be ignored, misused, or even degrade downstream decision-making. In this paper, we introduce WorldEvolver, a self-evolving world model framework tha...
Xuan Zhang; Wenxuan Zhang; See-Kiong Ng; Yang Deng
2026-06-29
2026
arXiv
arXiv
Accepted at EMNLP 2026 Findings
cs.AI
arxiv-api
2606.30639
2026-09-08T16:27:48
ale-0606
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Rethinking Continual Experience Internalization for Self-Evolving LLM Agents
https://arxiv.org/abs/2606.04703
external
arxiv.org
Finds that naively re-internalizing accumulated experience causes progressive capability collapse across self-improvement iterations, and identifies what keeps the loop stable: principle-level abstractions, step-wise injection for tool use, and off-policy distillation from stronger teacher trajectories.
Finds that naively re-internalizing accumulated experience causes progressive capability collapse across self-improvement iterations, and identifies what keeps the loop stable: principle-level abstractions, step-wise injection for tool use, and off-policy distillation from stronger teacher trajectories.
Finds that naively re-internalizing accumulated experience causes progressive capability collapse across self-improvement iterations, and identifies what keeps the loop stable: principle-level abstractions, step-wise injection for tool use, and off-policy distillation from stronger teacher trajectories.
Makes persistence and context management visible as runtime design choices. Finds that naively re-internalizing accumulated experience causes progressive capability collapse across self-improvement iterations, and identifies what keeps the loop stable: principle-level abstractions, step-wise injection for tool use, and...
Use Rethinking Continual Experience Internalization for Self-Evolving LLM Agents to carry context, state, and receipts across runs and failures.
Research source arXiv:2606.04703; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,293
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1293
Persist
persist
Carry context, state, and receipts across runs.
workspace
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2606.04703
[2606.04703] Rethinking Continual Experience Internalization for Self-Evolving LLM Agents
Experience internalization converts contextual experience from past interactions into reusable parametric capability, offering a promising path toward continual learning in large language models (LLMs). While prior work has predominantly focused on single-iteration transfer, we discover that under multi-iteration exper...
Jingwen Chen; Wenkai Yang; Shengda Fan; Wenbo Nie; Chenxing Sun; Shaodong Zheng; Yangen Hu; Lu Pan; Ke Zeng; Yankai Lin
2026-06-03
2026
arXiv
arXiv
10 pages, 8 figures
cs.CL
arxiv-api
2606.04703
2026-09-08T16:27:48
ale-0607
State, Memory, And Context Persistence
state-memory-and-context-persistence
Tool
🧰
GenericAgent
https://github.com/lsdefine/GenericAgent
external
github.com
Self-evolving agent that grows a skill tree from a small seed, crystallizing completed runs into layered memory and reusable skills, with a master-worker mode for long-horizon goals.
Self-evolving agent that grows a skill tree from a small seed, crystallizing completed runs into layered memory and reusable skills, with a master-worker mode for long-horizon goals.
Self-evolving agent that grows a skill tree from a small seed, crystallizing completed runs into layered memory and reusable skills, with a master-worker mode for long-horizon goals.
Persistent memory is treated as an external runtime artifact. Self-evolving agent that grows a skill tree from a small seed, crystallizing completed runs into layered memory and reusable skills, with a master-worker mode for long-horizon goals.
Use GenericAgent to carry context, state, and receipts across runs and failures.
Inspectable GitHub source; popularity is context, not proof of reliability.
medium
README.md
1,294
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1294
Persist
persist
Carry context, state, and receipts across runs.
objective;context
builder
harness
enabling
source-implementation
A
ok
https://github.com/lsdefine/GenericAgent
GitHub - lsdefine/GenericAgent: Self-evolving agent: grows skill tree from 3.3K-line seed, achieving full system control with 6x less token consumption · GitHub
Self-evolving agent: grows skill tree from 3.3K-line seed, achieving full system control with 6x less token consumption - lsdefine/GenericAgent
lsdefine/GenericAgent
GitHub
html-meta
lsdefine/GenericAgent
2026-09-08T16:27:48
ale-0608
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Self-GC: Self-Governing Context for Long-Horizon LLM Agents
https://arxiv.org/abs/2607.00692
external
arxiv.org
Governs long-horizon agent context as indexed lifecycle objects in an explicit nod to garbage collection, with a side-channel planner proposing fold, mask, and prune actions under harness-enforced recoverable sidecars, cutting production input tokens by 10-15%.
Governs long-horizon agent context as indexed lifecycle objects in an explicit nod to garbage collection, with a side-channel planner proposing fold, mask, and prune actions under harness-enforced recoverable sidecars, cutting production input tokens by 10-15%.
Governs long-horizon agent context as indexed lifecycle objects in an explicit nod to garbage collection, with a side-channel planner proposing fold, mask, and prune actions under harness-enforced recoverable sidecars, cutting production input tokens by 10-15%.
Context is managed as durable loop state rather than a single prompt payload. Governs long-horizon agent context as indexed lifecycle objects in an explicit nod to garbage collection, with a side-channel planner proposing fold, mask, and prune actions under harness-enforced recoverable sidecars, cutting production inpu...
Use Self-GC: Self-Governing Context for Long-Horizon LLM Agents to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.00692; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,295
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1295
Persist
persist
Carry context, state, and receipts across runs.
context;budget
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.00692
[2607.00692] Self-GC: Self-Governing Context for Long-Horizon LLM Agents
Long-horizon LLM agents accumulate tool results, files, plans, and user constraints that are too structured to be treated as a disposable text suffix. Current systems mostly rely on in-run heuristics such as chronological pruning and tool-output masking, or on final self-summary near a context limit. Heuristics are che...
Xubin Hao; Hongjin Meng; Xin Yin; Jiawei Zhu; Chenpeng Cao
2026-07-01
2026
arXiv
arXiv
cs.AI
arxiv-api
2607.00692
2026-09-08T16:27:48
ale-0609
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
CompactionRL: Reinforcement Learning with Context Compaction for Long-Horizon Agents
https://arxiv.org/abs/2607.05378
external
arxiv.org
Reinforcement-learning method that jointly optimizes task execution and compaction-summary generation so long-horizon agents can continue past finite context windows, lifting GLM-4.5-Air to 66.8% on SWE-bench Verified and shipping in the GLM-5.2 pipeline.
Reinforcement-learning method that jointly optimizes task execution and compaction-summary generation so long-horizon agents can continue past finite context windows, lifting GLM-4.5-Air to 66.8% on SWE-bench Verified and shipping in the GLM-5.2 pipeline.
Reinforcement-learning method that jointly optimizes task execution and compaction-summary generation so long-horizon agents can continue past finite context windows, lifting GLM-4.5-Air to 66.8% on SWE-bench Verified and shipping in the GLM-5.2 pipeline.
Verification is promoted from a final check to a loop-control signal. Reinforcement-learning method that jointly optimizes task execution and compaction-summary generation so long-horizon agents can continue past finite context windows, lifting GLM-4.5-Air to 66.8% on SWE-bench Verified and shipping in the GLM-5.2 pipe...
Use CompactionRL: Reinforcement Learning with Context Compaction for Long-Horizon Agents to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.05378; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,296
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1296
Persist
persist
Carry context, state, and receipts across runs.
context;verification
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.05378
[2607.05378] CompactionRL: Reinforcement Learning with Context Compaction for Long-Horizon Agents
Long-horizon agentic LLMs are increasingly limited by finite context windows, as extended interaction trajectories can exceed the maximum context length before a task is completed. Context compaction offers a natural solution by summarizing previous interaction states and continuing the rollout under a compressed conte...
Yujiang Li; Zhenyu Hou; Yi Jing; Jie Tang; Yuxiao Dong
2026-07-06
2026
arXiv
arXiv
cs.LG
arxiv-api
2607.05378
2026-09-08T16:27:48
ale-0610
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
SelfMem: Self-Optimizing Memory for AI Agents
https://arxiv.org/abs/2607.03726
external
arxiv.org
Memory framework in which the agent autonomously optimizes its own storage, retrieval, and summarization strategies per task instead of a fixed pipeline, improving BEAM's official score over the strongest baseline by 48.7%, 40.8%, and 41.9% at 100K, 500K, and 1M tokens, respectively.
Memory framework in which the agent autonomously optimizes its own storage, retrieval, and summarization strategies per task instead of a fixed pipeline, improving BEAM's official score over the strongest baseline by 48.7%, 40.8%, and 41.9% at 100K, 500K, and 1M tokens, respectively.
Memory framework in which the agent autonomously optimizes its own storage, retrieval, and summarization strategies per task instead of a fixed pipeline, improving BEAM's official score over the strongest baseline by 48.7%, 40.8%, and 41.9% at 100K, 500K, and 1M tokens, respectively.
Primary-source operational guidance rather than commentary. Memory framework in which the agent autonomously optimizes its own storage, retrieval, and summarization strategies per task instead of a fixed pipeline, improving BEAM's official score over the strongest baseline by 48.7%, 40.8%, and 41.9% at 100K, 500K, and ...
Use SelfMem: Self-Optimizing Memory for AI Agents to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.03726; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,297
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1297
Persist
persist
Carry context, state, and receipts across runs.
context;budget
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.03726
[2607.03726] SelfMem: Self-Optimizing Memory for AI Agents
While current AI agents support increasingly long context windows, tool use, and skill execution for long-horizon tasks, they still require memory systems to effectively leverage historical experience. Existing memory frameworks typically rely on fixed storage, retrieval, and summarization mechanisms, which can be rigi...
Shu Yang; Junchao Wu; Derek F. Wong; Di Wang
2026-07-04
2026
arXiv
arXiv
cs.CL
arxiv-api
2607.03726
2026-09-08T16:27:48
ale-0611
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Memory-Orchestrated Semantic System (MOSS): An Auditable Agentic Memory Architecture
https://arxiv.org/abs/2607.04391
external
arxiv.org
Model-, storage-, and API-agnostic agent memory architecture where the agent drives symbolic retrieval over a structured relational database, making long-term memory auditable and reproducible instead of opaque embedding search, validated in a year-long deployment.
Model-, storage-, and API-agnostic agent memory architecture where the agent drives symbolic retrieval over a structured relational database, making long-term memory auditable and reproducible instead of opaque embedding search, validated in a year-long deployment.
Model-, storage-, and API-agnostic agent memory architecture where the agent drives symbolic retrieval over a structured relational database, making long-term memory auditable and reproducible instead of opaque embedding search, validated in a year-long deployment.
Persistent memory is treated as an external runtime artifact. Model-, storage-, and API-agnostic agent memory architecture where the agent drives symbolic retrieval over a structured relational database, making long-term memory auditable and reproducible instead of opaque embedding search, validated in a year-long depl...
Use Memory-Orchestrated Semantic System (MOSS): An Auditable Agentic Memory Architecture to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.04391; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,298
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1298
Persist
persist
Carry context, state, and receipts across runs.
context;delegation
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.04391
[2607.04391] Memory-Orchestrated Semantic System (MOSS): An Auditable Agentic Memory Architecture
Long-term memory remains a structural weakness of AI agents. The dominant approach, retrieval-augmented generation (RAG), relies on embedding-based similarity search, which is opaque by construction, difficult to audit, and bounded by the theoretical limits of vector representations. We present the Memory-Orchestrated ...
Serge Lacasse; Jérémie Hatier; Alex Baker
2026-07-05
2026
arXiv
arXiv
22 pages, 2 figures
cs.CL
arxiv-api
2607.04391
2026-09-08T16:27:48
ale-0612
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
The Log Is the Agent: Event-Sourced Reactive Graphs for Auditable, Forkable Agentic Systems
https://arxiv.org/abs/2605.21997
external
arxiv.org
BabyAGI creator Yohei Nakajima makes an append-only event log the source of truth and the working graph a deterministic projection, giving long-running loops deterministic replay, cheap forking at any event, and end-to-end causal lineage from goal to model call.
BabyAGI creator Yohei Nakajima makes an append-only event log the source of truth and the working graph a deterministic projection, giving long-running loops deterministic replay, cheap forking at any event, and end-to-end causal lineage from goal to model call.
BabyAGI creator Yohei Nakajima makes an append-only event log the source of truth and the working graph a deterministic projection, giving long-running loops deterministic replay, cheap forking at any event, and end-to-end causal lineage from goal to model call.
Durable execution and replay are treated as first-class loop infrastructure. BabyAGI creator Yohei Nakajima makes an append-only event log the source of truth and the working graph a deterministic projection, giving long-running loops deterministic replay, cheap forking at any event, and end-to-end causal lineage from ...
Use The Log Is the Agent: Event-Sourced Reactive Graphs for Auditable, Forkable Agentic Systems to carry context, state, and receipts across runs and failures.
Research source arXiv:2605.21997; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,299
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1299
Persist
persist
Carry context, state, and receipts across runs.
objective;state
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2605.21997
[2605.21997] The Log is the Agent: Event-Sourced Reactive Graphs for Auditable, Forkable Agentic Systems
Most agent frameworks are built around the language model: a conversation loop comes first, then tools, then rules, and finally a logging layer bolted on for observability, with state persisted as retrievable "memory." We describe ActiveGraph, a runtime that inverts this arrangement. The append-only event log is the so...
Yohei Nakajima
2026-05-21
2026
arXiv
arXiv
11 pages, 1 figure. Open-source Apache-2.0 implementation with reproducible quickstart demo, deterministic replay, fork-and-diff, and lineage tracing
cs.AI
arxiv-api
2605.21997
2026-09-08T16:27:48
ale-0613
State, Memory, And Context Persistence
state-memory-and-context-persistence
Blog
📝
Agentics: Memorizing Session Transcripts Isn't Useful
https://12gramsofcarbon.com/p/agentics-memorizing-session-transcripts
external
12gramsofcarbon.com
From thousands of agent sessions at Nori, reports zero coding-task benefit from giving agents search over prior session transcripts and argues loop state belongs in distilled artifacts like commits and docs because agents never prune stale context.
From thousands of agent sessions at Nori, reports zero coding-task benefit from giving agents search over prior session transcripts and argues loop state belongs in distilled artifacts like commits and docs because agents never prune stale context.
From thousands of agent sessions at Nori, reports zero coding-task benefit from giving agents search over prior session transcripts and argues loop state belongs in distilled artifacts like commits and docs because agents never prune stale context.
Context is managed as durable loop state rather than a single prompt payload. From thousands of agent sessions at Nori, reports zero coding-task benefit from giving agents search over prior session transcripts and argues loop state belongs in distilled artifacts like commits and docs because agents never prune stale co...
Use Agentics: Memorizing Session Transcripts Isn't Useful to carry context, state, and receipts across runs and failures.
Contextual source from 12gramsofcarbon.com; useful for practice signals or boundary conditions, not independent validation.
contextual
README.md
1,300
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1300
Persist
persist
Carry context, state, and receipts across runs.
context;state
builder
harness
enabling
practitioner-analysis
B
ok
https://12gramsofcarbon.com/p/agentics-memorizing-session-transcripts
Agentics: Memorizing Session Transcripts Isn't Useful
Keep track of artifacts, not scratch. Alt title: Claude, please stop trying to memorize random crap
theahura
12gramsofcarbon.com
html-meta
2026-09-08T16:27:48
ale-0614
State, Memory, And Context Persistence
state-memory-and-context-persistence
Blog
📝
Long-Running Agents
https://addyo.substack.com/p/long-running-agents
external
addyo.substack.com
Addy Osmani's essay on the infrastructure behind agents that run for hours or days, naming three walls (finite context, missing persistent state, unreliable self-verification) and the patterns that address them: durable event logs, checkpoint-and-resume, external state, and a planner/worker/judge split.
Addy Osmani's essay on the infrastructure behind agents that run for hours or days, naming three walls (finite context, missing persistent state, unreliable self-verification) and the patterns that address them: durable event logs, checkpoint-and-resume, external state, and a planner/worker/judge split.
Addy Osmani's essay on the infrastructure behind agents that run for hours or days, naming three walls (finite context, missing persistent state, unreliable self-verification) and the patterns that address them: durable event logs, checkpoint-and-resume, external state, and a planner/worker/judge split.
Durable execution and replay are treated as first-class loop infrastructure. Addy Osmani's essay on the infrastructure behind agents that run for hours or days, naming three walls (finite context, missing persistent state, unreliable self-verification) and the patterns that address them: durable event logs, checkpoint-...
Use Long-Running Agents to carry context, state, and receipts across runs and failures.
Contextual source from addyo.substack.com; useful for practice signals or boundary conditions, not independent validation.
contextual
README.md
1,301
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1301
Persist
persist
Carry context, state, and receipts across runs.
context;verification;state
builder
harness
enabling
practitioner-analysis
B
ok
https://addyo.substack.com/p/long-running-agents
Long-running Agents - by Addy Osmani - Elevate
A long-running AI agent can keep making progress over hours, days, or weeks.
Addy Osmani
Substack
html-meta
2026-09-08T16:27:48
ale-0615
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
StateFuse: Deterministic Conflict-Preserving Memory for Multi-Agent Systems
https://arxiv.org/abs/2607.05844
external
arxiv.org
Conflict-aware replicated memory contract with immutable history, explicit conflict objects, and projection-time resolution, so agent systems that accumulate contradictory observations across branches, retries, and replicas record state auditably instead of silently overwriting it.
Conflict-aware replicated memory contract with immutable history, explicit conflict objects, and projection-time resolution, so agent systems that accumulate contradictory observations across branches, retries, and replicas record state auditably instead of silently overwriting it.
Conflict-aware replicated memory contract with immutable history, explicit conflict objects, and projection-time resolution, so agent systems that accumulate contradictory observations across branches, retries, and replicas record state auditably instead of silently overwriting it.
Persistent memory is treated as an external runtime artifact. Conflict-aware replicated memory contract with immutable history, explicit conflict objects, and projection-time resolution, so agent systems that accumulate contradictory observations across branches, retries, and replicas record state auditably instead of ...
Use StateFuse: Deterministic Conflict-Preserving Memory for Multi-Agent Systems to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.05844; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,302
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1302
Persist
persist
Carry context, state, and receipts across runs.
context;delegation;state;budget
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.05844
[2607.05844] StateFuse: Deterministic Conflict-Preserving Memory for Multi-Agent Systems
Agent systems accumulate conflicting observations across branches, retries, and replicas, yet many practical memory layers still collapse disagreement behind overwrite rules that are difficult to inspect or correct. We present StateFuse, a conflict-aware replicated memory contract built on standard OpSet/CRDT merge. St...
Sergey Volkov; Yang Li; Ye Luo
2026-07-07
2026
arXiv
arXiv
Code and supplementary materials available at: https://github.com/nZiben/statefuse
cs.AI
arxiv-api
2607.05844
2026-09-08T16:27:48
ale-0616
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Remember When It Matters: Proactive Memory Agent for Long-Horizon Agents
https://arxiv.org/abs/2607.08716
external
arxiv.org
Names the failure mode "behavioral state decay" (decision-relevant state such as prior attempts, diagnoses, and open subgoals gets buried or evicted as trajectories grow) and pairs the action agent with a proactive memory agent that maintains a structured memory bank and injects memory-grounded reminders only when need...
Names the failure mode "behavioral state decay" (decision-relevant state such as prior attempts, diagnoses, and open subgoals gets buried or evicted as trajectories grow) and pairs the action agent with a proactive memory agent that maintains a structured memory bank and injects memory-grounded reminders only when need...
Names the failure mode "behavioral state decay" (decision-relevant state such as prior attempts, diagnoses, and open subgoals gets buried or evicted as trajectories grow) and pairs the action agent with a proactive memory agent that maintains a structured memory bank and injects memory-grounded reminders only when need...
Persistent memory is treated as an external runtime artifact. Names the failure mode "behavioral state decay" (decision-relevant state such as prior attempts, diagnoses, and open subgoals gets buried or evicted as trajectories grow) and pairs the action agent with a proactive memory agent that maintains a structured me...
Use Remember When It Matters: Proactive Memory Agent for Long-Horizon Agents to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.08716; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,303
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1303
Persist
persist
Carry context, state, and receipts across runs.
context;state
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.08716
[2607.08716] Remember When It Matters: Proactive Memory Agent for Long-Horizon Agents
In long-horizon tasks, decision-relevant state is often scattered across an expanding trajectory, while the action agent must surface it and act. As trajectories grow, task requirements, environment facts, prior attempts, diagnoses, and open subgoals can be buried in the context window or pushed beyond it, failing to i...
Yifan Wu; Lizhu Zhang; Yuhang Zhou; Mingyi Wang; Bo Peng; Serena Li; Xiangjun Fan; Zhuokai Zhao
2026-07-09
2026
arXiv
arXiv
cs.AI
arxiv-api
2607.08716
2026-09-08T16:27:48
ale-0617
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
What to Keep, What to Forget: A Rate-Distortion View of Memory Compaction
https://arxiv.org/abs/2607.08032
external
arxiv.org
Rate-distortion framing that unifies KV-cache eviction, prompt compression, recurrent state, and cross-session agent memory compaction under a single objective with a layer-agnostic lower bound, showing why attention- and recency-based eviction discards information before future queries reveal what mattered.
Rate-distortion framing that unifies KV-cache eviction, prompt compression, recurrent state, and cross-session agent memory compaction under a single objective with a layer-agnostic lower bound, showing why attention- and recency-based eviction discards information before future queries reveal what mattered.
Rate-distortion framing that unifies KV-cache eviction, prompt compression, recurrent state, and cross-session agent memory compaction under a single objective with a layer-agnostic lower bound, showing why attention- and recency-based eviction discards information before future queries reveal what mattered.
Persistent memory is treated as an external runtime artifact. Rate-distortion framing that unifies KV-cache eviction, prompt compression, recurrent state, and cross-session agent memory compaction under a single objective with a layer-agnostic lower bound, showing why attention- and recency-based eviction discards info...
Use What to Keep, What to Forget: A Rate-Distortion View of Memory Compaction to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.08032; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,304
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1304
Persist
persist
Carry context, state, and receipts across runs.
objective;context;state
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.08032
[2607.08032] What to Keep, What to Forget: A Rate--Distortion View of Memory Compaction in LLMs and Agents
Large language models, and the agents built on them, spend an ever-growing share of their compute and memory on remembering: caching attention keys and values, carrying long prompts, maintaining recurrent state, and storing what happened in previous turns and sessions. Because none of this memory is free, four largely ...
Ashwin Gerard Colaco; Nada Lahjouji
2026-07-09
2026
arXiv
arXiv
cs.LG
arxiv-api
2607.08032
2026-09-08T16:27:48
ale-0618
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
A Hierarchical Memory Architecture Overcomes Context Limits in Long-Horizon Multi-Agent Modeling
https://arxiv.org/abs/2607.07666
external
arxiv.org
Three-layer hierarchical memory keeps injected context bounded and roughly constant across multi-session, long-horizon multi-agent workflows, demonstrated by the Ensemble QSP framework autonomously selecting pharmacokinetic-pharmacodynamic models across 104 runs with overseer agents handling verification and troublesho...
Three-layer hierarchical memory keeps injected context bounded and roughly constant across multi-session, long-horizon multi-agent workflows, demonstrated by the Ensemble QSP framework autonomously selecting pharmacokinetic-pharmacodynamic models across 104 runs with overseer agents handling verification and troublesho...
Three-layer hierarchical memory keeps injected context bounded and roughly constant across multi-session, long-horizon multi-agent workflows, demonstrated by the Ensemble QSP framework autonomously selecting pharmacokinetic-pharmacodynamic models across 104 runs with overseer agents handling verification and troublesho...
Verification is promoted from a final check to a loop-control signal. Three-layer hierarchical memory keeps injected context bounded and roughly constant across multi-session, long-horizon multi-agent workflows, demonstrated by the Ensemble QSP framework autonomously selecting pharmacokinetic-pharmacodynamic models acr...
Use A Hierarchical Memory Architecture Overcomes Context Limits in Long-Horizon Multi-Agent Modeling to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.07666; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,305
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1305
Persist
persist
Carry context, state, and receipts across runs.
context;delegation;verification
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.07666
[2607.07666] A hierarchical memory architecture overcomes context limits in long-horizon multi-agent computational modeling
Large language models (LLMs) demonstrate remarkable reasoning capabilities, yet their stateless architecture fundamentally limits deployment in long-horizon research workflows requiring multi-session continuity and quantitative rigor. Here we present Ensemble QSP, a multi-agent framework featuring a three-layer hierarc...
Shivendra G. Tewari; Holly Kimko
2026-07-08
2026
arXiv
arXiv
19 pages, 4 figures, 2 tables. Preprint submitted for publication
q-bio.QM
arxiv-api
2607.07666
2026-09-08T16:27:48
ale-0619
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
SkillCenter: A Large-Scale Source-Grounded Skill Library for Autonomous AI Agents
https://arxiv.org/abs/2607.07676
external
arxiv.org
Claims the largest open skill library for agents (216,938 structured skills across 24 domain bundles), built with an LLM quality gate (SkillGate) and iterative source-grounding that maps each retained claim to an exact source quotation, and shipped as offline-searchable SQLite FTS5 bundles, infrastructure for skills-as...
Claims the largest open skill library for agents (216,938 structured skills across 24 domain bundles), built with an LLM quality gate (SkillGate) and iterative source-grounding that maps each retained claim to an exact source quotation, and shipped as offline-searchable SQLite FTS5 bundles, infrastructure for skills-as...
Claims the largest open skill library for agents (216,938 structured skills across 24 domain bundles), built with an LLM quality gate (SkillGate) and iterative source-grounding that maps each retained claim to an exact source quotation, and shipped as offline-searchable SQLite FTS5 bundles, infrastructure for skills-as...
State persistence is explicit enough for repeated runs and handoff. Claims the largest open skill library for agents (216,938 structured skills across 24 domain bundles), built with an LLM quality gate (SkillGate) and iterative source-grounding that maps each retained claim to an exact source quotation, and shipped as ...
Use SkillCenter: A Large-Scale Source-Grounded Skill Library for Autonomous AI Agents to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.07676; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,306
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1306
Persist
persist
Carry context, state, and receipts across runs.
state
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.07676
[2607.07676] SkillCenter: A Large-Scale Source-Grounded Skill Library for Autonomous AI Agents
Autonomous AI agents can execute complex tasks with limited human review, yet they often lack the grounded operational knowledge to make their outputs not just executable but correct, secure, and maintainable. We introduce SkillCenter, to our knowledge the largest open skill library for agents by total count: 216,938 s...
Tianming Sha; Yue Zhao; Lichao Sun; Yushun Dong
2026-07-08
2026
arXiv
arXiv
44 pages, 5 figures. Code: https://github.com/LabRAI/SkillCenter ; Data: https://huggingface.co/datasets/Tommysha/skillcenter-bundles
cs.AI
arxiv-api
2607.07676
2026-09-08T16:27:48
ale-0620
State, Memory, And Context Persistence
state-memory-and-context-persistence
Blog
📝
How version control will evolve for the agent boom
https://entire.io/blog/how-version-control-will-evolve-for-the-agent-boom
external
entire.io
Thomas Dohmke (former GitHub CEO, now founder of Entire) argues that agent session logs (prompts, tool calls, and decision checkpoints) are becoming the most important artifact in software development and should be versioned alongside code so agent fleets stop repeating mistakes, and that Git hosting must re-decentrali...
Thomas Dohmke (former GitHub CEO, now founder of Entire) argues that agent session logs (prompts, tool calls, and decision checkpoints) are becoming the most important artifact in software development and should be versioned alongside code so agent fleets stop repeating mistakes, and that Git hosting must re-decentrali...
Thomas Dohmke (former GitHub CEO, now founder of Entire) argues that agent session logs (prompts, tool calls, and decision checkpoints) are becoming the most important artifact in software development and should be versioned alongside code so agent fleets stop repeating mistakes, and that Git hosting must re-decentrali...
Checkpointed state makes long-running agent work recoverable across failures. Thomas Dohmke (former GitHub CEO, now founder of Entire) argues that agent session logs (prompts, tool calls, and decision checkpoints) are becoming the most important artifact in software development and should be versioned alongside code so...
Use How version control will evolve for the agent boom to carry context, state, and receipts across runs and failures.
Contextual source from entire.io; useful for practice signals or boundary conditions, not independent validation.
contextual
README.md
1,307
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1307
Persist
persist
Carry context, state, and receipts across runs.
workspace;state;exit
builder
harness
enabling
practitioner-analysis
B
ok
https://entire.io/blog/how-version-control-will-evolve-for-the-agent-boom
How Version Control Will Evolve for the Agent Boom · Entire
To meet the demand of the agent boom, Git hosting must return to its original promise: a distributed network of many hosts.
2026-07-06
2026
Entire
html-meta
2026-09-08T16:27:48
ale-0621
State, Memory, And Context Persistence
state-memory-and-context-persistence
Tool
🧰
self-learning-skills
https://github.com/Kulaxyz/self-learning-skills
external
github.com
Meta-skill for Claude Code, Cursor, and AGENTS.md-compatible agents that recognizes when a session has earned a hard-won golden path (or hit a dead-end worth remembering), distills the procedure including failed approaches, and persists it as a skill or rule auto-loaded next run, turning each session's discoveries into...
Meta-skill for Claude Code, Cursor, and AGENTS.md-compatible agents that recognizes when a session has earned a hard-won golden path (or hit a dead-end worth remembering), distills the procedure including failed approaches, and persists it as a skill or rule auto-loaded next run, turning each session's discoveries into...
Meta-skill for Claude Code, Cursor, and AGENTS.md-compatible agents that recognizes when a session has earned a hard-won golden path (or hit a dead-end worth remembering), distills the procedure including failed approaches, and persists it as a skill or rule auto-loaded next run, turning each session's discoveries into...
Durable execution and replay are treated as first-class loop infrastructure. Meta-skill for Claude Code, Cursor, and AGENTS.md-compatible agents that recognizes when a session has earned a hard-won golden path (or hit a dead-end worth remembering), distills the procedure including failed approaches, and persists it as ...
Use self-learning-skills to carry context, state, and receipts across runs and failures.
Inspectable GitHub source; popularity is context, not proof of reliability.
medium
README.md
1,308
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1308
Persist
persist
Carry context, state, and receipts across runs.
state
builder
harness
enabling
source-implementation
A
ok
https://github.com/Kulaxyz/self-learning-skills
GitHub - Kulaxyz/self-learning-skills: A self-improving skill for AI coding agents (Claude Code, Cursor, AGENTS.md): recognize a hard-won golden path in a session and harvest it into a reusable skill/rule for next time. · GitHub
A self-improving skill for AI coding agents (Claude Code, Cursor, AGENTS.md): recognize a hard-won golden path in a session and harvest it into a reusable skill/rule for next time. - Kulaxyz/self-learning-skills
Kulaxyz/self-learning-skills
GitHub
html-meta
Kulaxyz/self-learning-skills
2026-09-08T16:27:48
ale-0622
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
GitLake: Git-for-data for the agentic lakehouse
https://arxiv.org/abs/2607.08319
external
arxiv.org
Git-for-data design for an agent-first lakehouse that lifts single-table Iceberg snapshots into lakehouse-wide commits, branches, and merges, so agents work on isolated branches while humans review and publish, and pipeline outputs become visible atomically or not at all, with production lessons and correctness insight...
Git-for-data design for an agent-first lakehouse that lifts single-table Iceberg snapshots into lakehouse-wide commits, branches, and merges, so agents work on isolated branches while humans review and publish, and pipeline outputs become visible atomically or not at all, with production lessons and correctness insight...
Git-for-data design for an agent-first lakehouse that lifts single-table Iceberg snapshots into lakehouse-wide commits, branches, and merges, so agents work on isolated branches while humans review and publish, and pipeline outputs become visible atomically or not at all, with production lessons and correctness insight...
Makes persistence and context management visible as runtime design choices. Git-for-data design for an agent-first lakehouse that lifts single-table Iceberg snapshots into lakehouse-wide commits, branches, and merges, so agents work on isolated branches while humans review and publish, and pipeline outputs become visib...
Use GitLake: Git-for-data for the agentic lakehouse to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.08319; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,309
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1309
Persist
persist
Carry context, state, and receipts across runs.
context;state
researcher;evaluator
harness
enabling
research-paper
A
ok
https://arxiv.org/abs/2607.08319
[2607.08319] GitLake: Git-for-data for the agentic lakehouse
We present GitLake, a Git-for-data design for an agent-first lakehouse. The system lifts single-table Iceberg snapshots into lakehouse-wide commits, branches, and merges, letting agents work on isolated branches while humans review and publish changes. Pipelines run on temporary branches and publish through a final mer...
Weiming Sheng; Jinlang Wang; Manuel Barros; Aldrin Montana; Jacopo Tagliabue; Luca Bigon
2026
2026
DASHSys Workshop at the International Conference on Very Large Data Bases (VLDB)
VLDB Endowment
Accepted at DASHSys Workshop at the International Conference on Very Large Data Bases (VLDB); the linked arXiv record is the available paper version.
cs.DB
Current arXiv acceptance note and official workshop page
2607.08319
2026-09-08T16:27:48
ale-0623
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Shared Selective Persistent Memory for Agentic LLM Systems
https://arxiv.org/abs/2607.09493
external
arxiv.org
Architecture that selectively persists four categories of reusable context (task specifications, data schemas, tool configurations, output constraints) across agent sessions while discarding session-specific reasoning traces, with cross-user sharing under access controls and a zero-token refresh path for recurring data...
Architecture that selectively persists four categories of reusable context (task specifications, data schemas, tool configurations, output constraints) across agent sessions while discarding session-specific reasoning traces, with cross-user sharing under access controls and a zero-token refresh path for recurring data...
Architecture that selectively persists four categories of reusable context (task specifications, data schemas, tool configurations, output constraints) across agent sessions while discarding session-specific reasoning traces, with cross-user sharing under access controls and a zero-token refresh path for recurring data...
Persistent memory is treated as an external runtime artifact. Architecture that selectively persists four categories of reusable context (task specifications, data schemas, tool configurations, output constraints) across agent sessions while discarding session-specific reasoning traces, with cross-user sharing under ac...
Use Shared Selective Persistent Memory for Agentic LLM Systems to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.09493; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,310
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1310
Persist
persist
Carry context, state, and receipts across runs.
workspace;context;state;budget;exit
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.09493
[2607.09493] Shared Selective Persistent Memory for Agentic LLM Systems
Agentic LLM systems that generate code through multi-turn tool use face a fundamental context problem: each session starts from zero, discarding the configuration choices, domain constraints, data schemas, and tool-use patterns that made previous sessions productive. Naively persisting entire conversation histories is ...
Sanjana Pedada; Aditya Dhavala; Neelraj Patil
2026-07-10
2026
arXiv
arXiv
11 pages, 2 figures, 4 tables
cs.AI
arxiv-api
2607.09493
2026-09-08T16:27:48
ale-0624
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Scoped Verification for Reliable Long-Horizon Agentic Context Evolution
https://arxiv.org/abs/2607.09175
external
arxiv.org
GRACE represents an agent's persistent instructions as a typed semantic graph and runs scoped verification over the local neighborhood of each proposed edit before committing it, so accumulated context evolves reliably across long-horizon deployment under distribution shift instead of drifting as flat text; evaluated o...
GRACE represents an agent's persistent instructions as a typed semantic graph and runs scoped verification over the local neighborhood of each proposed edit before committing it, so accumulated context evolves reliably across long-horizon deployment under distribution shift instead of drifting as flat text; evaluated o...
GRACE represents an agent's persistent instructions as a typed semantic graph and runs scoped verification over the local neighborhood of each proposed edit before committing it, so accumulated context evolves reliably across long-horizon deployment under distribution shift instead of drifting as flat text; evaluated o...
Control flow is represented as an inspectable graph rather than an opaque prompt loop. GRACE represents an agent's persistent instructions as a typed semantic graph and runs scoped verification over the local neighborhood of each proposed edit before committing it, so accumulated context evolves reliably across long-ho...
Use Scoped Verification for Reliable Long-Horizon Agentic Context Evolution to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.09175; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,311
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1311
Persist
persist
Carry context, state, and receipts across runs.
context;verification;state
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.09175
[2607.09175] Scoped Verification for Reliable Long-Horizon Agentic Context Evolution under Distribution Shift
Deployed LLM agents rely on agentic context, the model-external textual control content assembled by an operational harness. In this work, the mutable component of that context is a persistent system-level instruction that is updated from operational experience while the model, tools, and harness remain fixed. Over lon...
Dan C. Hsu; Luke Lu
2026-07-10
2026
arXiv
arXiv
18 pages, 3 figs
cs.AI
arxiv-api
2607.09175
2026-09-08T16:27:48
ale-0625
State, Memory, And Context Persistence
state-memory-and-context-persistence
Tool
🧰
AgentMemory
https://github.com/rohitg00/agentmemory
external
github.com
Persistent cross-session memory for coding agents: auto-capture lifecycle hooks, an MCP server, and a REST API so progress, decisions, and context survive across runs and tools.
Persistent cross-session memory for coding agents: auto-capture lifecycle hooks, an MCP server, and a REST API so progress, decisions, and context survive across runs and tools.
Persistent cross-session memory for coding agents: auto-capture lifecycle hooks, an MCP server, and a REST API so progress, decisions, and context survive across runs and tools.
Persistent memory is treated as an external runtime artifact. Persistent cross-session memory for coding agents: auto-capture lifecycle hooks, an MCP server, and a REST API so progress, decisions, and context survive across runs and tools.
Use AgentMemory to carry context, state, and receipts across runs and failures.
Inspectable GitHub source; popularity is context, not proof of reliability.
medium
README.md
1,312
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1312
Persist
persist
Carry context, state, and receipts across runs.
workspace;context;state
builder
harness
enabling
source-implementation
A
ok
https://github.com/rohitg00/agentmemory
GitHub - rohitg00/agentmemory: #1 Persistent memory for AI coding agents based on real-world benchmarks · GitHub
#1 Persistent memory for AI coding agents based on real-world benchmarks - rohitg00/agentmemory
rohitg00/agentmemory
GitHub
html-meta
rohitg00/agentmemory
2026-09-08T16:27:48
ale-0626
State, Memory, And Context Persistence
state-memory-and-context-persistence
Tool
🧰
TencentDB-Agent-Memory
https://github.com/TencentCloud/TencentDB-Agent-Memory
external
github.com
Tencent Cloud's open-source local-first long-term memory for AI agents: a four-tier progressive pipeline from capture through consolidation with zero external API dependencies.
Tencent Cloud's open-source local-first long-term memory for AI agents: a four-tier progressive pipeline from capture through consolidation with zero external API dependencies.
Tencent Cloud's open-source local-first long-term memory for AI agents: a four-tier progressive pipeline from capture through consolidation with zero external API dependencies.
Persistent memory is treated as an external runtime artifact. Tencent Cloud's open-source local-first long-term memory for AI agents: a four-tier progressive pipeline from capture through consolidation with zero external API dependencies.
Use TencentDB-Agent-Memory to carry context, state, and receipts across runs and failures.
Inspectable GitHub source; popularity is context, not proof of reliability.
medium
README.md
1,313
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1313
Persist
persist
Carry context, state, and receipts across runs.
context
builder
harness
enabling
source-implementation
A
ok
https://github.com/TencentCloud/TencentDB-Agent-Memory
GitHub - TencentCloud/TencentDB-Agent-Memory: TencentDB Agent Memory is a team-level memory hub for AI Agents — turning conversations, docs, and code into four reusable memory assets (Chat Memory, Skill, LLM-Wiki, Code-Graph) that are governed, shared, and equipped across agents and frameworks. · GitHub
TencentDB Agent Memory is a team-level memory hub for AI Agents — turning conversations, docs, and code into four reusable memory assets (Chat Memory, Skill, LLM-Wiki, Code-Graph) that are governed, shared, and equipped across agents and frameworks. - TencentCloud/TencentDB-Agent-Memory
TencentCloud/TencentDB-Agent-Memory
GitHub
html-meta
TencentCloud/TencentDB-Agent-Memory
2026-09-08T16:27:48
ale-0627
State, Memory, And Context Persistence
state-memory-and-context-persistence
Tool
🧰
agent-memory (Neo4j Labs)
https://github.com/neo4j-labs/agent-memory
external
github.com
Official Neo4j Labs graph-native memory system that stores conversations, builds knowledge graphs from agent interactions, and lets agents learn from their own reasoning traces.
Official Neo4j Labs graph-native memory system that stores conversations, builds knowledge graphs from agent interactions, and lets agents learn from their own reasoning traces.
Official Neo4j Labs graph-native memory system that stores conversations, builds knowledge graphs from agent interactions, and lets agents learn from their own reasoning traces.
Primary-source operational guidance rather than commentary. Official Neo4j Labs graph-native memory system that stores conversations, builds knowledge graphs from agent interactions, and lets agents learn from their own reasoning traces.
Use agent-memory (Neo4j Labs) to carry context, state, and receipts across runs and failures.
Inspectable GitHub source; popularity is context, not proof of reliability.
medium
README.md
1,314
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1314
Persist
persist
Carry context, state, and receipts across runs.
context
builder
harness
enabling
source-implementation
A
ok
https://github.com/neo4j-labs/agent-memory
GitHub - neo4j-labs/agent-memory: A graph-native memory system for AI agents and context graphs. Store conversations, build knowledge graphs, and let your agents learn from their own reasoning — all backed by Neo4j. · GitHub
A graph-native memory system for AI agents and context graphs. Store conversations, build knowledge graphs, and let your agents learn from their own reasoning — all backed by Neo4j. - neo4j-labs/agent-memory
neo4j-labs/agent-memory
GitHub
html-meta
neo4j-labs/agent-memory
2026-09-08T16:27:48
ale-0628
State, Memory, And Context Persistence
state-memory-and-context-persistence
Tool
🧰
re_gent
https://github.com/regent-vcs/re_gent
external
github.com
Agent-native version control layered on top of Git that records agent activity at the prompt level, so you can answer why the agent did something and undo agent work without losing your own.
Agent-native version control layered on top of Git that records agent activity at the prompt level, so you can answer why the agent did something and undo agent work without losing your own.
Agent-native version control layered on top of Git that records agent activity at the prompt level, so you can answer why the agent did something and undo agent work without losing your own.
Makes persistence and context management visible as runtime design choices. Agent-native version control layered on top of Git that records agent activity at the prompt level, so you can answer why the agent did something and undo agent work without losing your own.
Use re_gent to carry context, state, and receipts across runs and failures.
Inspectable GitHub source; popularity is context, not proof of reliability.
medium
README.md
1,315
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1315
Persist
persist
Carry context, state, and receipts across runs.
context;state
builder
harness
enabling
source-implementation
A
ok
https://github.com/regent-vcs/re_gent
GitHub - regent-vcs/re_gent: Version control for AI agents — track what your agent did, blame any line to a prompt, inspect any step. · GitHub
Version control for AI agents — track what your agent did, blame any line to a prompt, inspect any step. - regent-vcs/re_gent
regent-vcs/re_gent
GitHub
html-meta
regent-vcs/re_gent
2026-09-08T16:27:48
ale-0629
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
StructAgent: Harness Long-Horizon Digital Agents with Unified Causal Structure
https://arxiv.org/abs/2607.11388
external
arxiv.org
State-centered framework that structures a long-horizon computer-use agent's state around a unified causal representation of task progress, regulating every update through verifier-backed state transitions with checkpointing and targeted failure recovery, lifting Qwen3.5-27B from 31.6% to 62.2% on OSWorld-Verified.
State-centered framework that structures a long-horizon computer-use agent's state around a unified causal representation of task progress, regulating every update through verifier-backed state transitions with checkpointing and targeted failure recovery, lifting Qwen3.5-27B from 31.6% to 62.2% on OSWorld-Verified.
State-centered framework that structures a long-horizon computer-use agent's state around a unified causal representation of task progress, regulating every update through verifier-backed state transitions with checkpointing and targeted failure recovery, lifting Qwen3.5-27B from 31.6% to 62.2% on OSWorld-Verified.
Checkpointed state makes long-running agent work recoverable across failures. State-centered framework that structures a long-horizon computer-use agent's state around a unified causal representation of task progress, regulating every update through verifier-backed state transitions with checkpointing and targeted fail...
Use StructAgent: Harness Long-Horizon Digital Agents with Unified Causal Structure to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.11388; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,316
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1316
2026-07-15
Persist
persist
Carry context, state, and receipts across runs.
verification;state
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.11388
[2607.11388] StructAgent: Harness Long-horizon Digital Agents with Unified Causal Structure
Recent advances in large language models (LLMs) and vision-language models (VLMs) have enabled increasingly capable digital agents for computer use. However, real-world tasks are often long-horizon and involve evolving contexts containing accumulated observations, intermediate edits, failed attempts, and partially comp...
Wenyi Wu; Sibo Zhu; Kun Zhou; Aayush Salvi; Zixuan Song; Biwei Huang
2026-07-13
2026
arXiv
arXiv
cs.AI
arxiv-api
2607.11388
2026-09-08T16:27:48
ale-0630
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
The Compliance Trap: Diagnosing How AI Agents Consume Conflicting Memory
https://arxiv.org/abs/2607.10608
external
arxiv.org
Diagnoses how agents resolve contradictions in their own persistent memory, finding they tend to comply with the most recent or most assertive entry rather than the correct one, a failure mode for any loop that accumulates state across runs.
Diagnoses how agents resolve contradictions in their own persistent memory, finding they tend to comply with the most recent or most assertive entry rather than the correct one, a failure mode for any loop that accumulates state across runs.
Diagnoses how agents resolve contradictions in their own persistent memory, finding they tend to comply with the most recent or most assertive entry rather than the correct one, a failure mode for any loop that accumulates state across runs.
Persistent memory is treated as an external runtime artifact. Diagnoses how agents resolve contradictions in their own persistent memory, finding they tend to comply with the most recent or most assertive entry rather than the correct one, a failure mode for any loop that accumulates state across runs.
Use The Compliance Trap: Diagnosing How AI Agents Consume Conflicting Memory to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.10608; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,317
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1317
2026-07-15
Persist
persist
Carry context, state, and receipts across runs.
context;state
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.10608
[2607.10608] The Compliance Trap: Diagnosing How AI Agents Consume Conflicting Memory
Memory is becoming a core component of long-horizon AI agents, allowing agents to reuse past experience when operating web browsers, software tools, and other interactive environments. Existing work mostly treats memory as a supply problem, asking what experience to write, how to store it, and which entry to retrieve f...
Yixiong Chen; Xinyi Bai; Alan Yuille
2026-07-12
2026
arXiv
arXiv
cs.AI
arxiv-api
2607.10608
2026-09-08T16:27:48
ale-0631
State, Memory, And Context Persistence
state-memory-and-context-persistence
Docs
📚
Conversational Context: Session, State, and Memory
https://adk.dev/sessions/
external
adk.dev
Official ADK model separating a conversation thread, its mutable state, and searchable cross-session memory, with service backends for durable persistence.
Official ADK model separating a conversation thread, its mutable state, and searchable cross-session memory, with service backends for durable persistence.
Official ADK model separating a conversation thread, its mutable state, and searchable cross-session memory, with service backends for durable persistence.
Primary-source operational guidance rather than commentary. Official ADK model separating a conversation thread, its mutable state, and searchable cross-session memory, with service backends for durable persistence.
Use Conversational Context: Session, State, and Memory to carry context, state, and receipts across runs and failures.
Primary official documentation from adk.dev; use it for current product or standard behavior.
high
README.md
1,318
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1318
2026-07-17
Persist
persist
Carry context, state, and receipts across runs.
context;state
builder
harness
enabling
official-documentation
A
ok
https://adk.dev/sessions/
Conversational context - Agent Development Kit (ADK) Agent Development Kit (ADK)
Build powerful multi-agent systems with Agent Development Kit (ADK)
Google Agent Development Kit
Google Agent Development Kit
Google
primary-page
2026-09-08T16:27:48
ale-0632
State, Memory, And Context Persistence
state-memory-and-context-persistence
Docs
📚
Persistence
https://docs.langchain.com/oss/python/langgraph/persistence
external
docs.langchain.com
Official LangGraph checkpoint model: save state at every super-step, retain pending writes, recover interrupted execution, support human review, and enable memory and time travel.
Official LangGraph checkpoint model: save state at every super-step, retain pending writes, recover interrupted execution, support human review, and enable memory and time travel.
Official LangGraph checkpoint model: save state at every super-step, retain pending writes, recover interrupted execution, support human review, and enable memory and time travel.
Primary-source operational guidance rather than commentary. Official LangGraph checkpoint model: save state at every super-step, retain pending writes, recover interrupted execution, support human review, and enable memory and time travel.
Use Persistence to carry context, state, and receipts across runs and failures.
Primary official documentation from docs.langchain.com; use it for current product or standard behavior.
high
README.md
1,319
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1319
2026-07-17
Persist
persist
Carry context, state, and receipts across runs.
context;state;escalation
builder
harness
enabling
official-documentation
A
ok
https://docs.langchain.com/oss/python/langgraph/persistence
Persistence - Docs by LangChain
LangGraph's persistence layer gives agents short-term memory through checkpointers and long-term memory through stores.
LangChain
LangGraph
LangChain
primary-page
2026-09-08T16:27:48
ale-0633
State, Memory, And Context Persistence
state-memory-and-context-persistence
Docs
📚
Workflow checkpoints
https://learn.microsoft.com/en-us/agent-framework/workflows/checkpoints
external
learn.microsoft.com
Official checkpointing guide covering super-step state, pending messages and requests, shared state, in-memory and durable storage providers, and resuming long-running workflows.
Official checkpointing guide covering super-step state, pending messages and requests, shared state, in-memory and durable storage providers, and resuming long-running workflows.
Official checkpointing guide covering super-step state, pending messages and requests, shared state, in-memory and durable storage providers, and resuming long-running workflows.
Primary-source operational guidance rather than commentary. Official checkpointing guide covering super-step state, pending messages and requests, shared state, in-memory and durable storage providers, and resuming long-running workflows.
Use Workflow checkpoints to carry context, state, and receipts across runs and failures.
Primary official documentation from learn.microsoft.com; use it for current product or standard behavior.
high
README.md
1,320
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1320
2026-07-17
Persist
persist
Carry context, state, and receipts across runs.
context;state
builder
harness
enabling
official-documentation
A
ok
https://learn.microsoft.com/en-us/agent-framework/workflows/checkpoints
Microsoft Agent Framework Workflows - Checkpoints | Microsoft Learn
In-depth look at Checkpoints in Microsoft Agent Framework Workflows.
Microsoft
Microsoft Agent Framework
Microsoft
primary-page
2026-09-08T16:27:48
ale-0634
State, Memory, And Context Persistence
state-memory-and-context-persistence
Docs
📚
Agent state
https://strandsagents.com/docs/user-guide/concepts/agents/state/
external
strandsagents.com
Official state guide separating conversation, agent, and invocation lifetimes, with validation hooks and cross-session persistence for durable agent behavior.
Official state guide separating conversation, agent, and invocation lifetimes, with validation hooks and cross-session persistence for durable agent behavior.
Official state guide separating conversation, agent, and invocation lifetimes, with validation hooks and cross-session persistence for durable agent behavior.
Primary-source operational guidance rather than commentary. Official state guide separating conversation, agent, and invocation lifetimes, with validation hooks and cross-session persistence for durable agent behavior.
Use Agent state to carry context, state, and receipts across runs and failures.
Primary official documentation from strandsagents.com; use it for current product or standard behavior.
high
README.md
1,321
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1321
2026-07-17
Persist
persist
Carry context, state, and receipts across runs.
state
builder
harness
enabling
official-documentation
A
ok
https://strandsagents.com/docs/user-guide/concepts/agents/state/
State Management | Strands Agents
Maintain state in Strands agents across turns and sessions: conversation history, key-value agent state, and per-request invocation state.
Strands Agents
Strands Agents
Strands Agents
primary-page
2026-09-08T16:27:48
ale-0635
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Memory as a Controlled Process: Learned Adaptive Memory Management for LLM Agents
https://arxiv.org/abs/2607.13591
external
arxiv.org
Learns when to retrieve, consolidate, and forget rather than treating memory as passive storage; across six benchmarks, three frameworks, and three LLMs, it reports up to 15.2 points higher task success with 5-20% fewer tokens.
Learns when to retrieve, consolidate, and forget rather than treating memory as passive storage; across six benchmarks, three frameworks, and three LLMs, it reports up to 15.2 points higher task success with 5-20% fewer tokens.
Learns when to retrieve, consolidate, and forget rather than treating memory as passive storage; across six benchmarks, three frameworks, and three LLMs, it reports up to 15.2 points higher task success with 5-20% fewer tokens.
The work turns loop quality into a measurable task or score. Learns when to retrieve, consolidate, and forget rather than treating memory as passive storage; across six benchmarks, three frameworks, and three LLMs, it reports up to 15.2 points higher task success with 5-20% fewer tokens.
Use Memory as a Controlled Process: Learned Adaptive Memory Management for LLM Agents to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.13591; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,322
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1322
2026-07-17
Persist
persist
Carry context, state, and receipts across runs.
context;verification;budget
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.13591
[2607.13591] Memory as a Controlled Process: Learned Adaptive Memory Management for LLM Agents
Large Language Model (LLM) agents increasingly rely on external memory systems to accumulate experience across tasks. Yet nearly all existing approaches, from graph-structured memories to reflective insight stores, access memory through fixed, hand-designed heuristics. We argue that this static view of memory is a core...
Eric Hanchen Jiang; Zhi Zhang; Yuchen Wu; Levina Li; Dong Liu; Xiao Liang; Rui Sun; Yubei Li; Edward Sun; Haozheng Luo; Zhaolu Kang; Aylin Caliskan; Kai-Wei Chang; Ying Nian Wu
2026-07-15
2026
arXiv
arXiv
cs.CL
arxiv-api
2607.13591
2026-09-08T16:27:48
ale-0636
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Why Git Is the Memory Solution for the Agentic Development Lifecycle
https://arxiv.org/abs/2607.14390
external
arxiv.org
Binds agent memory to versioned Git artifacts and evaluates retrieval across eight corpora; reported best retrieval reaches about 0.31 MRR and decision synthesis 0.83 sufficiency at 382-980 tokens per query, with capture quality still the main constraint.
Binds agent memory to versioned Git artifacts and evaluates retrieval across eight corpora; reported best retrieval reaches about 0.31 MRR and decision synthesis 0.83 sufficiency at 382-980 tokens per query, with capture quality still the main constraint.
Binds agent memory to versioned Git artifacts and evaluates retrieval across eight corpora; reported best retrieval reaches about 0.31 MRR and decision synthesis 0.83 sufficiency at 382-980 tokens per query, with capture quality still the main constraint.
Persistent memory is treated as an external runtime artifact. Binds agent memory to versioned Git artifacts and evaluates retrieval across eight corpora; reported best retrieval reaches about 0.31 MRR and decision synthesis 0.83 sufficiency at 382-980 tokens per query, with capture quality still the main constraint.
Use Why Git Is the Memory Solution for the Agentic Development Lifecycle to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.14390; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,323
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1323
2026-07-17
Persist
persist
Carry context, state, and receipts across runs.
context;budget
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.14390
[2607.14390] Why Git Is the Memory Solution for the Agentic Development Lifecycle
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 u...
Frank Guo
2026-07-15
2026
arXiv
arXiv
8 pages
cs.SE
arxiv-api
2607.14390
2026-09-08T16:27:48
ale-0637
State, Memory, And Context Persistence
state-memory-and-context-persistence
Benchmark
🧪
ContinuityBench: A Benchmark and Systems Study of Stateful Failover in Multi-Provider LLM Routing
https://arxiv.org/abs/2607.15899
external
arxiv.org
Defines continuity-preservation and latency-overhead metrics for provider failover and evaluates a state-forwarding proxy over 750 failover events; reported context preservation reaches 99.20% versus near zero for stateless routing, with asynchronous backoff and jitter preventing retry storms.
Defines continuity-preservation and latency-overhead metrics for provider failover and evaluates a state-forwarding proxy over 750 failover events; reported context preservation reaches 99.20% versus near zero for stateless routing, with asynchronous backoff and jitter preventing retry storms.
Defines continuity-preservation and latency-overhead metrics for provider failover and evaluates a state-forwarding proxy over 750 failover events; reported context preservation reaches 99.20% versus near zero for stateless routing, with asynchronous backoff and jitter preventing retry storms.
The work turns loop quality into a measurable task or score. Defines continuity-preservation and latency-overhead metrics for provider failover and evaluates a state-forwarding proxy over 750 failover events; reported context preservation reaches 99.20% versus near zero for stateless routing, with asynchronous backoff ...
Use ContinuityBench: A Benchmark and Systems Study of Stateful Failover in Multi-Provider LLM Routing to carry context, state, and receipts across runs and failures.
Benchmark or leaderboard source with repeatable tasks or scores; compare systems only after checking setup and scope.
high
README.md
1,324
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1324
2026-07-20
Persist
persist
Carry context, state, and receipts across runs.
context;verification;state;budget
researcher;evaluator
harness
enabling
benchmark
A
ok
https://arxiv.org/abs/2607.15899
[2607.15899] ContinuityBench: A Benchmark and Systems Study of Stateful Failover in Multi-Provider LLM Routing
In production large language model (LLM) deployments, high API availability guarantees do not equate to conversational continuity. When a primary provider experiences an outage or strict rate-limiting, naive stateless failover mechanisms successfully maintain uptime but silently discard conversation history, severely d...
Vishal Pandey; Gopal Singh
2026-07-17
2026
arXiv
arXiv
16 pages, 2 figures
cs.LG
arxiv-api
2607.15899
2026-09-08T16:27:48
ale-0638
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Presentation, Not Mechanism: A Render Confound in Deprecation-Aware Memory Evaluation
https://arxiv.org/abs/2607.16019
external
arxiv.org
Uses a render-matched control over 2,907 evidence-revision questions to show that an apparent +0.182 structured-memory gain is mostly presentation, leaving a +0.021 to +0.025 mechanism residual indistinguishable from zero; memory evaluations should hold rendering fixed and prefer the coarsest retained state that answer...
Uses a render-matched control over 2,907 evidence-revision questions to show that an apparent +0.182 structured-memory gain is mostly presentation, leaving a +0.021 to +0.025 mechanism residual indistinguishable from zero; memory evaluations should hold rendering fixed and prefer the coarsest retained state that answer...
Uses a render-matched control over 2,907 evidence-revision questions to show that an apparent +0.182 structured-memory gain is mostly presentation, leaving a +0.021 to +0.025 mechanism residual indistinguishable from zero; memory evaluations should hold rendering fixed and prefer the coarsest retained state that answer...
Evaluation data is used as the feedback signal for improving loop behavior. Uses a render-matched control over 2,907 evidence-revision questions to show that an apparent +0.182 structured-memory gain is mostly presentation, leaving a +0.021 to +0.025 mechanism residual indistinguishable from zero; memory evaluations sh...
Use Presentation, Not Mechanism: A Render Confound in Deprecation-Aware Memory Evaluation to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.16019; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,325
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1325
2026-07-20
Persist
persist
Carry context, state, and receipts across runs.
context;verification;state
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.16019
[2607.16019] Presentation, Not Mechanism: A Render Confound in Deprecation-Aware Memory Evaluation
AI systems increasingly retrieve from records that revise themselves: issue threads, encyclopedic histories, policy logs, and long conversations. The challenge is not only finding relevant evidence, but deciding which claims remain in force, which were superseded, and when to abstain. Structured memories promise to sol...
Zhaoyang Jiang; Zhizhong Fu; Zicheng Li; Yunsoo Kim; Jiacong Mi; Xuanqi Peng; Fei Teng; Honghan Wu
2026-07-17
2026
arXiv
arXiv
cs.LG
arxiv-api
2607.16019
2026-09-08T16:27:48
ale-0639
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
ReasoningBank: Scaling Agent Self-Evolving with Reasoning Memory
https://arxiv.org/abs/2509.25140
external
arxiv.org
Distills successful and failed trajectories into reusable reasoning memories, then uses memory-aware test-time scaling to turn additional exploration into better guidance for future runs.
Distills successful and failed trajectories into reusable reasoning memories, then uses memory-aware test-time scaling to turn additional exploration into better guidance for future runs.
Distills successful and failed trajectories into reusable reasoning memories, then uses memory-aware test-time scaling to turn additional exploration into better guidance for future runs.
Persistent memory is treated as an external runtime artifact. Distills successful and failed trajectories into reusable reasoning memories, then uses memory-aware test-time scaling to turn additional exploration into better guidance for future runs.
Use ReasoningBank: Scaling Agent Self-Evolving with Reasoning Memory to carry context, state, and receipts across runs and failures.
Research source arXiv:2509.25140; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,329
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1329
2026-07-18
Persist
persist
Carry context, state, and receipts across runs.
context;verification
researcher;evaluator
harness
enabling
research-paper
A
ok
https://openreview.net/forum?id=jL7fwchScm
[2509.25140] ReasoningBank: Scaling Agent Self-Evolving with Reasoning Memory
With the growing adoption of large language model agents in persistent real-world roles, they naturally encounter continuous streams of tasks. A key limitation, however, is their failure to learn from the accumulated interaction history, forcing them to discard valuable insights and repeat past errors. We propose Reaso...
Siru Ouyang; Jun Yan; I-Hung Hsu; Yanfei Chen; Ke Jiang; Zifeng Wang; Rujun Han; Long T. Le; Samira Daruki; Xiangru Tang; Vishy Tirumalashetty; George Lee; Mahsan Rofouei; Hangfei Lin; Jiawei Han; Chen-Yu Lee; Tomas Pfister
2026
2026
International Conference on Learning Representations (ICLR)
International Conference on Learning Representations
Published in International Conference on Learning Representations (ICLR); the linked arXiv record remains available for open access.
cs.AI
OpenReview proceedings record
2509.25140
2026-09-08T16:27:48
ale-0640
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Scaling Long-Horizon LLM Agent via Context-Folding
https://arxiv.org/abs/2510.11967
external
arxiv.org
Lets an agent branch into temporary sub-trajectories and fold completed work into compact continuation state, pairing the mechanism with FoldGRPO to operate under a 32K active-context budget.
Lets an agent branch into temporary sub-trajectories and fold completed work into compact continuation state, pairing the mechanism with FoldGRPO to operate under a 32K active-context budget.
Lets an agent branch into temporary sub-trajectories and fold completed work into compact continuation state, pairing the mechanism with FoldGRPO to operate under a 32K active-context budget.
Context is managed as durable loop state rather than a single prompt payload. Lets an agent branch into temporary sub-trajectories and fold completed work into compact continuation state, pairing the mechanism with FoldGRPO to operate under a 32K active-context budget.
Use Scaling Long-Horizon LLM Agent via Context-Folding to carry context, state, and receipts across runs and failures.
Research source arXiv:2510.11967; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,330
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1330
2026-07-18
Persist
persist
Carry context, state, and receipts across runs.
context;state;budget
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2510.11967
[2510.11967] Scaling Long-Horizon LLM Agent via Context-Folding
Large language model (LLM) agents are fundamentally constrained by context length on long-horizon tasks. We introduce Context-Folding, a framework that empowers agents to actively manage their working context. An agent can procedurally branch into a sub-trajectory to handle a subtask and then fold it upon completion, c...
Weiwei Sun; Miao Lu; Zhan Ling; Kang Liu; Xuesong Yao; Yiming Yang; Jiecao Chen
2025-10-13
2025
arXiv
arXiv
cs.CL
arxiv-api
2510.11967
2026-09-08T16:27:48
ale-0641
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Experience Memory Graph: One-Shot Error Correction for Agents
https://arxiv.org/abs/2607.13884
external
arxiv.org
Converts failed and successful past trajectories into directed action-decision graphs and matches them at test time to retrieve explicit correction strategies, letting agents fix known error patterns in a single loop-free execution instead of iterative reflection, cross-run experience memory applied to error recovery, ...
Converts failed and successful past trajectories into directed action-decision graphs and matches them at test time to retrieve explicit correction strategies, letting agents fix known error patterns in a single loop-free execution instead of iterative reflection, cross-run experience memory applied to error recovery, ...
Converts failed and successful past trajectories into directed action-decision graphs and matches them at test time to retrieve explicit correction strategies, letting agents fix known error patterns in a single loop-free execution instead of iterative reflection, cross-run experience memory applied to error recovery, ...
Control flow is represented as an inspectable graph rather than an opaque prompt loop. Converts failed and successful past trajectories into directed action-decision graphs and matches them at test time to retrieve explicit correction strategies, letting agents fix known error patterns in a single loop-free execution i...
Use Experience Memory Graph: One-Shot Error Correction for Agents to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.13884; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,331
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1331
2026-07-22
Persist
persist
Carry context, state, and receipts across runs.
context;verification
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.13884
[2607.13884] Experience Memory Graph: One-Shot Error Correction for Agents
Large Language Model (LLM) agents have shown remarkable capabilities in autonomous decision-making by generating sequential trajectories of states, actions, and observations. However, in complex, long-horizon tasks, these agents frequently suffer from compounding errors and struggle to recover from failures. Existing s...
Wenjun Wang; Yuchen Fang; Fengrui Liu; Zibo Liang; Kai Zheng
2026-07-15
2026
arXiv
arXiv
11 pages, 6 figures
cs.AI
arxiv-api
2607.13884
2026-09-08T16:27:48
ale-0642
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Always-On Agents: A Survey of Persistent Memory, State, and Governance in LLM Agents
https://arxiv.org/abs/2606.30306
external
arxiv.org
Survey of 435 works on agents whose behavior is shaped by durable state accumulated across runs, memories, task records, permissions, audit trails, analyzed along six diagnostic dimensions (authority, scope, mutability, provenance, recoverability, actionability); finds the literature over-indexes on accumulating and re...
Survey of 435 works on agents whose behavior is shaped by durable state accumulated across runs, memories, task records, permissions, audit trails, analyzed along six diagnostic dimensions (authority, scope, mutability, provenance, recoverability, actionability); finds the literature over-indexes on accumulating and re...
Survey of 435 works on agents whose behavior is shaped by durable state accumulated across runs, memories, task records, permissions, audit trails, analyzed along six diagnostic dimensions (authority, scope, mutability, provenance, recoverability, actionability); finds the literature over-indexes on accumulating and re...
Durable execution and replay are treated as first-class loop infrastructure. Survey of 435 works on agents whose behavior is shaped by durable state accumulated across runs, memories, task records, permissions, audit trails, analyzed along six diagnostic dimensions (authority, scope, mutability, provenance, recoverabil...
Use Always-On Agents: A Survey of Persistent Memory, State, and Governance in LLM Agents to carry context, state, and receipts across runs and failures.
Research source arXiv:2606.30306; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,332
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1332
2026-07-22
Persist
persist
Carry context, state, and receipts across runs.
workspace;context;verification;state
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2606.30306
[2606.30306] Always-OnAgents:A Survey of Persistent Memory, State, and Governance in LLMAgents
Always-on agents are systems whose future behavior depends on durable state accumulated across earlier interactions. We treat them as persistent-state systems: the operative system includes retrievable memories, but also task ledgers, permissions, credentials, commitments, provenance and audit records, shared state, tr...
Tianyu Ding; Aditya Nannapaneni; Bingfan Liu; Ling Zhang
2026-06-29
2026
arXiv
arXiv
cs.MA
arxiv-api
2606.30306
2026-09-08T16:27:48
ale-0643
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
When Continual Learning Moves to Memory: A Study of Experience Reuse in LLM Agents
https://arxiv.org/abs/2604.27003
external
arxiv.org
Shows the stability-plasticity dilemma of continual learning resurfaces at the memory level in memory-augmented agents, old and new experiences compete during retrieval under bounded context, and proposes a framework for representing and organizing cross-run experience to maximize transfer while minimizing forgetting, ...
Shows the stability-plasticity dilemma of continual learning resurfaces at the memory level in memory-augmented agents, old and new experiences compete during retrieval under bounded context, and proposes a framework for representing and organizing cross-run experience to maximize transfer while minimizing forgetting, ...
Shows the stability-plasticity dilemma of continual learning resurfaces at the memory level in memory-augmented agents, old and new experiences compete during retrieval under bounded context, and proposes a framework for representing and organizing cross-run experience to maximize transfer while minimizing forgetting, ...
Persistent memory is treated as an external runtime artifact. Shows the stability-plasticity dilemma of continual learning resurfaces at the memory level in memory-augmented agents, old and new experiences compete during retrieval under bounded context, and proposes a framework for representing and organizing cross-run...
Use When Continual Learning Moves to Memory: A Study of Experience Reuse in LLM Agents to carry context, state, and receipts across runs and failures.
Research source arXiv:2604.27003; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,333
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1333
2026-07-22
Persist
persist
Carry context, state, and receipts across runs.
context
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2604.27003
[2604.27003] When Continual Learning Moves to Memory: A Study of Experience Reuse in LLM Agents
Memory-augmented LLM agents offer an appealing shortcut to continual learning: rather than updating model parameters, they accumulate experience in external memory, seemingly sidestepping the stability-plasticity dilemma of parametric learning. We show that this challenge does not disappear but resurfaces at the memory...
Qisheng Hu; Quanyu Long; Wenya Wang
2026-04-29
2026
arXiv
arXiv
Working in progress
cs.LG
arxiv-api
2604.27003
2026-09-08T16:27:48
ale-0644
State, Memory, And Context Persistence
state-memory-and-context-persistence
Tool
🧰
deja-vu
https://github.com/vshulcz/deja-vu
external
github.com
Memory layer over coding-agent session logs that mines past sessions for reusable context and recalls it into future runs.
Memory layer over coding-agent session logs that mines past sessions for reusable context and recalls it into future runs.
Memory layer over coding-agent session logs that mines past sessions for reusable context and recalls it into future runs.
Persistent memory is treated as an external runtime artifact. Memory layer over coding-agent session logs that mines past sessions for reusable context and recalls it into future runs.
Use deja-vu to carry context, state, and receipts across runs and failures.
Inspectable GitHub source; popularity is context, not proof of reliability.
medium
README.md
1,334
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1334
2026-07-22
Persist
persist
Carry context, state, and receipts across runs.
context
builder
harness
enabling
source-implementation
A
ok
https://github.com/vshulcz/deja-vu
GitHub - vshulcz/deja-vu: One memory shared by Claude Code, Codex, Cursor, Copilot CLI, OpenClaw and 20 more coding agents, built from the session history already on disk. A fix found in one agent comes back in any of them, including months of sessions from before you installed it. No LLM, no embeddings, one local Go b...
One memory shared by Claude Code, Codex, Cursor, Copilot CLI, OpenClaw and 20 more coding agents, built from the session history already on disk. A fix found in one agent comes back in any of them, including months of sessions from before you installed it. No LLM, no embeddings, one local Go binary. - vshulcz/deja-vu
vshulcz/deja-vu
GitHub
html-meta
vshulcz/deja-vu
2026-09-08T16:27:48
ale-0645
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Oracle Agent Memory as an Enterprise Memory Substrate for Long-Horizon AI Agents
https://arxiv.org/abs/2607.13157
external
arxiv.org
Oracle's database-native memory layer for long-running agents: task-state retention across extended conversations, procedural knowledge accumulation from prior outcomes, and layered active/passive memory with per-user and per-agent scope control, the second major database vendor (after Tencent) to ship agent memory as ...
Oracle's database-native memory layer for long-running agents: task-state retention across extended conversations, procedural knowledge accumulation from prior outcomes, and layered active/passive memory with per-user and per-agent scope control, the second major database vendor (after Tencent) to ship agent memory as ...
Oracle's database-native memory layer for long-running agents: task-state retention across extended conversations, procedural knowledge accumulation from prior outcomes, and layered active/passive memory with per-user and per-agent scope control, the second major database vendor (after Tencent) to ship agent memory as ...
Persistent memory is treated as an external runtime artifact. Oracle's database-native memory layer for long-running agents: task-state retention across extended conversations, procedural knowledge accumulation from prior outcomes, and layered active/passive memory with per-user and per-agent scope control, the second ...
Use Oracle Agent Memory as an Enterprise Memory Substrate for Long-Horizon AI Agents to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.13157; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,335
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1335
2026-07-22
Persist
persist
Carry context, state, and receipts across runs.
context;state
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.13157
[2607.13157] Oracle Agent Memory as an Enterprise Memory Substrate for Long-Horizon AI Agents
Agent memory is a systems problem for long-horizon agents. Practical deployments require retention of task state across extended conversations, recovery of user-specific facts and preferences across sessions, and accumulation of procedural knowledge from prior outcomes. These requirements extend beyond document retriev...
Richmond Alake; Cesare Bernardis; Paul Cayet; Luca Engel; Damien Hilloulin; Sungpack Hong; Allen Hosler; Nickolas Kavantzas; Ingo Kossyk; Son Le; Rhicheek Patra; Kartik Talamadupula; Valentin Venzin
2026-07-14
2026
arXiv
arXiv
23 pages, 7 figures. Technical report on Oracle Agent Memory
cs.AI
arxiv-api
2607.13157
2026-09-08T16:27:48
ale-0646
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
KnowAct-GUIClaw: Personal GUI Assistant with Self-Evolving Memory
https://arxiv.org/abs/2607.12625
external
arxiv.org
Cross-platform GUI agent (Android, iOS, HarmonyOS, Windows) with a Know-Route-Act-Reflect loop that accumulates user experience into memory and grows a self-evolving skill library across sessions, a concrete instance of recurring-stateful-loop design applied to personal device automation.
Cross-platform GUI agent (Android, iOS, HarmonyOS, Windows) with a Know-Route-Act-Reflect loop that accumulates user experience into memory and grows a self-evolving skill library across sessions, a concrete instance of recurring-stateful-loop design applied to personal device automation.
Cross-platform GUI agent (Android, iOS, HarmonyOS, Windows) with a Know-Route-Act-Reflect loop that accumulates user experience into memory and grows a self-evolving skill library across sessions, a concrete instance of recurring-stateful-loop design applied to personal device automation.
Persistent memory is treated as an external runtime artifact. Cross-platform GUI agent (Android, iOS, HarmonyOS, Windows) with a Know-Route-Act-Reflect loop that accumulates user experience into memory and grows a self-evolving skill library across sessions, a concrete instance of recurring-stateful-loop design applied...
Use KnowAct-GUIClaw: Personal GUI Assistant with Self-Evolving Memory to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.12625; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,336
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1336
2026-07-22
Persist
persist
Carry context, state, and receipts across runs.
context;state
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.12625
[2607.12625] KnowAct-GUIClaw: Know Deeply, Act Perfectly, Personal GUI Assistant with Self-Evolving Memory and Skill
OpenClaw has emerged as a leading agent framework for complex task automation, yet it faces insufficient cross-platform GUI interaction support and a well-built self-evolution mechanism. These flaws limit its adaptation to diverse device ecosystems and prevent performance improvements through continuous learning from e...
Yunxin Li; Jinchao Li; Shibo Su; Zhenran Xu; Chenrui Zhao; Tongshu Bian; Xiaoman Liang; Meishan Zhang; Baotian Hu; Min Zhang
2026-07-14
2026
arXiv
arXiv
29 pages, 9 figures
cs.CL
arxiv-api
2607.12625
2026-09-08T16:27:48
ale-0647
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
PRO-LONG: Programmatic Memory Enables Long-Horizon Reasoning
https://arxiv.org/abs/2607.20064
external
arxiv.org
Minimal context-management framework that keeps a complete, structured interaction log as programmatic memory and reuses coding-agent capabilities to search that history rather than stuffing the context window, reporting +18.0 points over a base coding agent on ARC-AGI-3 while using 4.2-5.8x fewer tokens than specializ...
Minimal context-management framework that keeps a complete, structured interaction log as programmatic memory and reuses coding-agent capabilities to search that history rather than stuffing the context window, reporting +18.0 points over a base coding agent on ARC-AGI-3 while using 4.2-5.8x fewer tokens than specializ...
Minimal context-management framework that keeps a complete, structured interaction log as programmatic memory and reuses coding-agent capabilities to search that history rather than stuffing the context window, reporting +18.0 points over a base coding agent on ARC-AGI-3 while using 4.2-5.8x fewer tokens than specializ...
Persistent memory is treated as an external runtime artifact. Minimal context-management framework that keeps a complete, structured interaction log as programmatic memory and reuses coding-agent capabilities to search that history rather than stuffing the context window, reporting +18.0 points over a base coding agent...
Use PRO-LONG: Programmatic Memory Enables Long-Horizon Reasoning to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.20064; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,337
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1337
2026-07-23
Persist
persist
Carry context, state, and receipts across runs.
context;budget
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.20064
[2607.20064] PRO-LONG: Programmatic Memory Enables Long-Horizon Reasoning
Long-horizon tasks require sustained perception, reasoning, and exploration, and are a persistent challenge for large language model (LLM) agents. This gap is reflected in their limited performance on continual learning benchmarks such as ARC-AGI-3, especially when models are evaluated out of the box. Various agent har...
Alexis Fox; Junlin Wang; Paul Rosu; Bhuwan Dhingra
2026-07-22
2026
arXiv
arXiv
cs.AI
arxiv-api
2607.20064
2026-09-08T16:27:48
ale-0648
State, Memory, And Context Persistence
state-memory-and-context-persistence
Tool
🧰
engram
https://github.com/Gentleman-Programming/engram
external
github.com
Agent-agnostic persistent memory system for coding agents, shipped as a single Go binary, so lessons, decisions, and context survive across sessions and tools instead of cold-starting each run.
Agent-agnostic persistent memory system for coding agents, shipped as a single Go binary, so lessons, decisions, and context survive across sessions and tools instead of cold-starting each run.
Agent-agnostic persistent memory system for coding agents, shipped as a single Go binary, so lessons, decisions, and context survive across sessions and tools instead of cold-starting each run.
Persistent memory is treated as an external runtime artifact. Agent-agnostic persistent memory system for coding agents, shipped as a single Go binary, so lessons, decisions, and context survive across sessions and tools instead of cold-starting each run.
Use engram to carry context, state, and receipts across runs and failures.
Inspectable GitHub source; popularity is context, not proof of reliability.
medium
README.md
1,338
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1338
2026-07-23
Persist
persist
Carry context, state, and receipts across runs.
workspace;context;state
builder
harness
enabling
source-implementation
A
ok
https://github.com/Gentleman-Programming/engram
GitHub - Gentleman-Programming/engram: Persistent memory system for AI coding agents. Agent-agnostic Go binary with SQLite + FTS5, MCP server, HTTP API, CLI, and TUI. · GitHub
Persistent memory system for AI coding agents. Agent-agnostic Go binary with SQLite + FTS5, MCP server, HTTP API, CLI, and TUI. - Gentleman-Programming/engram
Gentleman-Programming/engram
GitHub
html-meta
Gentleman-Programming/engram
2026-09-08T16:27:48
ale-0649
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Delivery, Not Storage: Cue-Anchored Working Memory as a Harness Property for Coding Agents
https://arxiv.org/abs/2607.20972
external
arxiv.org
Position paper arguing coding agents need a second memory tier beyond deliberately authored documents: situationally-bound operational facts (gotchas, locations, conventions) captured as a side effect of work and delivered automatically when situations cue them, implemented as a harness property, not an agent choice.
Position paper arguing coding agents need a second memory tier beyond deliberately authored documents: situationally-bound operational facts (gotchas, locations, conventions) captured as a side effect of work and delivered automatically when situations cue them, implemented as a harness property, not an agent choice.
Position paper arguing coding agents need a second memory tier beyond deliberately authored documents: situationally-bound operational facts (gotchas, locations, conventions) captured as a side effect of work and delivered automatically when situations cue them, implemented as a harness property, not an agent choice.
Persistent memory is treated as an external runtime artifact. Position paper arguing coding agents need a second memory tier beyond deliberately authored documents: situationally-bound operational facts (gotchas, locations, conventions) captured as a side effect of work and delivered automatically when situations cue t...
Use Delivery, Not Storage: Cue-Anchored Working Memory as a Harness Property for Coding Agents to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.20972; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,339
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1339
2026-07-24
Persist
persist
Carry context, state, and receipts across runs.
context
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.20972
[2607.20972] Delivery, Not Storage: Cue-Anchored Working Memory as a Harness Property for Coding Agents
Coding agents ship with one kind of memory: documents. Instruction files, plan artifacts, and auto-written memory directories are deliberately authored and deliberately retrieved: the agent must choose to write them and choose to read them back. Human expertise runs on a second tier that never gets written down: situat...
Swapnanil Saha
2026-07-23
2026
arXiv
arXiv
cs.AI
arxiv-api
2607.20972
2026-09-08T16:27:48
ale-0650
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
MemTools: A Unified Research Framework for Interoperable Agent Memory
https://arxiv.org/abs/2607.21404
external
arxiv.org
Interoperability framework that decouples agent memory-system components from deployment environments, standardizing the memory lifecycle through declarative contracts so symbolic, neural, and multimodal memory types, and their evaluation protocols, become interchangeable rather than entangled.
Interoperability framework that decouples agent memory-system components from deployment environments, standardizing the memory lifecycle through declarative contracts so symbolic, neural, and multimodal memory types, and their evaluation protocols, become interchangeable rather than entangled.
Interoperability framework that decouples agent memory-system components from deployment environments, standardizing the memory lifecycle through declarative contracts so symbolic, neural, and multimodal memory types, and their evaluation protocols, become interchangeable rather than entangled.
Evaluation data is used as the feedback signal for improving loop behavior. Interoperability framework that decouples agent memory-system components from deployment environments, standardizing the memory lifecycle through declarative contracts so symbolic, neural, and multimodal memory types, and their evaluation proto...
Use MemTools: A Unified Research Framework for Interoperable Agent Memory to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.21404; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,340
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1340
2026-07-24
Persist
persist
Carry context, state, and receipts across runs.
context;verification
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.21404
[2607.21404] MemTools: A Unified Research Framework for Interoperable Agent Memory
While memory systems are essential for agent architectures, pervasive architectural fragmentation restricts systematic research. Existing implementations typically couple different stages of the memory lifecycle, entangle evaluation logic with specific datasets, and provide limited support for the management of heterog...
Chengfeng Zhao; Jinhui Chen; Sirui Liang; Shizhu He; Yequan Wang; Jun Zhao; Kang Liu
2026-07-23
2026
arXiv
arXiv
Work in progress
cs.CL
arxiv-api
2607.21404
2026-09-08T16:27:48
ale-0651
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
AttriMem: Attribution-Guided Process Feedback for Agent Memory Learning
https://arxiv.org/abs/2607.21106
external
arxiv.org
Learns the memory-construction policy itself, what to extract, store, update, compress, or discard, by attributing token-level contributions of stored memory to the final answer and using them as local process rewards, replacing coarse outcome-level RL signals and heuristic memory rules; beats retrieval, heuristic, and...
Learns the memory-construction policy itself, what to extract, store, update, compress, or discard, by attributing token-level contributions of stored memory to the final answer and using them as local process rewards, replacing coarse outcome-level RL signals and heuristic memory rules; beats retrieval, heuristic, and...
Learns the memory-construction policy itself, what to extract, store, update, compress, or discard, by attributing token-level contributions of stored memory to the final answer and using them as local process rewards, replacing coarse outcome-level RL signals and heuristic memory rules; beats retrieval, heuristic, and...
The work turns loop quality into a measurable task or score. Learns the memory-construction policy itself, what to extract, store, update, compress, or discard, by attributing token-level contributions of stored memory to the final answer and using them as local process rewards, replacing coarse outcome-level RL signal...
Use AttriMem: Attribution-Guided Process Feedback for Agent Memory Learning to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.21106; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,341
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1341
2026-07-24
Persist
persist
Carry context, state, and receipts across runs.
context;verification;budget
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.21106
[2607.21106] AttriMem: Attribution-Guided Process Feedback for Agent Memory Construction
Effective memory is crucial for LLM agents, yet constructing it effectively remains challenging. A memory-construction policy decides what information to extract, store, update, compress, or discard as interactions accumulate. Heuristic memory methods rely on subjective, task-specific rules, which can misalign with dow...
Qinfeng Li; Yuntai Bao; Xinyan Yu; Hongze Chen; Yanming Liu; Huifeng Zhu; Yier Jin; Jintao Chen; Wenqi Zhang; Xuhong Zhang
2026-07-23
2026
arXiv
arXiv
cs.AI
arxiv-api
2607.21106
2026-09-08T16:27:48
ale-0652
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Agentic Context Management: Solving Agent Memory and Cost by Treating Them as Lifecycle and Architecture Problems
https://arxiv.org/abs/2607.21503
external
arxiv.org
Reframes agent memory and runaway token cost from a storage-and-retrieval problem to a context lifecycle and architecture problem, arguing production agent failures stem from unmanaged reasoning context (histories, tool definitions, ballooning tool outputs).
Reframes agent memory and runaway token cost from a storage-and-retrieval problem to a context lifecycle and architecture problem, arguing production agent failures stem from unmanaged reasoning context (histories, tool definitions, ballooning tool outputs).
Reframes agent memory and runaway token cost from a storage-and-retrieval problem to a context lifecycle and architecture problem, arguing production agent failures stem from unmanaged reasoning context (histories, tool definitions, ballooning tool outputs).
Persistent memory is treated as an external runtime artifact. Reframes agent memory and runaway token cost from a storage-and-retrieval problem to a context lifecycle and architecture problem, arguing production agent failures stem from unmanaged reasoning context (histories, tool definitions, ballooning tool outputs).
Use Agentic Context Management: Solving Agent Memory and Cost by Treating Them as Lifecycle and Architecture Problems to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.21503; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,342
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1342
2026-07-25
Persist
persist
Carry context, state, and receipts across runs.
workspace;context;budget
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.21503
[2607.21503] Agentic Context Management: Solving Agent Memory and Cost by Treating Them as Lifecycle and Architecture Problems
Production AI agents' failures are less often due to an inability to reason well and more often because they cannot manage what is in their reasoning context: conversation histories, large prompts, large tool definitions, and ballooning tool outputs. Agents drown in their own accumulating history while paying a token c...
Gaurav Dadhich
2026-07-23
2026
arXiv
arXiv
23 pages, 6 figures, 4 tables. Evaluation harness and study data: github.com/maximem-ai
cs.AI
arxiv-api
2607.21503
2026-09-08T16:27:48
ale-0653
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
FedAgentKE: Federated Semantic Knowledge Evolution for Heterogeneous Agents
https://arxiv.org/abs/2607.21361
external
arxiv.org
Federated evolution of agent knowledge: heterogeneous agent frameworks iteratively distill, aggregate, and adapt semantic reasoning abstractions so locally-learned experience (workflow reuse, memory) transfers across systems without sharing raw reasoning trajectories, extending memory-driven self-improvement from singl...
Federated evolution of agent knowledge: heterogeneous agent frameworks iteratively distill, aggregate, and adapt semantic reasoning abstractions so locally-learned experience (workflow reuse, memory) transfers across systems without sharing raw reasoning trajectories, extending memory-driven self-improvement from singl...
Federated evolution of agent knowledge: heterogeneous agent frameworks iteratively distill, aggregate, and adapt semantic reasoning abstractions so locally-learned experience (workflow reuse, memory) transfers across systems without sharing raw reasoning trajectories, extending memory-driven self-improvement from singl...
Persistent memory is treated as an external runtime artifact. Federated evolution of agent knowledge: heterogeneous agent frameworks iteratively distill, aggregate, and adapt semantic reasoning abstractions so locally-learned experience (workflow reuse, memory) transfers across systems without sharing raw reasoning tra...
Use FedAgentKE: Federated Semantic Knowledge Evolution for Heterogeneous Agents to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.21361; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,343
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1343
2026-07-25
Persist
persist
Carry context, state, and receipts across runs.
context
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.21361
[2607.21361] FedAgentKE: Federated Semantic Knowledge Evolution for Heterogeneous Agents
Large language model (LLM)-based agents increasingly rely on reasoning, tool use, and iterative execution, yet existing agent frameworks still operate largely in isolation. While recent memory-based agent systems improve individual agents through local retrieval and workflow reuse, local experiences remain fragmented a...
Weihao Li; Jun Bai; Ziyang Song
2026-07-23
2026
arXiv
arXiv
9 pages (including appendix)
cs.MA
arxiv-api
2607.21361
2026-09-08T16:27:48
ale-0654
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
MemTX: Transactional Belief Commit for Stateful Agent Memory
https://arxiv.org/abs/2607.23929
external
arxiv.org
Treats an agent memory write as a database transaction rather than instant truth. Records carry evidence, permissions, provenance, and validity; writes are staged under snapshot isolation and admitted by a validate-and-commit pipeline; irreversible tool calls are gated on in-flight belief state; retracting a belief tri...
Treats an agent memory write as a database transaction rather than instant truth. Records carry evidence, permissions, provenance, and validity; writes are staged under snapshot isolation and admitted by a validate-and-commit pipeline; irreversible tool calls are gated on in-flight belief state; retracting a belief tri...
Treats an agent memory write as a database transaction rather than instant truth. Records carry evidence, permissions, provenance, and validity; writes are staged under snapshot isolation and admitted by a validate-and-commit pipeline; irreversible tool calls are gated on in-flight belief state; retracting a belief tri...
Persistent memory is treated as an external runtime artifact. Treats an agent memory write as a database transaction rather than instant truth. Records carry evidence, permissions, provenance, and validity; writes are staged under snapshot isolation and admitted by a validate-and-commit pipeline; irreversible tool call...
Use MemTX: Transactional Belief Commit for Stateful Agent Memory to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.23929; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,344
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1344
2026-07-28
Persist
persist
Carry context, state, and receipts across runs.
trigger;workspace;context;state
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.23929
[2607.23929] MemTX: Transactional Belief Commit for Stateful Agent Memory
LLM agents increasingly coordinate through persistent shared memory: one agent's write becomes another agent's premise, and eventually a tool call with real side effects. Current agent memory systems treat every accepted write as immediately actionable truth, so a polluted tool result, a stale update, or a teammate's h...
Xiaoyang Li; Yiqi Wang; Haohui Lu; Zhi Chen; Mo Li; Pingan Song; Mingkai Zheng; Taotao Cai
2026-07-27
2026
arXiv
arXiv
Preprint
cs.AI
arxiv-api
2607.23929
2026-09-08T16:27:48
ale-0655
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
ConsistencyGate: Preventing Memory Contamination in LLM Agents via Self-Consistency Admission Control
https://arxiv.org/abs/2607.22962
external
arxiv.org
Identifies memory contamination -- a hallucinated fact written once persists as a false premise for every subsequent step -- and points out existing memory management handles retrieval and capacity but never write-time correctness.
Identifies memory contamination -- a hallucinated fact written once persists as a false premise for every subsequent step -- and points out existing memory management handles retrieval and capacity but never write-time correctness.
Identifies memory contamination -- a hallucinated fact written once persists as a false premise for every subsequent step -- and points out existing memory management handles retrieval and capacity but never write-time correctness.
Persistent memory is treated as an external runtime artifact. Identifies memory contamination -- a hallucinated fact written once persists as a false premise for every subsequent step -- and points out existing memory management handles retrieval and capacity but never write-time correctness.
Use ConsistencyGate: Preventing Memory Contamination in LLM Agents via Self-Consistency Admission Control to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.22962; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,345
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1345
2026-07-28
Persist
persist
Carry context, state, and receipts across runs.
context;state
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.22962
[2607.22962] ConsistencyGate: Preventing Memory Contamination in LLM Agents via Self-Consistency Admission Control
LLM agents that operate over many turns accumulate facts in an external memory store and reuse them as premises for downstream reasoning. A hallucinated fact written at one step therefore persists as a false premise for every subsequent step, a failure mode we call memory contamination. Existing memory management addre...
Yan Zhang; Shibo Li
2026-07-25
2026
arXiv
arXiv
24 pages, 2 figures, 6 tables, 1 algorithm; includes appendices
cs.AI
arxiv-api
2607.22962
2026-09-08T16:27:48
ale-0656
State, Memory, And Context Persistence
state-memory-and-context-persistence
Benchmark
🧪
Ground Truth First: A Longitudinal Evaluation Instrument for Agent Memory, and the Tenure Crossover in Memory-Architecture Rankings
https://arxiv.org/abs/2607.21962
external
arxiv.org
Inverts the standard agent-memory benchmark pipeline: instead of generating conversations then extracting answer keys (with documented label-error and contamination problems), a seeded life-script sampler emits facts with validity intervals, volatility classes, and source channels before any text exists, then renders c...
Inverts the standard agent-memory benchmark pipeline: instead of generating conversations then extracting answer keys (with documented label-error and contamination problems), a seeded life-script sampler emits facts with validity intervals, volatility classes, and source channels before any text exists, then renders c...
Inverts the standard agent-memory benchmark pipeline: instead of generating conversations then extracting answer keys (with documented label-error and contamination problems), a seeded life-script sampler emits facts with validity intervals, volatility classes, and source channels before any text exists, then renders c...
Evaluation data is used as the feedback signal for improving loop behavior. Inverts the standard agent-memory benchmark pipeline: instead of generating conversations then extracting answer keys (with documented label-error and contamination problems), a seeded life-script sampler emits facts with validity intervals, vo...
Use Ground Truth First: A Longitudinal Evaluation Instrument for Agent Memory, and the Tenure Crossover in Memory-Architecture Rankings to carry context, state, and receipts across runs and failures.
Benchmark or leaderboard source with repeatable tasks or scores; compare systems only after checking setup and scope.
high
README.md
1,346
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1346
2026-07-28
Persist
persist
Carry context, state, and receipts across runs.
context;verification
researcher;evaluator
harness
enabling
benchmark
A
ok
https://arxiv.org/abs/2607.21962
[2607.21962] Ground Truth First: A Longitudinal Evaluation Instrument for Agent Memory, and the Tenure Crossover in Memory-Architecture Rankings
Benchmarks for LLM-agent memory typically generate conversations first and extract answer keys afterwards -- with documented label-error and contamination problems -- and they overwhelmingly measure short interaction histories. We invert the pipeline: a seeded life-script sampler emits facts with validity intervals, vo...
Quentin Spencer
2026-07-24
2026
arXiv
arXiv
25 pages, 2 figures. Code: https://github.com/veracium-ai/Veracium
cs.CL
arxiv-api
2607.21962
2026-09-08T16:27:48
ale-0657
State, Memory, And Context Persistence
state-memory-and-context-persistence
Benchmark
🧪
Keep It InMind: Benchmarking the Implicit-Association Blind Spot in Agent Memory
https://arxiv.org/abs/2607.24368
external
arxiv.org
Exposes the unstated assumption behind every retrieval-based long-term memory system: that a needed memory resembles the query needing it. A stored tree-nut allergy should change the answer to a macaron request via almond flour, yet the two texts share no retrievable cue.
Exposes the unstated assumption behind every retrieval-based long-term memory system: that a needed memory resembles the query needing it. A stored tree-nut allergy should change the answer to a macaron request via almond flour, yet the two texts share no retrievable cue.
Exposes the unstated assumption behind every retrieval-based long-term memory system: that a needed memory resembles the query needing it. A stored tree-nut allergy should change the answer to a macaron request via almond flour, yet the two texts share no retrievable cue.
Persistent memory is treated as an external runtime artifact. Exposes the unstated assumption behind every retrieval-based long-term memory system: that a needed memory resembles the query needing it. A stored tree-nut allergy should change the answer to a macaron request via almond flour, yet the two texts share no re...
Use Keep It InMind: Benchmarking the Implicit-Association Blind Spot in Agent Memory to carry context, state, and receipts across runs and failures.
Benchmark or leaderboard source with repeatable tasks or scores; compare systems only after checking setup and scope.
high
README.md
1,347
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1347
2026-07-28
Persist
persist
Carry context, state, and receipts across runs.
context
researcher;evaluator
harness
enabling
benchmark
A
ok
https://arxiv.org/abs/2607.24368
[2607.24368] Keep It InMind: Benchmarking the Implicit-Association Blind Spot in Agent Memory
Long-term memory systems store what a user says in an external store and retrieve it when a related query arrives. This interface rests on an assumption so natural that it is rarely stated: a memory that is needed will resemble the query that needs it. World knowledge breaks the assumption. A tree-nut allergy should ch...
Ruizhe Li; Mingxuan Du; Benfeng Xu; Zhendong Mao
2026-07-27
2026
arXiv
arXiv
cs.CL
arxiv-api
2607.24368
2026-09-08T16:27:48
ale-0658
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
ACM: Agentic Context Management for Long Horizon Tasks
https://arxiv.org/abs/2607.23809
external
arxiv.org
Gives the agent purpose-built context editing tools so it decides when to compress rather than firing on rigid heuristic thresholds, offloading discarded content to external memory and querying it on demand -- lossless context management modeled on short-term/long-term human memory, plus a post-training pipeline that t...
Gives the agent purpose-built context editing tools so it decides when to compress rather than firing on rigid heuristic thresholds, offloading discarded content to external memory and querying it on demand -- lossless context management modeled on short-term/long-term human memory, plus a post-training pipeline that t...
Gives the agent purpose-built context editing tools so it decides when to compress rather than firing on rigid heuristic thresholds, offloading discarded content to external memory and querying it on demand -- lossless context management modeled on short-term/long-term human memory, plus a post-training pipeline that t...
Persistent memory is treated as an external runtime artifact. Gives the agent purpose-built context editing tools so it decides when to compress rather than firing on rigid heuristic thresholds, offloading discarded content to external memory and querying it on demand -- lossless context management modeled on short-ter...
Use ACM: Agentic Context Management for Long Horizon Tasks to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.23809; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,348
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1348
2026-07-28
Persist
persist
Carry context, state, and receipts across runs.
workspace;context;escalation
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.23809
[2607.23809] ACM: Agentic Context Management for Long Horizon Tasks
Agentic tasks are inherently long-horizon and multi-turn, constantly accumulating context through interactions with the environment. Existing context compression methods inevitably incur information loss and are triggered by rigid heuristic rules, leaving them misaligned with the agent's evolving reasoning focus. We pr...
Xiaochuan Li; Ryan Ming; Meng Chu; Shuai Shao; Rong Jin; Chenyan Xiong
2026-07-26
2026
arXiv
arXiv
cs.AI
arxiv-api
2607.23809
2026-09-08T16:27:48
ale-0659
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
MemChain: Learning Interpretable Memory Traces for Memory-Augmented LLM Agents
https://arxiv.org/abs/2607.24097
external
arxiv.org
Challenges the retrieval-as-evidence default where retrieved memories are dumped straight into the answer model, leaving it to resolve redundancy, conflicts, and weak relevance at high context cost.
Challenges the retrieval-as-evidence default where retrieved memories are dumped straight into the answer model, leaving it to resolve redundancy, conflicts, and weak relevance at high context cost.
Challenges the retrieval-as-evidence default where retrieved memories are dumped straight into the answer model, leaving it to resolve redundancy, conflicts, and weak relevance at high context cost.
Persistent memory is treated as an external runtime artifact. Challenges the retrieval-as-evidence default where retrieved memories are dumped straight into the answer model, leaving it to resolve redundancy, conflicts, and weak relevance at high context cost.
Use MemChain: Learning Interpretable Memory Traces for Memory-Augmented LLM Agents to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.24097; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,349
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1349
2026-07-28
Persist
persist
Carry context, state, and receipts across runs.
context;budget
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.24097
[2607.24097] MemChain: Learning Interpretable Memory Traces for Memory-Augmented LLM Agents
Memory-augmented LLM agents typically answer queries by retrieving relevant memories and feeding them directly to an answer model. This retrieval-as-evidence paradigm assumes retrieved memories are already suitable for reasoning, leaving the answer model to resolve redundancy, conflicts, and weak relevance while incurr...
Yiwen Ma; Songjun Tu; Qichao Zhang; Dong Li; Linjing Li; Dongbin Zhao
2026-07-27
2026
arXiv
arXiv
cs.AI
arxiv-api
2607.24097
2026-09-08T16:27:48
ale-0660
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Eviction as Estimation: A Fixed-Lag Smoothing View of Test-Time Memory, and When Measuring Beats Accumulating
https://arxiv.org/abs/2607.24667
external
arxiv.org
Recasts bounded working-memory eviction as estimation of a hidden signal (will this item be reused), placing StreamingLLM, H2O, and SnapKV on a single axis: commit lag H. All deployed methods commit at H=0; Belady's optimum sits at full future knowledge.
Recasts bounded working-memory eviction as estimation of a hidden signal (will this item be reused), placing StreamingLLM, H2O, and SnapKV on a single axis: commit lag H. All deployed methods commit at H=0; Belady's optimum sits at full future knowledge.
Recasts bounded working-memory eviction as estimation of a hidden signal (will this item be reused), placing StreamingLLM, H2O, and SnapKV on a single axis: commit lag H. All deployed methods commit at H=0; Belady's optimum sits at full future knowledge.
Persistent memory is treated as an external runtime artifact. Recasts bounded working-memory eviction as estimation of a hidden signal (will this item be reused), placing StreamingLLM, H2O, and SnapKV on a single axis: commit lag H. All deployed methods commit at H=0; Belady's optimum sits at full future knowledge.
Use Eviction as Estimation: A Fixed-Lag Smoothing View of Test-Time Memory, and When Measuring Beats Accumulating to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.24667; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,350
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1350
2026-07-28
Persist
persist
Carry context, state, and receipts across runs.
context;verification
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.24667
[2607.24667] Eviction as Estimation: A Fixed-Lag Smoothing View of Test-Time Memory, and When Measuring Beats Accumulating
A language model with a bounded working memory must repeatedly decide which stored items to keep. Every deployed method decides the moment an item arrives, from the past (StreamingLLM, H2O) or from a guess about the future (SnapKV). We recast the choice as an estimation problem on a hidden signal, whether an item will ...
Maruthi Vemula; Neeraj Praneeth Gajula
2026-07-27
2026
arXiv
arXiv
8 pages, 3 figures, 3 tables
cs.AI
arxiv-api
2607.24667
2026-09-08T16:27:48
ale-0661
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
StateAct: Program State, before Pixels, for Long-Horizon Computer-Use Agents
https://arxiv.org/abs/2607.22798
external
arxiv.org
Argues computer-use agents are over-invested in perception when a screenshot is a lossy rendering of the real target -- files, application backends, DOM. StateAct is a code-first multi-agent harness where the main agent manipulates program state through code and a GUI subagent handles screenshot-and-click only where re...
Argues computer-use agents are over-invested in perception when a screenshot is a lossy rendering of the real target -- files, application backends, DOM. StateAct is a code-first multi-agent harness where the main agent manipulates program state through code and a GUI subagent handles screenshot-and-click only where re...
Argues computer-use agents are over-invested in perception when a screenshot is a lossy rendering of the real target -- files, application backends, DOM. StateAct is a code-first multi-agent harness where the main agent manipulates program state through code and a GUI subagent handles screenshot-and-click only where re...
The work separates roles across agents, verifiers, or orchestration layers. Argues computer-use agents are over-invested in perception when a screenshot is a lossy rendering of the real target -- files, application backends, DOM. StateAct is a code-first multi-agent harness where the main agent manipulates program stat...
Use StateAct: Program State, before Pixels, for Long-Horizon Computer-Use Agents to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.22798; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,351
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1351
2026-07-28
Persist
persist
Carry context, state, and receipts across runs.
delegation;state
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.22798
[2607.22798] StateAct: Program State, before Pixels, for Long-Horizon Computer-Use Agents
Computer-use agents are usually improved by strengthening perception: better models for reading a screenshot and choosing where to click. Yet a screenshot is only a lossy rendering of the underlying program state, e.g., the files, application backends, and DOM that hold the task data. Different states can produce the s...
Yan Yang; Xiangru Jian; Ziyang Luo; Zirui Zhao; Yutong Dai; Ziji Shi; Hanshu Yan; Jun Hao Liew; Silvio Savarese; Junnan Li
2026-07-24
2026
arXiv
arXiv
cs.SE
arxiv-api
2607.22798
2026-09-08T16:27:48
ale-0662
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
MEMENTO: Memory-Guided Memetic Code-as-Policy Evolution
https://arxiv.org/abs/2607.22832
external
arxiv.org
Frames long-horizon embodied policy improvement as execution-guided program search: represent the policy as an inspectable control program, revise it after rollout evaluation, re-execute, compare. Notes that existing LLM-driven evolutionary approaches only select among independently generated variants and skip sequenti...
Frames long-horizon embodied policy improvement as execution-guided program search: represent the policy as an inspectable control program, revise it after rollout evaluation, re-execute, compare. Notes that existing LLM-driven evolutionary approaches only select among independently generated variants and skip sequenti...
Frames long-horizon embodied policy improvement as execution-guided program search: represent the policy as an inspectable control program, revise it after rollout evaluation, re-execute, compare. Notes that existing LLM-driven evolutionary approaches only select among independently generated variants and skip sequenti...
Evaluation data is used as the feedback signal for improving loop behavior. Frames long-horizon embodied policy improvement as execution-guided program search: represent the policy as an inspectable control program, revise it after rollout evaluation, re-execute, compare. Notes that existing LLM-driven evolutionary app...
Use MEMENTO: Memory-Guided Memetic Code-as-Policy Evolution to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.22832; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,352
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1352
2026-07-28
Persist
persist
Carry context, state, and receipts across runs.
context;verification
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.22832
[2607.22832] MEMENTO: Memory-Guided Memetic Code-as-Policy Evolution
Long-horizon embodied tasks require policies that execute many dependent actions before task success can be observed. Representing policies as executable control pro- grams (code-as-policy) enables their decision logic to be inspected and revised after rollout evaluation. Revised programs can then be executed and compa...
Alkis Sygkounas; Victor Aregbede; Amy Loutfi; Andreas Persson
2026-07-24
2026
arXiv
arXiv
cs.LG
arxiv-api
2607.22832
2026-09-08T16:27:48
ale-0663
State, Memory, And Context Persistence
state-memory-and-context-persistence
Tool
🧰
OptMem
https://github.com/VictorTaelin/OptMem
external
github.com
Minimalist persistent memory for agents: an append-only LOG.txt plus a binary tree of pairwise summaries at increasing levels, exposed through three commands (`memo wake` to rehydrate at session start, `memo note` to record ≤280 chars, `memo recall <regex>` to search the full log).
Minimalist persistent memory for agents: an append-only LOG.txt plus a binary tree of pairwise summaries at increasing levels, exposed through three commands (`memo wake` to rehydrate at session start, `memo note` to record ≤280 chars, `memo recall <regex>` to search the full log).
Minimalist persistent memory for agents: an append-only LOG.txt plus a binary tree of pairwise summaries at increasing levels, exposed through three commands (`memo wake` to rehydrate at session start, `memo note` to record ≤280 chars, `memo recall <regex>` to search the full log).
Persistent memory is treated as an external runtime artifact. Minimalist persistent memory for agents: an append-only LOG.txt plus a binary tree of pairwise summaries at increasing levels, exposed through three commands (`memo wake` to rehydrate at session start, `memo note` to record ≤280 chars, `memo recall <regex>` ...
Use OptMem to carry context, state, and receipts across runs and failures.
Inspectable GitHub source; popularity is context, not proof of reliability.
medium
README.md
1,353
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1353
2026-07-28
Persist
persist
Carry context, state, and receipts across runs.
context;state
builder
harness
enabling
source-implementation
A
ok
https://github.com/VictorTaelin/OptMem
GitHub - VictorTaelin/OptMem: Permanent memory for AI agents. A 426-token prompt, a script, plug and play. · GitHub
Permanent memory for AI agents. A 426-token prompt, a script, plug and play. - VictorTaelin/OptMem
VictorTaelin/OptMem
GitHub
html-meta
VictorTaelin/OptMem
2026-09-08T16:27:48
ale-0664
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
UniMem: Complementary Episodic-to-Parametric Memory for Boundary-Agnostic Task Streams
https://arxiv.org/abs/2607.26017
external
arxiv.org
Learnable routing tokens arbitrate between an episodic buffer for novel tasks and expandable parametric memory for recurring patterns, addressing stability-plasticity for agents on task streams with no clean task boundaries. The consolidation direction, episodic experience graduating into parameters, is the piece most ...
Learnable routing tokens arbitrate between an episodic buffer for novel tasks and expandable parametric memory for recurring patterns, addressing stability-plasticity for agents on task streams with no clean task boundaries. The consolidation direction, episodic experience graduating into parameters, is the piece most ...
Learnable routing tokens arbitrate between an episodic buffer for novel tasks and expandable parametric memory for recurring patterns, addressing stability-plasticity for agents on task streams with no clean task boundaries. The consolidation direction, episodic experience graduating into parameters, is the piece most ...
Persistent memory is treated as an external runtime artifact. Learnable routing tokens arbitrate between an episodic buffer for novel tasks and expandable parametric memory for recurring patterns, addressing stability-plasticity for agents on task streams with no clean task boundaries. The consolidation direction, epis...
Use UniMem: Complementary Episodic-to-Parametric Memory for Boundary-Agnostic Task Streams to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.26017; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,354
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1354
2026-07-30
Persist
persist
Carry context, state, and receipts across runs.
context;budget
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.26017
[2607.26017] UniMem: Complementary Episodic-to-Parametric Memory for Boundary-Agnostic Task Streams
Memory is essential for LLM agents to accumulate task experience and reuse task-specific execution strategies. However, real-world deployment over boundary-agnostic and evolving task streams exposes a fundamental stability-plasticity dilemma. External retrieval-based memory can rapidly absorb new evidence, but it often...
Siyu Xia; Chenheng Zhang; Yanting Wu; Haoxuan Li; Jiajun Chai; Xiaohan Wang; Guojun Yin; Wei Lin; Zhouchen Lin; Haifeng Zhang; Jun Wang
2026-07-28
2026
arXiv
arXiv
cs.CL
arxiv-api
2607.26017
2026-09-08T16:27:48
ale-0665
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Addressable Recall Compaction for Long Context-Window Control in AI Agents
https://arxiv.org/abs/2607.25066
external
arxiv.org
ARC stores tool observations in an append-only addressable log and swaps in compact ID citations under context pressure, letting the agent recall detail without re-running tools, 99.40% on Needle-in-a-Haystack and 29.97% on LongBench-v2 Hard. A concrete alternative to lossy summarization compaction.
ARC stores tool observations in an append-only addressable log and swaps in compact ID citations under context pressure, letting the agent recall detail without re-running tools, 99.40% on Needle-in-a-Haystack and 29.97% on LongBench-v2 Hard. A concrete alternative to lossy summarization compaction.
ARC stores tool observations in an append-only addressable log and swaps in compact ID citations under context pressure, letting the agent recall detail without re-running tools, 99.40% on Needle-in-a-Haystack and 29.97% on LongBench-v2 Hard. A concrete alternative to lossy summarization compaction.
Context is managed as durable loop state rather than a single prompt payload. ARC stores tool observations in an append-only addressable log and swaps in compact ID citations under context pressure, letting the agent recall detail without re-running tools, 99.40% on Needle-in-a-Haystack and 29.97% on LongBench-v2 Hard....
Use Addressable Recall Compaction for Long Context-Window Control in AI Agents to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.25066; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,355
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1355
2026-07-30
Persist
persist
Carry context, state, and receipts across runs.
workspace;context
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.25066
[2607.25066] Addressable Recall Compaction for Long Context-Window Control in AI Agents
Long-horizon LLM agents accumulate reasoning traces, actions, and tool observations that can eventually exceed a model's fixed context window. Existing compaction methods address this limitation by discarding, summarizing, or retrieving earlier information, but they may remove task-critical details or fail to recover t...
Thang Dang; Yuma Ichikawa; Sakina Fatima; Koichi Shirahata
2026-07-27
2026
arXiv
arXiv
20 pages, 2 figures
cs.AI
arxiv-api
2607.25066
2026-09-08T16:27:48
ale-0666
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Filesystem-Based Memory for LLM Agents: Organization, Evolution, and Sustainability
https://arxiv.org/abs/2607.26637
external
arxiv.org
First systematic study of the now-ubiquitous markdown-files-on-disk memory pattern, asking whether agents can keep a growing store organized and whether that organization actually pays off in retrieval cost, answer quality, and memory health. Evaluates the default memory design most practitioners already run without ev...
First systematic study of the now-ubiquitous markdown-files-on-disk memory pattern, asking whether agents can keep a growing store organized and whether that organization actually pays off in retrieval cost, answer quality, and memory health. Evaluates the default memory design most practitioners already run without ev...
First systematic study of the now-ubiquitous markdown-files-on-disk memory pattern, asking whether agents can keep a growing store organized and whether that organization actually pays off in retrieval cost, answer quality, and memory health. Evaluates the default memory design most practitioners already run without ev...
Persistent memory is treated as an external runtime artifact. First systematic study of the now-ubiquitous markdown-files-on-disk memory pattern, asking whether agents can keep a growing store organized and whether that organization actually pays off in retrieval cost, answer quality, and memory health. Evaluates the d...
Use Filesystem-Based Memory for LLM Agents: Organization, Evolution, and Sustainability to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.26637; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,356
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1356
2026-07-30
Persist
persist
Carry context, state, and receipts across runs.
context;budget
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.26637
[2607.26637] Filesystem-Based Memory for LLM Agents: Organization, Evolution, and Sustainability
Deployed LLM agents increasingly keep their long-term memory as a filesystem: a directory tree of markdown files that the agent itself reads, writes, and reorganizes through generic file tools. Yet research has largely passed over this medium: prior systems design bespoke memory representations and study retrieval over...
Sizhe Zhou; Sheldon Yu; Hui Wei; Junda Wu; Siru Ouyang; Yizhu Jiao; Shijia Pan; Julian McAuley; Yu Zhang; Tong Yu; Jiawei Han
2026-07-29
2026
arXiv
arXiv
59 pages, 12 figures, 18 tables
cs.CL
arxiv-api
2607.26637
2026-09-08T16:27:48
ale-0667
State, Memory, And Context Persistence
state-memory-and-context-persistence
Tool
🧰
MemLens: A Value-Aware Memory Management System with Interactive Analytics for LLM-based Agents
https://arxiv.org/abs/2607.25992
external
arxiv.org
Scores individual memory records with Shapley-style attribution and exposes an interactive dashboard for inspecting memory value, hierarchy, and competing management strategies. Rare tooling for the operator side of agent memory, what is in there, what is it worth, what should be evicted.
Scores individual memory records with Shapley-style attribution and exposes an interactive dashboard for inspecting memory value, hierarchy, and competing management strategies. Rare tooling for the operator side of agent memory, what is in there, what is it worth, what should be evicted.
Scores individual memory records with Shapley-style attribution and exposes an interactive dashboard for inspecting memory value, hierarchy, and competing management strategies. Rare tooling for the operator side of agent memory, what is in there, what is it worth, what should be evicted.
Persistent memory is treated as an external runtime artifact. Scores individual memory records with Shapley-style attribution and exposes an interactive dashboard for inspecting memory value, hierarchy, and competing management strategies. Rare tooling for the operator side of agent memory, what is in there, what is it...
Use MemLens: A Value-Aware Memory Management System with Interactive Analytics for LLM-based Agents to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.25992; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,357
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1357
2026-07-30
Persist
persist
Carry context, state, and receipts across runs.
context
builder
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.25992
[2607.25992] MemLens: A Value-Aware Memory Management System with Interactive Analytics for LLM-based Agents
Recently, memory management has become a key infrastructure for LLM-based agents, as it directly affects long-horizon reasoning, personalized responses, and knowledge reuse. However, existing LLM memory systems typically adopt a coarse-grained (utility-agnostic) manner that treats heterogeneous user-LLM interaction rec...
Shuyue Wei; Chang Liu; Zimu Zhou; Yongxin Tong; Lizhen Cui
2026-07-28
2026
arXiv
arXiv
cs.DB
arxiv-api
2607.25992
2026-09-08T16:27:48
ale-0668
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
A Graph-Native Bitemporal Memory Store for Conversational AI Agents
https://arxiv.org/abs/2607.26520
external
arxiv.org
Agent-local Neo4j store combining vector indexes with bitemporal modeling so facts carry both valid-time and transaction-time, reaching 46.7% overall recall and 80% on knowledge-update questions where naive stores overwrite silently. Honest about weak temporal-reasoning performance, which is useful signal.
Agent-local Neo4j store combining vector indexes with bitemporal modeling so facts carry both valid-time and transaction-time, reaching 46.7% overall recall and 80% on knowledge-update questions where naive stores overwrite silently. Honest about weak temporal-reasoning performance, which is useful signal.
Agent-local Neo4j store combining vector indexes with bitemporal modeling so facts carry both valid-time and transaction-time, reaching 46.7% overall recall and 80% on knowledge-update questions where naive stores overwrite silently. Honest about weak temporal-reasoning performance, which is useful signal.
Control flow is represented as an inspectable graph rather than an opaque prompt loop. Agent-local Neo4j store combining vector indexes with bitemporal modeling so facts carry both valid-time and transaction-time, reaching 46.7% overall recall and 80% on knowledge-update questions where naive stores overwrite silently....
Use A Graph-Native Bitemporal Memory Store for Conversational AI Agents to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.26520; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,358
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1358
2026-07-30
Persist
persist
Carry context, state, and receipts across runs.
context
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.26520
[2607.26520] A Graph-Native Bitemporal Memory Store for Conversational AI Agents
Conversational AI agents commonly lack persistent memory across sessions. The obvious fixes like injecting full chat histories into the context window, or delegating to a third-party memory service, either exhaust the model's context budget or send personal data through infrastructure the user does not control. We desc...
Alp Niksarli; Gopesh Baheti
2026-07-29
2026
arXiv
arXiv
cs.DB
arxiv-api
2607.26520
2026-09-08T16:27:48
ale-0669
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
HiSkill: Empowering LLM Agents with Hierarchical Skill Graphs
https://arxiv.org/abs/2607.25853
external
arxiv.org
Organizes past trajectories into a directed graph linking high-level skills to executable action templates across several relation types, then retrieves task-relevant subgraphs at inference to guide skill switching and action selection. Turns accumulated experience into navigable structure rather than a flat trajectory...
Organizes past trajectories into a directed graph linking high-level skills to executable action templates across several relation types, then retrieves task-relevant subgraphs at inference to guide skill switching and action selection. Turns accumulated experience into navigable structure rather than a flat trajectory...
Organizes past trajectories into a directed graph linking high-level skills to executable action templates across several relation types, then retrieves task-relevant subgraphs at inference to guide skill switching and action selection. Turns accumulated experience into navigable structure rather than a flat trajectory...
Control flow is represented as an inspectable graph rather than an opaque prompt loop. Organizes past trajectories into a directed graph linking high-level skills to executable action templates across several relation types, then retrieves task-relevant subgraphs at inference to guide skill switching and action selecti...
Use HiSkill: Empowering LLM Agents with Hierarchical Skill Graphs to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.25853; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,359
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1359
2026-07-30
Persist
persist
Carry context, state, and receipts across runs.
context;state
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.25853
[2607.25853] HiSkill: Empowering LLM Agents with Hierarchical Skill Graphs
Skills have become an important abstraction for enabling large language model (LLM) agents to reuse past experience in long-horizon interactive tasks. However, existing trajectory-to-skill methods often produce flat collections of high-level textual skills that are stored and retrieved independently, leaving skill rela...
Yu Hao; Jinxuan Cai; Qi Zhang; Yawen Li; Zhiqiang Zhang; Chuan Shi; Cheng Yang
2026-07-28
2026
arXiv
arXiv
cs.AI
arxiv-api
2607.25853
2026-09-08T16:27:48
ale-0670
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
VITAL-RAG: Invariance Race for Context Allocation in Coding Agents
https://arxiv.org/abs/2607.26937
external
arxiv.org
Names an invariance race between cutting redundancy and preserving task-relevant code, then organizes retrieved evidence by code object with companion selection to raise recall while spending fewer tokens. Sharpens the context-budget problem specific to repository-scale coding loops.
Names an invariance race between cutting redundancy and preserving task-relevant code, then organizes retrieved evidence by code object with companion selection to raise recall while spending fewer tokens. Sharpens the context-budget problem specific to repository-scale coding loops.
Names an invariance race between cutting redundancy and preserving task-relevant code, then organizes retrieved evidence by code object with companion selection to raise recall while spending fewer tokens. Sharpens the context-budget problem specific to repository-scale coding loops.
Context is managed as durable loop state rather than a single prompt payload. Names an invariance race between cutting redundancy and preserving task-relevant code, then organizes retrieved evidence by code object with companion selection to raise recall while spending fewer tokens. Sharpens the context-budget problem ...
Use VITAL-RAG: Invariance Race for Context Allocation in Coding Agents to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.26937; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,360
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1360
2026-07-30
Persist
persist
Carry context, state, and receipts across runs.
context;budget
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.26937
[2607.26937] VITAL-RAG: Invariance Race for Context Allocation in Coding Agents
Coding agents often retrieve code from an entire repository, but only limited evidence can fit into the final model input. Conventional retrieval-augmented generation (RAG) for coding agents treats fragments from the same code object as separate results, so redundant views can occupy multiple context positions and crow...
Zijian Lu; Yonghua Lu; Mingcai Chen; Yiping Zuo; Xin He; Weijun Wang; Weibei Fan
2026-07-29
2026
arXiv
arXiv
8 pages, 2 figures
cs.SE
arxiv-api
2607.26937
2026-09-08T16:27:48
ale-0671
State, Memory, And Context Persistence
state-memory-and-context-persistence
Tool
🧰
CodeNib: A Multi-View Data System for Serving Repository Context to Coding Agents
https://arxiv.org/abs/2607.25431
external
arxiv.org
Maintains lexical, dense, and structural views per repository commit that persist incrementally across code changes, giving agents ranked search and symbol navigation at far fewer tokens than grep-driven exploration. Treats repo context as a served, versioned index rather than something re-derived every session.
Maintains lexical, dense, and structural views per repository commit that persist incrementally across code changes, giving agents ranked search and symbol navigation at far fewer tokens than grep-driven exploration. Treats repo context as a served, versioned index rather than something re-derived every session.
Maintains lexical, dense, and structural views per repository commit that persist incrementally across code changes, giving agents ranked search and symbol navigation at far fewer tokens than grep-driven exploration. Treats repo context as a served, versioned index rather than something re-derived every session.
Context is managed as durable loop state rather than a single prompt payload. Maintains lexical, dense, and structural views per repository commit that persist incrementally across code changes, giving agents ranked search and symbol navigation at far fewer tokens than grep-driven exploration. Treats repo context as a ...
Use CodeNib: A Multi-View Data System for Serving Repository Context to Coding Agents to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.25431; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,361
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1361
2026-07-30
Persist
persist
Carry context, state, and receipts across runs.
context;state;budget
builder
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.25431
[2607.25431] CodeNib: A Multi-View Data System for Serving Repository Context to Coding Agents
Coding agents repeatedly search, navigate, and retain context from evolving repositories, but disconnected indexes, language servers, and task-local histories force repeated discovery and obscure lifecycle costs. CodeNib builds reusable lexical, dense, and structural views per repository commit, maps outputs to reposit...
Zhongming Yu; Hengjia Yu; Boqin Yuan; Shuting Zhao; Yizhao Chen; Aryan Dokania; Mihir Jagtap; Jiayu Chang; Yitong Ma; Yash Jayswal; Wentao Ni; Hejia Zhang; Zhaoling Chen; Gangda Deng; Jishen Zhao
2026-07-28
2026
arXiv
arXiv
cs.SE
arxiv-api
2607.25431
2026-09-08T16:27:48
ale-0672
State, Memory, And Context Persistence
state-memory-and-context-persistence
Blog
📝
Stateless MCP Has Recaptured My Interest
https://simonwillison.net/2026/Jul/31/stateless-mcp/
external
simonwillison.net
Simon Willison's practitioner read on what the 2026-07-28 MCP specification revision actually changes for people running agents at scale.
Simon Willison's practitioner read on what the 2026-07-28 MCP specification revision actually changes for people running agents at scale.
Simon Willison's practitioner read on what the 2026-07-28 MCP specification revision actually changes for people running agents at scale.
Makes persistence and context management visible as runtime design choices. Simon Willison's practitioner read on what the 2026-07-28 MCP specification revision actually changes for people running agents at scale.
Use Stateless MCP Has Recaptured My Interest to carry context, state, and receipts across runs and failures.
Contextual source from simonwillison.net; useful for practice signals or boundary conditions, not independent validation.
contextual
README.md
1,362
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1362
2026-08-02
Persist
persist
Carry context, state, and receipts across runs.
context;state
builder
harness
enabling
practitioner-analysis
B
ok
https://simonwillison.net/2026/Jul/31/stateless-mcp/
Stateless MCP has recaptured my interest (and inspired mcp-explorer and datasette-mcp)
Tuesday was Stateless MCP day—the rollout of MCP 2.0, or the 2026-07-28 Model Context Protocol specification to use the more formal but less memorable name. This is the most significant …
Simon Willison
2026
Simon Willison’s Weblog
html-meta
2026-09-08T16:27:48
ale-0673
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
MemTxn: A Transaction Boundary for Source-Supported Updates and Complete-State Recovery in Agent Memory
https://arxiv.org/abs/2607.27834
external
arxiv.org
Applies database transaction semantics to agent memory: an Ordered PatchTest validates every write against its source, a Temporal Resolver picks versions, and a durable snapshot journal recovers complete state after corruption. This is the memory-durability primitive most long-running loops lack, writes are gated, not ...
Applies database transaction semantics to agent memory: an Ordered PatchTest validates every write against its source, a Temporal Resolver picks versions, and a durable snapshot journal recovers complete state after corruption. This is the memory-durability primitive most long-running loops lack, writes are gated, not ...
Applies database transaction semantics to agent memory: an Ordered PatchTest validates every write against its source, a Temporal Resolver picks versions, and a durable snapshot journal recovers complete state after corruption. This is the memory-durability primitive most long-running loops lack, writes are gated, not ...
Durable execution and replay are treated as first-class loop infrastructure. Applies database transaction semantics to agent memory: an Ordered PatchTest validates every write against its source, a Temporal Resolver picks versions, and a durable snapshot journal recovers complete state after corruption. This is the mem...
Use MemTxn: A Transaction Boundary for Source-Supported Updates and Complete-State Recovery in Agent Memory to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.27834; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,363
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1363
2026-08-02
Persist
persist
Carry context, state, and receipts across runs.
context;state
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.27834
[2607.27834] MemTxn: A Transaction Boundary for Source-Supported Updates and Complete-State Recovery in Agent Memory
Persistent memory lets long-running large language model agents reuse information across sessions and tasks. Yet errors in writable memory can persist and corrupt future behavior. Existing systems improve storage and retrieval, but they do not provide a transaction boundary for reliable updates and recovery. We therefo...
Hanshuai Cui; Zhiqing Tang; Zhi Yao; Fanshuai Meng; Qianli Ma; Weijia Jia
2026-07-30
2026
arXiv
arXiv
cs.AI
arxiv-api
2607.27834
2026-09-08T16:27:48
ale-0674
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Zero-Mem: Zero-Token Memory Operations for LLM Agents
https://arxiv.org/abs/2607.29377
external
arxiv.org
Asks whether structured memory access requires generation at all, and answers no. Most agent memory systems spend extra LLM calls generating intermediate records and mediating retrieval, adding recurring token and latency cost while merged or omitted details obscure the original evidence.
Asks whether structured memory access requires generation at all, and answers no. Most agent memory systems spend extra LLM calls generating intermediate records and mediating retrieval, adding recurring token and latency cost while merged or omitted details obscure the original evidence.
Asks whether structured memory access requires generation at all, and answers no. Most agent memory systems spend extra LLM calls generating intermediate records and mediating retrieval, adding recurring token and latency cost while merged or omitted details obscure the original evidence.
Persistent memory is treated as an external runtime artifact. Asks whether structured memory access requires generation at all, and answers no. Most agent memory systems spend extra LLM calls generating intermediate records and mediating retrieval, adding recurring token and latency cost while merged or omitted details...
Use Zero-Mem: Zero-Token Memory Operations for LLM Agents to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.29377; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,364
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1364
2026-08-05
Persist
persist
Carry context, state, and receipts across runs.
context;budget
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.29377
[2607.29377] Zero-Mem: Zero-Token Memory Operations for LLM Agents
LLM agents need memory to act consistently over long interactions, yet many systems use additional LLM calls to operate that memory. Generating intermediate records and mediating their retrieval adds recurring token and time costs, while omitted or merged details can obscure the original evidence. We ask whether struct...
Yilin Xiao; Zhehan Zhu; Yujing Zhang; Jin Chen; Zijin Hong; Luyao Zhuang; Qinggang Zhang; Shengyuan Chen; Xiaocao Ouyang; Lingfei Ren; Xiao Huang
2026-07-31
2026
arXiv
arXiv
cs.CL
arxiv-api
2607.29377
2026-09-08T16:27:48
ale-0675
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Reusing Past Repairs Through Hierarchical Trajectory Abstraction for Coding Agents
https://arxiv.org/abs/2607.29658
external
arxiv.org
Repair agents treat every issue independently and throw away the procedural knowledge from previous repairs.
Repair agents treat every issue independently and throw away the procedural knowledge from previous repairs.
Repair agents treat every issue independently and throw away the procedural knowledge from previous repairs.
Makes persistence and context management visible as runtime design choices. Repair agents treat every issue independently and throw away the procedural knowledge from previous repairs.
Use Reusing Past Repairs Through Hierarchical Trajectory Abstraction for Coding Agents to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.29658; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,365
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1365
2026-08-05
Persist
persist
Carry context, state, and receipts across runs.
intake
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.29658
[2607.29658] Reusing Past Repairs Through Hierarchical Trajectory Abstraction for Coding Agents
Although LLM-driven repair agents can tackle complex, repository-level issues, they treat every issue independently and discard the procedural knowledge accumulated from previous repairs. We introduce STAIR, a framework that converts historical repair trajectories into hierarchical, reusable plans that can be adapted t...
Yisen Xu; Jiayuan Zhou; Ruiqi Pan; Tse-Hsun Chen
2026-07-31
2026
arXiv
arXiv
10 pages
cs.SE
arxiv-api
2607.29658
2026-09-08T16:27:48
ale-0676
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
MemHarness: Memory Is Reconstructed, Not Replayed
https://arxiv.org/abs/2607.28272
external
arxiv.org
Argues that treating retrieved experience as a static record to replay causes negative transfer when past runs do not match present state, and trains a unified policy (GRPO) that reconstructs memory conditioned on the current state instead.
Argues that treating retrieved experience as a static record to replay causes negative transfer when past runs do not match present state, and trains a unified policy (GRPO) that reconstructs memory conditioned on the current state instead.
Argues that treating retrieved experience as a static record to replay causes negative transfer when past runs do not match present state, and trains a unified policy (GRPO) that reconstructs memory conditioned on the current state instead.
Durable execution and replay are treated as first-class loop infrastructure. Argues that treating retrieved experience as a static record to replay causes negative transfer when past runs do not match present state, and trains a unified policy (GRPO) that reconstructs memory conditioned on the current state instead.
Use MemHarness: Memory Is Reconstructed, Not Replayed to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.28272; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,366
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1366
2026-08-05
Persist
persist
Carry context, state, and receipts across runs.
context;state
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.28272
[2607.28272] MemHarness: Memory Is Reconstructed, Not Replayed
Retrieving past experiences has become a common strategy to enhance large language model agents. However, most existing memory-augmented agents treat retrieved experiences as static records to be replayed verbatim, injecting them into the context regardless of whether they align with the agent's current situation. This...
Rong Wu; Daocheng Fu; Licheng Wen; Xuemeng Yang; Shu Zou; Jianbiao Mei; Yuxin Wang; Hairong Zhang; Yu Yang; Tao Hu; Cong Zhang; Botian Shi; Pinlong Cai
2026-07-30
2026
arXiv
arXiv
20 pages, 13 figures
cs.AI
arxiv-api
2607.28272
2026-09-08T16:27:48
ale-0677
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
ChronoMem: Version Control and Semantic Rollback for Large Language Model Agent Memory
https://arxiv.org/abs/2607.27773
external
arxiv.org
Commits a memory snapshot on every write and supports natural-language rollback requests via hybrid retrieval, so an agent can inspect and revert prior memory states rather than living with whatever it last wrote. Claims to be the first open-source system plus benchmark for systematic semantic global memory rollback.
Commits a memory snapshot on every write and supports natural-language rollback requests via hybrid retrieval, so an agent can inspect and revert prior memory states rather than living with whatever it last wrote. Claims to be the first open-source system plus benchmark for systematic semantic global memory rollback.
Commits a memory snapshot on every write and supports natural-language rollback requests via hybrid retrieval, so an agent can inspect and revert prior memory states rather than living with whatever it last wrote. Claims to be the first open-source system plus benchmark for systematic semantic global memory rollback.
The work turns loop quality into a measurable task or score. Commits a memory snapshot on every write and supports natural-language rollback requests via hybrid retrieval, so an agent can inspect and revert prior memory states rather than living with whatever it last wrote. Claims to be the first open-source system plu...
Use ChronoMem: Version Control and Semantic Rollback for Large Language Model Agent Memory to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.27773; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,367
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1367
2026-08-05
Persist
persist
Carry context, state, and receipts across runs.
context;verification
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.27773
[2607.27773] ChronoMem: Version Control and Semantic Rollback for Large Language Model Agent Memory
LLM agents increasingly rely on long-term memory to support multi-session interaction and personalization. However, existing agent memory systems are designed around forward-only evolution, continuously accumulating, consolidating, and overwriting knowledge, with no principled mechanism to inspect, version, or revert p...
Yongye Su; Wujiang Xu; Chaoji Zuo; Elisa Bertino
2026-07-30
2026
arXiv
arXiv
cs.CL
arxiv-api
2607.27773
2026-09-08T16:27:48
ale-0678
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Σ-Mem: An Online Reliability Memory for LLM-based Multi-Agent Systems
https://arxiv.org/abs/2607.27958
external
arxiv.org
Maintains symmetric matrices of per-peer competence and peer-relationship evidence, updated online from correctness feedback with spectral bounds for stability, then reuses that memory for residual steering, response-free peer routing, or reliability-weighted voting.
Maintains symmetric matrices of per-peer competence and peer-relationship evidence, updated online from correctness feedback with spectral bounds for stability, then reuses that memory for residual steering, response-free peer routing, or reliability-weighted voting.
Maintains symmetric matrices of per-peer competence and peer-relationship evidence, updated online from correctness feedback with spectral bounds for stability, then reuses that memory for residual steering, response-free peer routing, or reliability-weighted voting.
Persistent memory is treated as an external runtime artifact. Maintains symmetric matrices of per-peer competence and peer-relationship evidence, updated online from correctness feedback with spectral bounds for stability, then reuses that memory for residual steering, response-free peer routing, or reliability-weighte...
Use Σ-Mem: An Online Reliability Memory for LLM-based Multi-Agent Systems to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.27958; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,368
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1368
2026-08-05
Persist
persist
Carry context, state, and receipts across runs.
context;delegation
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.27958
[2607.27958] $Σ$-Mem: An Online Reliability Memory for LLM-based Multi-Agent Systems
Memory is central to long-horizon LLM agents, yet existing memory systems primarily preserve interaction content rather than modeling which agents can be trusted and under what conditions. This limitation is particularly important in multi-agent systems, where a central model may be unable to directly verify plausible ...
Peilin Feng; Suorong Yang; Soujanya Poria
2026-07-30
2026
arXiv
arXiv
cs.MA
arxiv-api
2607.27958
2026-09-08T16:27:48
ale-0679
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
SkillRise: Agentic Reinforcement Learning for Cross-Task Skill Evolution
https://arxiv.org/abs/2607.26784
external
arxiv.org
Treats a persistent, evolving skill document as the agent's carried state across a sequence of tasks, rather than as a retrieval index consulted per task.
Treats a persistent, evolving skill document as the agent's carried state across a sequence of tasks, rather than as a retrieval index consulted per task.
Treats a persistent, evolving skill document as the agent's carried state across a sequence of tasks, rather than as a retrieval index consulted per task.
State persistence is explicit enough for repeated runs and handoff. Treats a persistent, evolving skill document as the agent's carried state across a sequence of tasks, rather than as a retrieval index consulted per task.
Use SkillRise: Agentic Reinforcement Learning for Cross-Task Skill Evolution to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.26784; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,369
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1369
2026-08-07
Persist
persist
Carry context, state, and receipts across runs.
context;state
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.26784
[2607.26784] SkillRise: Agentic Reinforcement Learning for Cross-Task Skill Evolution
Large language model agents often encounter related yet distinct tasks that share reusable solution patterns. Yet standard agentic reinforcement learning treats tasks as independent episodes, while existing approaches to skill learning either focus on repeated attempts of one task or use pipelines with multiple stages ...
Zhiyuan Yao; Yuxin Chen; Zhengxi Lu; Zishan Xu; Yueqing Sun; Yifu Guo; Yuquan Lu; Zhengzhou Cai; Kangning Zhang; Zhuowen Han; Zi-Han Wang; Ziang Ye; Qi Gu; Xunliang Cai; Weiwen Liu; Yongliang Shen
2026-07-29
2026
arXiv
arXiv
cs.LG
arxiv-api
2607.26784
2026-09-08T16:27:48
ale-0680
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
A-TMA: Decoupling State-Aware Memory Failures in Long-Term Agent Memory
https://arxiv.org/abs/2607.01935
external
arxiv.org
Names ghost memory: superseded, current, and transition facts coexisting in the bank and being retrieved together, so the answer model cannot tell what is true now. Keeps superseded records but labels them, and argues bank, retrieval, and answer-time failures need decoupled evaluation.
Names ghost memory: superseded, current, and transition facts coexisting in the bank and being retrieved together, so the answer model cannot tell what is true now. Keeps superseded records but labels them, and argues bank, retrieval, and answer-time failures need decoupled evaluation.
Names ghost memory: superseded, current, and transition facts coexisting in the bank and being retrieved together, so the answer model cannot tell what is true now. Keeps superseded records but labels them, and argues bank, retrieval, and answer-time failures need decoupled evaluation.
Evaluation data is used as the feedback signal for improving loop behavior. Names ghost memory: superseded, current, and transition facts coexisting in the bank and being retrieved together, so the answer model cannot tell what is true now. Keeps superseded records but labels them, and argues bank, retrieval, and answe...
Use A-TMA: Decoupling State-Aware Memory Failures in Long-Term Agent Memory to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.01935; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,370
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1370
2026-08-10
Persist
persist
Carry context, state, and receipts across runs.
context;verification;state
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.01935
[2607.01935] A-TMA: Decoupling State-Aware Memory Failures in Long-Term Agent Memory
Long term memory lets LLM agents act as persistent assistants, but user facts change. A useful memory system must know what is true now, what used to be true, and what changed. We study \emph{ghost memory}, a state coordination failure in which old, current, and transition facts coexist in the memory bank, remain mixed...
Zitong Shi; Yixuan Tang; Anthony Kum Hoe Tung
2026-07-02
2026
arXiv
arXiv
cs.AI
arxiv-api
2607.01935
2026-09-08T16:27:48
ale-0681
State, Memory, And Context Persistence
state-memory-and-context-persistence
Benchmark
🧪
Personalize-then-Store: Benchmarking and Learning Personalized Memory for Long-horizon Agents
https://arxiv.org/abs/2605.25535
external
arxiv.org
Challenges the assumption that one retention policy suits every user, since a fixed policy spends budget on transient sessions while dropping context long-horizon work needs. Proposes session-level storage gating that skips memory writes for sessions judged transient.
Challenges the assumption that one retention policy suits every user, since a fixed policy spends budget on transient sessions while dropping context long-horizon work needs. Proposes session-level storage gating that skips memory writes for sessions judged transient.
Challenges the assumption that one retention policy suits every user, since a fixed policy spends budget on transient sessions while dropping context long-horizon work needs. Proposes session-level storage gating that skips memory writes for sessions judged transient.
Persistent memory is treated as an external runtime artifact. Challenges the assumption that one retention policy suits every user, since a fixed policy spends budget on transient sessions while dropping context long-horizon work needs. Proposes session-level storage gating that skips memory writes for sessions judged ...
Use Personalize-then-Store: Benchmarking and Learning Personalized Memory for Long-horizon Agents to carry context, state, and receipts across runs and failures.
Benchmark or leaderboard source with repeatable tasks or scores; compare systems only after checking setup and scope.
high
README.md
1,371
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1371
2026-08-10
Persist
persist
Carry context, state, and receipts across runs.
context;budget
researcher;evaluator
harness
enabling
benchmark
A
ok
https://arxiv.org/abs/2605.25535
[2605.25535] Personalize-then-Store: Benchmarking and Learning Personalized Memory for Long-horizon Agents
Existing large language model (LLM) based memory systems apply universal, static policies that overlook a fundamental reality: the contexts that are worth storing in memory are different across users. This misalignment wastes limited memory budget on transient interactions while failing to preserve critical context for...
Yeonjun In; Wonjoong Kim; Sangwu Park; Kanghoon Yoon; Chanyoung Park
2026-05-25
2026
arXiv
arXiv
preprint
cs.AI
arxiv-api
2605.25535
2026-09-08T16:27:48
ale-0682
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
SmoothAgent: Efficient Long-Horizon LLM-Based Agent Serving with Lookahead Context Engineering
https://arxiv.org/abs/2607.00151
external
arxiv.org
Prices the context strategies agent frameworks rely on. Offloading, reduction, and isolation each invalidate the KV cache and force a re-prefill, raising time-to-first-token. Observes that these transformations are segment-decomposable, so a prefix can be transformed ahead of time rather than on the critical path.
Prices the context strategies agent frameworks rely on. Offloading, reduction, and isolation each invalidate the KV cache and force a re-prefill, raising time-to-first-token. Observes that these transformations are segment-decomposable, so a prefix can be transformed ahead of time rather than on the critical path.
Prices the context strategies agent frameworks rely on. Offloading, reduction, and isolation each invalidate the KV cache and force a re-prefill, raising time-to-first-token. Observes that these transformations are segment-decomposable, so a prefix can be transformed ahead of time rather than on the critical path.
Context is managed as durable loop state rather than a single prompt payload. Prices the context strategies agent frameworks rely on. Offloading, reduction, and isolation each invalidate the KV cache and force a re-prefill, raising time-to-first-token. Observes that these transformations are segment-decomposable, so a ...
Use SmoothAgent: Efficient Long-Horizon LLM-Based Agent Serving with Lookahead Context Engineering to carry context, state, and receipts across runs and failures.
Research source arXiv:2607.00151; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,372
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1372
2026-08-13
Persist
persist
Carry context, state, and receipts across runs.
context;budget
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.00151
[2607.00151] SmoothAgent: Efficient Long-Horizon LLM-Based Agent Serving with Lookahead Context Engineering
LLM-based agents execute multi-turn workflows with continuously growing contexts, where LLM calls are interleaved with tool invocations and environment feedback. To maintain model quality, modern agent frameworks rely on context engineering strategies such as offloading, reduction, and isolation to control the context ...
Zaifeng Pan; Qianxu Wang; Zhengding Hu; Chang Chen; Yue Guan; Yanbo Zhou; Steven Swanson; Yufei Ding
2026-06-30
2026
arXiv
arXiv
cs.DC
arxiv-api
2607.00151
2026-09-08T16:27:48
ale-0683
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Everything is Context: Agentic File System Abstraction for Context Engineering
https://arxiv.org/abs/2512.05470
external
arxiv.org
Argues prompt engineering, RAG, and tool integration stay fragmented and leave transient artefacts that cannot be traced or audited. Applies the Unix everything-is-a-file idea to context, giving memory, tools, knowledge, and human input one persistent mount-based interface with governance attached.
Argues prompt engineering, RAG, and tool integration stay fragmented and leave transient artefacts that cannot be traced or audited. Applies the Unix everything-is-a-file idea to context, giving memory, tools, knowledge, and human input one persistent mount-based interface with governance attached.
Argues prompt engineering, RAG, and tool integration stay fragmented and leave transient artefacts that cannot be traced or audited. Applies the Unix everything-is-a-file idea to context, giving memory, tools, knowledge, and human input one persistent mount-based interface with governance attached.
Persistent memory is treated as an external runtime artifact. Argues prompt engineering, RAG, and tool integration stay fragmented and leave transient artefacts that cannot be traced or audited. Applies the Unix everything-is-a-file idea to context, giving memory, tools, knowledge, and human input one persistent mount-...
Use Everything is Context: Agentic File System Abstraction for Context Engineering to carry context, state, and receipts across runs and failures.
Research source arXiv:2512.05470; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,373
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1373
2026-08-13
Persist
persist
Carry context, state, and receipts across runs.
workspace;context;state;escalation
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2512.05470
[2512.05470] Everything is Context: Agentic File System Abstraction for Context Engineering
Generative AI (GenAI) has reshaped software system design by introducing foundation models as pre-trained subsystems that redefine architectures and operations. The emerging challenge is no longer model fine-tuning but context engineering-how systems capture, structure, and govern external knowledge, memory, tools, and...
Xiwei Xu; Robert Mao; Quan Bai; Xuewu Gu; Yechao Li; Liming Zhu
2025-12-05
2025
arXiv
arXiv
Submitted
cs.SE
arxiv-api
2512.05470
2026-09-08T16:27:48
ale-0684
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
AgentProg: Empowering Long-Horizon GUI Agents with Program-Guided Context Management
https://arxiv.org/abs/2512.10371
external
arxiv.org
Reframes interaction history as a program with variables and control flow, so what to retain and what to discard follows program structure instead of recency or similarity. Adds a global belief state for partial observability, aimed at the context overhead that grows with every step of a long GUI task.
Reframes interaction history as a program with variables and control flow, so what to retain and what to discard follows program structure instead of recency or similarity. Adds a global belief state for partial observability, aimed at the context overhead that grows with every step of a long GUI task.
Reframes interaction history as a program with variables and control flow, so what to retain and what to discard follows program structure instead of recency or similarity. Adds a global belief state for partial observability, aimed at the context overhead that grows with every step of a long GUI task.
Context is managed as durable loop state rather than a single prompt payload. Reframes interaction history as a program with variables and control flow, so what to retain and what to discard follows program structure instead of recency or similarity. Adds a global belief state for partial observability, aimed at the co...
Use AgentProg: Empowering Long-Horizon GUI Agents with Program-Guided Context Management to carry context, state, and receipts across runs and failures.
Research source arXiv:2512.10371; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,374
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1374
2026-08-17
Persist
persist
Carry context, state, and receipts across runs.
context;state
researcher;evaluator
harness
enabling
research-paper
A
ok
https://doi.org/10.1145/3745756.3809245
[2512.10371] AgentProg: Empowering Long-Horizon GUI Agents with Program-Guided Context Management
The rapid development of mobile GUI agents has stimulated growing research interest in long-horizon task automation. However, building agents for these tasks faces a critical bottleneck: the reliance on ever-expanding interaction history incurs substantial context overhead. Existing context management and compression t...
Shizuo Tian; Hao Wen; Yuxuan Chen; Jiacheng Liu; Shanhui Zhao; Guohong Liu; Ju Ren; Yunxin Liu; Yuanchun Li
2026-06-21
2026
MobiSys '26: 24th Annual International Conference on Mobile Systems, Applications and Services 2026
ACM
10.1145/3745756.3809245
Published in MobiSys '26: 24th Annual International Conference on Mobile Systems, Applications and Services 2026; the linked arXiv record remains available for open access.
cs.AI
DBLP API + primary publication record
2512.10371
2026-09-08T16:27:48
ale-0685
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Context Compaction Theory
https://arxiv.org/abs/2608.01326
external
arxiv.org
First formal analysis of a process every long-running agent already runs. Models compaction as two games, one selecting a subset of accumulated state to retain and one generating a replacement, and studies the compounding case that matters in practice: later compactions operate on states that already contain earlier su...
First formal analysis of a process every long-running agent already runs. Models compaction as two games, one selecting a subset of accumulated state to retain and one generating a replacement, and studies the compounding case that matters in practice: later compactions operate on states that already contain earlier su...
First formal analysis of a process every long-running agent already runs. Models compaction as two games, one selecting a subset of accumulated state to retain and one generating a replacement, and studies the compounding case that matters in practice: later compactions operate on states that already contain earlier su...
Context is managed as durable loop state rather than a single prompt payload. First formal analysis of a process every long-running agent already runs. Models compaction as two games, one selecting a subset of accumulated state to retain and one generating a replacement, and studies the compounding case that matters in...
Use Context Compaction Theory to carry context, state, and receipts across runs and failures.
Research source arXiv:2608.01326; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,375
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1375
2026-08-28
Persist
persist
Carry context, state, and receipts across runs.
context;state
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2608.01326
[2608.01326] Context Compaction Theory
Large Language Models (LLMs) have a bounded context window. The context window is the maximum input size an LLM can consume for a single inference. AI agents rely on a process called context compaction to fit their state within the context window when calling an LLM. Despite its ubiquity, context compaction has receive...
Hayder Tirmazi; Sam Markelon; Allison Bishop; Michael Mitzenmacher
2026-08-02
2026
arXiv
arXiv
21 pages, 2 figures, Preliminary version
cs.DS
arxiv-api
2608.01326
2026-09-08T16:27:48
ale-0686
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Beyond Compaction: Structured Context Eviction for Long-Horizon Agents
https://arxiv.org/abs/2606.11213
external
arxiv.org
Annotates the trajectory as typed, dependency-linked episodes, then evicts in priority order using a deterministic policy that calls no model. Preserves user turns and the context under active reasoning while shedding action episodes whose effects are already persisted in the environment, holding active context near a ...
Annotates the trajectory as typed, dependency-linked episodes, then evicts in priority order using a deterministic policy that calls no model. Preserves user turns and the context under active reasoning while shedding action episodes whose effects are already persisted in the environment, holding active context near a ...
Annotates the trajectory as typed, dependency-linked episodes, then evicts in priority order using a deterministic policy that calls no model. Preserves user turns and the context under active reasoning while shedding action episodes whose effects are already persisted in the environment, holding active context near a ...
Context is managed as durable loop state rather than a single prompt payload. Annotates the trajectory as typed, dependency-linked episodes, then evicts in priority order using a deterministic policy that calls no model. Preserves user turns and the context under active reasoning while shedding action episodes whose ef...
Use Beyond Compaction: Structured Context Eviction for Long-Horizon Agents to carry context, state, and receipts across runs and failures.
Research source arXiv:2606.11213; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,376
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1376
2026-08-28
Persist
persist
Carry context, state, and receipts across runs.
context;state
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2606.11213
[2606.11213] Beyond Compaction: Structured Context Eviction for Long-Horizon Agents
We present Context Window Lifecycle (CWL), a context-management scheme that gives long-horizon LLM agents an effectively unbounded working horizon. As a session accumulates history, CWL keeps the context within budget through graduated, semantically-aware eviction: the agent annotates its trajectory as typed, dependenc...
Andrew Semenov; Svyatoslav Dorofeev
2026-05-01
2026
arXiv
arXiv
cs.CL
arxiv-api
2606.11213
2026-09-08T16:27:48
ale-0687
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Parallel Context Compaction for Long-Horizon LLM Agent Serving
https://arxiv.org/abs/2605.23296
external
arxiv.org
Prices summarization-based compaction as an operational cost rather than a design detail: the blocking call stalls agent inference for tens of seconds, prompt instructions barely control summary volume, and retained information fluctuates run to run. Proposes running compaction alongside the agent loop instead of in it...
Prices summarization-based compaction as an operational cost rather than a design detail: the blocking call stalls agent inference for tens of seconds, prompt instructions barely control summary volume, and retained information fluctuates run to run. Proposes running compaction alongside the agent loop instead of in it...
Prices summarization-based compaction as an operational cost rather than a design detail: the blocking call stalls agent inference for tens of seconds, prompt instructions barely control summary volume, and retained information fluctuates run to run. Proposes running compaction alongside the agent loop instead of in it...
Context is managed as durable loop state rather than a single prompt payload. Prices summarization-based compaction as an operational cost rather than a design detail: the blocking call stalls agent inference for tens of seconds, prompt instructions barely control summary volume, and retained information fluctuates run...
Use Parallel Context Compaction for Long-Horizon LLM Agent Serving to carry context, state, and receipts across runs and failures.
Research source arXiv:2605.23296; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,377
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1377
2026-08-28
Persist
persist
Carry context, state, and receipts across runs.
context;budget
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2605.23296
[2605.23296] Parallel Context Compaction for Long-Horizon LLM Agent Serving
Long-horizon LLM agents accumulate growing conversation histories that eventually exceed the model's context window. Context compaction via LLM-based summarization keeps the conversation bounded, but summarization is inherently lossy and the blocking call stalls agent inference for tens of seconds. Moreover, the operat...
Musa Cim; Burak Topcu; Chita Das; Mahmut Taylan Kandemir
2026-05-22
2026
arXiv
arXiv
cs.AI
arxiv-api
2605.23296
2026-09-08T16:27:48
ale-0688
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Agent Memory: Characterization and System Implications of Stateful Long-Horizon Workloads
https://arxiv.org/abs/2606.06448
external
arxiv.org
A systems characterization of agent memory rather than another memory design. Classifies ten representative systems on four axes and profiles them with a phase-aware harness that attributes cost separately to construction, retrieval, and generation, which is where memory overhead usually hides.
A systems characterization of agent memory rather than another memory design. Classifies ten representative systems on four axes and profiles them with a phase-aware harness that attributes cost separately to construction, retrieval, and generation, which is where memory overhead usually hides.
A systems characterization of agent memory rather than another memory design. Classifies ten representative systems on four axes and profiles them with a phase-aware harness that attributes cost separately to construction, retrieval, and generation, which is where memory overhead usually hides.
Persistent memory is treated as an external runtime artifact. A systems characterization of agent memory rather than another memory design. Classifies ten representative systems on four axes and profiles them with a phase-aware harness that attributes cost separately to construction, retrieval, and generation, which is...
Use Agent Memory: Characterization and System Implications of Stateful Long-Horizon Workloads to carry context, state, and receipts across runs and failures.
Research source arXiv:2606.06448; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,378
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1378
2026-09-04
Persist
persist
Carry context, state, and receipts across runs.
context;state;budget
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2606.06448
[2606.06448] Agent Memory: Characterization and System Implications of Stateful Long-Horizon Workloads
LLM agents are increasingly deployed on long-horizon tasks requiring sustained reasoning over extended interaction histories. Realizing this at scale requires agents to persistently store, retrieve, and update their own memory across sessions. A rich ecosystem of agent memory systems has emerged spanning flat retrieval...
Yasmine Omri; Ziyu Gan; Zachary Broveak; Robin Geens; Zexue He; Alex Pentland; Marian Verhelst; Tsachy Weissman; Thierry Tambe
2026-06-04
2026
arXiv
arXiv
cs.AI
arxiv-api
2606.06448
2026-09-08T16:27:48
ale-0689
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
When Memory Becomes Authority: Benchmarking Authority Collapse at the Memory Consolidation Boundary
https://arxiv.org/abs/2608.01679
external
arxiv.org
Names authority collapse: consolidation keeps a claim but drops the source constraints governing its use, so stored memory later implies more authority than its origin permitted and an offhand remark can return as a standing instruction. The benchmark holds claim and task fixed and varies only source authority.
Names authority collapse: consolidation keeps a claim but drops the source constraints governing its use, so stored memory later implies more authority than its origin permitted and an offhand remark can return as a standing instruction. The benchmark holds claim and task fixed and varies only source authority.
Names authority collapse: consolidation keeps a claim but drops the source constraints governing its use, so stored memory later implies more authority than its origin permitted and an offhand remark can return as a standing instruction. The benchmark holds claim and task fixed and varies only source authority.
The work turns loop quality into a measurable task or score. Names authority collapse: consolidation keeps a claim but drops the source constraints governing its use, so stored memory later implies more authority than its origin permitted and an offhand remark can return as a standing instruction. The benchmark holds c...
Use When Memory Becomes Authority: Benchmarking Authority Collapse at the Memory Consolidation Boundary to carry context, state, and receipts across runs and failures.
Research source arXiv:2608.01679; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,379
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1379
2026-09-04
Persist
persist
Carry context, state, and receipts across runs.
context;verification
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2608.01679
[2608.01679] When Memory Becomes Authority: Benchmarking Authority Collapse at the Memory Consolidation Boundary
Persistent memory allows (self-evolving) LLM agents to adapt across tasks by consolidating heterogeneous interaction histories into reusable facts, preferences, observations, and rules. Yet consolidation also imposes an implicit authorization boundary: it determines whether stored information may later be consumed as a...
Qiuyang Zhan; Rui Zhang; Sheng Guo; Lepeng Zhao; Zhuotao Liu
2026-08-03
2026
arXiv
arXiv
38 pages, 2 figures
cs.AI
arxiv-api
2608.01679
2026-09-08T16:27:48
ale-0690
State, Memory, And Context Persistence
state-memory-and-context-persistence
Paper
📄
Graph-based Agent Memory: Taxonomy, Techniques, and Applications
https://arxiv.org/abs/2602.05665
external
arxiv.org
Survey of agent memory through graph structure, taxonomized on three axes: short-term against long-term, knowledge against experience, and non-structural against structural. Argues graphs fit the job because they carry the relational dependencies and hierarchy that flat retrieval flattens away.
Survey of agent memory through graph structure, taxonomized on three axes: short-term against long-term, knowledge against experience, and non-structural against structural. Argues graphs fit the job because they carry the relational dependencies and hierarchy that flat retrieval flattens away.
Survey of agent memory through graph structure, taxonomized on three axes: short-term against long-term, knowledge against experience, and non-structural against structural. Argues graphs fit the job because they carry the relational dependencies and hierarchy that flat retrieval flattens away.
Control flow is represented as an inspectable graph rather than an opaque prompt loop. Survey of agent memory through graph structure, taxonomized on three axes: short-term against long-term, knowledge against experience, and non-structural against structural. Argues graphs fit the job because they carry the relational...
Use Graph-based Agent Memory: Taxonomy, Techniques, and Applications to carry context, state, and receipts across runs and failures.
Research source arXiv:2602.05665; inspect its method and evaluation before treating results as production evidence.
medium
README.md
1,380
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1380
2026-09-04
Persist
persist
Carry context, state, and receipts across runs.
context
researcher;evaluator
harness
enabling
research-preprint
A
ok
https://arxiv.org/abs/2602.05665
[2602.05665] Graph-based Agent Memory: Taxonomy, Techniques, and Applications
Memory emerges as the core module in the Large Language Model (LLM)-based agents for long-horizon complex tasks (e.g., multi-turn dialogue, game playing, scientific discovery), where memory can enable knowledge accumulation, iterative reasoning and self-evolution. Among diverse paradigms, graph stands out as a powerful...
Chang Yang; Chuang Zhou; Yilin Xiao; Su Dong; Luyao Zhuang; Yujing Zhang; Zhu Wang; Zijin Hong; Zheng Yuan; Zhishang Xiang; Shengyuan Chen; Huachi Zhou; Qinggang Zhang; Ninghao Liu; Jinsong Su; Xinrun Wang; Yi Chang; Xiao Huang
2026-02-05
2026
arXiv
arXiv
cs.AI
arxiv-api
2602.05665
2026-09-08T16:27:48
ale-0691
Orchestration And Multi-Agent Delegation
orchestration-and-multi-agent-delegation
Tool
🧰
AutoGen
https://github.com/microsoft/autogen
external
github.com
Multi-agent programming framework for conversations, tool use, and orchestration; active development has moved to the Microsoft Agent Framework.
Multi-agent programming framework for conversations, tool use, and orchestration; active development has moved to the Microsoft Agent Framework.
Multi-agent programming framework for conversations, tool use, and orchestration; active development has moved to the Microsoft Agent Framework.
The work separates roles across agents, verifiers, or orchestration layers. Multi-agent programming framework for conversations, tool use, and orchestration; active development has moved to the Microsoft Agent Framework.
Use AutoGen to choose an implementation surface for repeatable agent work.
Inspectable GitHub source; popularity is context, not proof of reliability.
medium
README.md
1,388
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1388
Build
build
Choose runtimes, tools, and delegation surfaces.
workspace;delegation
builder
workflow
enabling
source-implementation
A
ok
https://github.com/microsoft/autogen
GitHub - microsoft/autogen: A programming framework for agentic AI · GitHub
A programming framework for agentic AI. Contribute to microsoft/autogen development by creating an account on GitHub.
microsoft/autogen
GitHub
html-meta
microsoft/autogen
2026-09-08T16:27:48
ale-0692
Orchestration And Multi-Agent Delegation
orchestration-and-multi-agent-delegation
Tool
🧰
Microsoft Agent Framework
https://github.com/microsoft/agent-framework
external
github.com
Microsoft's successor to AutoGen and Semantic Kernel for building and orchestrating multi-agent workflows in Python and .NET.
Microsoft's successor to AutoGen and Semantic Kernel for building and orchestrating multi-agent workflows in Python and .NET.
Microsoft's successor to AutoGen and Semantic Kernel for building and orchestrating multi-agent workflows in Python and .NET.
The work separates roles across agents, verifiers, or orchestration layers. Microsoft's successor to AutoGen and Semantic Kernel for building and orchestrating multi-agent workflows in Python and .NET.
Use Microsoft Agent Framework to choose an implementation surface for repeatable agent work.
Inspectable GitHub source; popularity is context, not proof of reliability.
medium
README.md
1,389
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1389
Build
build
Choose runtimes, tools, and delegation surfaces.
delegation
builder
workflow
enabling
source-implementation
A
ok
https://github.com/microsoft/agent-framework
GitHub - microsoft/agent-framework: A framework for building, orchestrating and deploying AI agents and multi-agent workflows with support for Python and .NET. · GitHub
A framework for building, orchestrating and deploying AI agents and multi-agent workflows with support for Python and .NET. - microsoft/agent-framework
microsoft/agent-framework
GitHub
html-meta
microsoft/agent-framework
2026-09-08T16:27:48
ale-0693
Orchestration And Multi-Agent Delegation
orchestration-and-multi-agent-delegation
Tool
🧰
LangGraph
https://github.com/langchain-ai/langgraph
external
github.com
Graph-based framework for controllable agent workflows, persistence, and human-in-the-loop steps.
Graph-based framework for controllable agent workflows, persistence, and human-in-the-loop steps.
Graph-based framework for controllable agent workflows, persistence, and human-in-the-loop steps.
Control flow is represented as an inspectable graph rather than an opaque prompt loop. Graph-based framework for controllable agent workflows, persistence, and human-in-the-loop steps.
Use LangGraph to choose an implementation surface for repeatable agent work.
Inspectable GitHub source; popularity is context, not proof of reliability.
medium
README.md
1,390
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1390
Build
build
Choose runtimes, tools, and delegation surfaces.
state;escalation
builder
workflow
enabling
source-implementation
A
ok
https://github.com/langchain-ai/langgraph
GitHub - langchain-ai/langgraph: Build resilient agents. · GitHub
Build resilient agents. Contribute to langchain-ai/langgraph development by creating an account on GitHub.
langchain-ai/langgraph
GitHub
html-meta
langchain-ai/langgraph
2026-09-08T16:27:48
ale-0694
Orchestration And Multi-Agent Delegation
orchestration-and-multi-agent-delegation
Tool
🧰
CrewAI
https://github.com/crewAIInc/crewAI
external
github.com
Framework for multi-agent workflows organized around roles, tasks, and crews.
Framework for multi-agent workflows organized around roles, tasks, and crews.
Framework for multi-agent workflows organized around roles, tasks, and crews.
The work separates roles across agents, verifiers, or orchestration layers. Framework for multi-agent workflows organized around roles, tasks, and crews.
Use CrewAI to choose an implementation surface for repeatable agent work.
Inspectable GitHub source; popularity is context, not proof of reliability.
medium
README.md
1,391
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1391
Build
build
Choose runtimes, tools, and delegation surfaces.
delegation
builder
workflow
enabling
source-implementation
A
ok
https://github.com/crewAIInc/crewAI
GitHub - crewAIInc/crewAI: Framework for orchestrating role-playing, autonomous AI agents. By fostering collaborative intelligence, CrewAI empowers agents to work together seamlessly, tackling complex tasks. · GitHub
Framework for orchestrating role-playing, autonomous AI agents. By fostering collaborative intelligence, CrewAI empowers agents to work together seamlessly, tackling complex tasks. - crewAIInc/crewAI
crewAIInc/crewAI
GitHub
html-meta
crewAIInc/crewAI
2026-09-08T16:27:48
ale-0695
Orchestration And Multi-Agent Delegation
orchestration-and-multi-agent-delegation
Docs
📚
LlamaIndex Workflows
https://developers.llamaindex.ai/python/llamaagents/workflows/
external
developers.llamaindex.ai
Event-driven workflow abstraction for agentic applications.
Event-driven workflow abstraction for agentic applications.
Event-driven workflow abstraction for agentic applications.
Shows how delegation, handoff, and workflow control turn one agent into a coordinated loop. Event-driven workflow abstraction for agentic applications.
Use LlamaIndex Workflows to choose an implementation surface for repeatable agent work.
Primary documentation from a platform, SDK, standard, or framework; strong implementation signal.
high
README.md
1,392
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1392
Build
build
Choose runtimes, tools, and delegation surfaces.
trigger
builder
workflow
enabling
technical-documentation
A
ok
https://developers.llamaindex.ai/python/llamaagents/workflows/
Introduction | Developer Documentation
Developer Documentation
html-meta
2026-09-08T16:27:48
ale-0696
Orchestration And Multi-Agent Delegation
orchestration-and-multi-agent-delegation
Docs
📚
OpenAI Agents SDK handoffs
https://openai.github.io/openai-agents-python/handoffs/
external
openai.github.io
First-class delegation between specialized agents.
First-class delegation between specialized agents.
First-class delegation between specialized agents.
Shows how delegation, handoff, and workflow control turn one agent into a coordinated loop. First-class delegation between specialized agents.
Use OpenAI Agents SDK handoffs to choose an implementation surface for repeatable agent work.
Primary official documentation from openai.github.io; use it for current product or standard behavior.
high
README.md
1,393
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1393
Build
build
Choose runtimes, tools, and delegation surfaces.
delegation
builder
workflow
enabling
official-documentation
A
ok
https://openai.github.io/openai-agents-python/handoffs/
Handoffs - OpenAI Agents SDK
openai.github.io
domain-fallback
2026-09-08T16:27:48
ale-0697
Orchestration And Multi-Agent Delegation
orchestration-and-multi-agent-delegation
Docs
📚
Agent Protocol
https://agentprotocol.ai/
external
agentprotocol.ai
API protocol for agent interaction, useful for separating loop managers from agent runtimes.
API protocol for agent interaction, useful for separating loop managers from agent runtimes.
API protocol for agent interaction, useful for separating loop managers from agent runtimes.
Shows how delegation, handoff, and workflow control turn one agent into a coordinated loop. API protocol for agent interaction, useful for separating loop managers from agent runtimes.
Use Agent Protocol to choose an implementation surface for repeatable agent work.
Primary documentation from a platform, SDK, standard, or framework; strong implementation signal.
high
README.md
1,394
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1394
Build
build
Choose runtimes, tools, and delegation surfaces.
delegation;state
builder
workflow
enabling
technical-documentation
A
ok
https://agentprotocol.ai/
AgentProtocol.ai — AI agent protocols explained
AgentProtocol.ai is an independent, vendor-neutral guide to AI agent communication standards — MCP, A2A, Agent Protocol and interoperability.
AgentProtocol.ai
AgentProtocol.ai
html-meta
2026-09-08T16:27:48
ale-0698
Orchestration And Multi-Agent Delegation
orchestration-and-multi-agent-delegation
Tool
🧰
AgentKit
https://github.com/inngest/agent-kit
external
github.com
TypeScript toolkit for durable, event-driven agents on workflow infrastructure.
TypeScript toolkit for durable, event-driven agents on workflow infrastructure.
TypeScript toolkit for durable, event-driven agents on workflow infrastructure.
Durable execution and replay are treated as first-class loop infrastructure. TypeScript toolkit for durable, event-driven agents on workflow infrastructure.
Use AgentKit to choose an implementation surface for repeatable agent work.
Inspectable GitHub source; popularity is context, not proof of reliability.
medium
README.md
1,395
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1395
Build
build
Choose runtimes, tools, and delegation surfaces.
trigger
builder
workflow
enabling
source-implementation
A
ok
https://github.com/inngest/agent-kit
GitHub - inngest/agent-kit: AgentKit: Build multi-agent networks in TypeScript with deterministic routing and rich tooling via MCP. · GitHub
AgentKit: Build multi-agent networks in TypeScript with deterministic routing and rich tooling via MCP. - inngest/agent-kit
inngest/agent-kit
GitHub
html-meta
inngest/agent-kit
2026-09-08T16:27:48
ale-0699
Orchestration And Multi-Agent Delegation
orchestration-and-multi-agent-delegation
Tool
🧰
deepagents
https://github.com/langchain-ai/deepagents
external
github.com
LangChain project for deeper, longer-running agents with middleware and harness patterns.
LangChain project for deeper, longer-running agents with middleware and harness patterns.
LangChain project for deeper, longer-running agents with middleware and harness patterns.
Shows how delegation, handoff, and workflow control turn one agent into a coordinated loop. LangChain project for deeper, longer-running agents with middleware and harness patterns.
Use deepagents to choose an implementation surface for repeatable agent work.
Inspectable GitHub source; popularity is context, not proof of reliability.
medium
README.md
1,396
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1396
Build
build
Choose runtimes, tools, and delegation surfaces.
delegation;state
builder
workflow
enabling
source-implementation
A
ok
https://github.com/langchain-ai/deepagents
GitHub - langchain-ai/deepagents: The batteries-included agent harness. · GitHub
The batteries-included agent harness. Contribute to langchain-ai/deepagents development by creating an account on GitHub.
langchain-ai/deepagents
GitHub
html-meta
langchain-ai/deepagents
2026-09-08T16:27:48
ale-0700
Orchestration And Multi-Agent Delegation
orchestration-and-multi-agent-delegation
Docs
📚
Temporal for AI
https://temporal.io/solutions/ai
external
temporal.io
Durable execution for long-running agent workflows: crash-proof state, automatic retries, and human-in-the-loop signals.
Durable execution for long-running agent workflows: crash-proof state, automatic retries, and human-in-the-loop signals.
Durable execution for long-running agent workflows: crash-proof state, automatic retries, and human-in-the-loop signals.
Durable execution and replay are treated as first-class loop infrastructure. Durable execution for long-running agent workflows: crash-proof state, automatic retries, and human-in-the-loop signals.
Use Temporal for AI to choose an implementation surface for repeatable agent work.
Primary documentation from a platform, SDK, standard, or framework; strong implementation signal.
high
README.md
1,397
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1397
Build
build
Choose runtimes, tools, and delegation surfaces.
state;budget;escalation
builder
workflow
enabling
technical-documentation
A
ok
https://temporal.io/solutions/ai
AI Applications & Agents With Temporal | Temporal
Build the most capable AI applications and agents on Temporal's open foundation, powered by durable execution
temporal.io
domain-fallback
2026-09-08T16:27:48