Memory mechanisms in agentic systems
Type: types/note.md
Each row uses one main-review result and its stated memory boundary. Values carry their evidence basis. Absence, inapplicability, uninspected mechanisms, and indeterminate classifications remain distinct. This is the selected population, not the historical memory-review corpus.
code-grounded (28)
| System | Compared boundary | Storage | Read-back | Push selection | Trace learning | Authority | Evidence |
|---|---|---|---|---|---|---|---|
| Apache Maka | Retained runtime traces and their model projections; text and provider-native continuation checkpoints; manual MEMORY.md/PENDING.md entries; automatically extracted SQLite memory items and their access metadata; acquired or edited local skills, preferences and turn inventories; archived tool results and delegated-session reads. Static shipped prompt text and uninstalled bundled skills are contextual evidence only. Operator-only recap output is a checked boundary exclusion. External provider internals and deployment outcomes are unavailable. | files, in-memory, sqlite [wired] | pull, push [wired] | coarse, identifier [wired] | yes [wired] | instruction, knowledge, ranking, routing, validation [wired] | AAS-2026-09-05-apache-maka-06 |
| AREX-Skill | Repository operating skills accumulated or revised through Creator construction, verification, refresh and extension; private construction handoffs/reports with later consumers; router/access metadata and managed deployment state; Researcher and documented exported-agent reads. Static imported library content establishes an admission alternative, not learning through use. Host conversation memory, paper construction and unrelated workflows are excluded. | files [afforded] | pull, push [afforded] | coarse [afforded] | yes [afforded] | enforcement, instruction, knowledge, routing, validation [afforded] | AAS-2026-09-25-arex-skill-01 |
| arsumbris | Accumulated or revised workspace graph content and schemas, derived access structures and launch artifacts, correction-derived guidance, and retained session continuity facts in the 22 pinned release repositories. Ordinary current-run read/served hashes and freshness notices are auxiliary context controls outside this normalized memory scope. Shipped unchanged instructions are delivery machinery, not accumulated memory. External harness/model storage and compaction internals are excluded. | files, graph, in-memory, repo [wired] | pull, push [wired] | coarse, identifier [wired] | yes [afforded] | instruction, knowledge, routing, validation [afforded] | AAS-2026-09-23-arsumbris-02 |
| ContextPilot | Use-accumulated inference and training notes, plans, histories, context replacements, relation/access caches, branch replay state, imported wrong-answer retry records and repository-trained actor parameters/checkpoints. Covers default FSM and base alternatives, conditional embeddings/compression and callable nondefault memory methods. Excludes ordinary incoming corpus/search indexes, static prompts, external SDK internals and externally pretrained weights. | files, graph, in-memory, model-weights, vector [wired] | pull, push [wired] | not-determinable | yes [wired] | enforcement, instruction, knowledge, learning, ranking, routing [wired] | AAS-2026-09-25-contextpilot-01 |
| DualGraph | Named DualGraph subsystem: accumulated within-report outline, evidence, knowledge graph, embeddings/clusters/chains, query/URL history, report prefix/products, optional Neo4j projection and human demo/live display read-back. Excludes static prompts as memory, other DKI_LLM subsystems, external provider internals and benchmark/generated-output execution evidence. Wired denotes inspected code connections, not successful operation; RTE-9 records default-path defects. | files, graph, in-memory [wired] | pull, push [wired] | not-determinable | yes [wired] | enforcement, instruction, knowledge, ranking, routing, validation [wired] | AAS-2026-09-25-dualgraph-01 |
| EAL-bench | Core EAL-bench one-shot and incremental free-text/typed authorization profiles, accepted-state continuation and repair, frozen executor replay, and packaged writer-memory retrieval through the reusable evaluation interface. Includes retained payloads, pre-executor file checkpoints and their operative identity/provenance structures. Static policy, oracle-exact and deliberately altered diagnostic controls are comparison inputs, not acquired memory; extension studies and provider/LangMem internals are excluded. Domain-independent mechanics are covered; detailed semantic and rationale conclusions are limited to procurement. | files, in-memory [wired] | pull, push [afforded] | not-determinable | yes [wired] | enforcement, knowledge, routing, validation [wired] | AAS-2026-09-25-eal-bench-01 |
| Ecdysis | Whole method-level retained-memory boundary: extracted failure evidence, FDCR transcript/specification/checkpoint and continuation, accepted generic HarnessT through training and later inference, decision records, and skill-artifact authoring/storage utilities. Includes opaque harness payloads; excludes internals of external adapters, providers, and task environments. Skill storage has no established task-consumer route. | not-determinable | push [wired] | coarse [wired] | yes [wired] | not-determinable | AAS-2026-09-25-ecdysis-01 |
| Enoch | Core-owned use-modified private memory/identity, journals, retained task and recurring-job briefs, evidence/candidates/curation, imported lineage assessments, learning receipts and governed body/skill revision artifacts at the frozen pin; includes core session-reference interfaces and process-local delivery metadata. Provider-owned opaque session payloads, model weights, installed dependency revisions and actual deployments are outside comparison assurance and remain explicit continuity uncertainties. | files, in-memory, repo [wired] | pull, push [wired] | coarse, identifier, inferred-judgment, inferred-lexical [wired] | yes [wired] | not-determinable | AAS-2026-09-25-enoch-01 |
| EvoOntology | Accumulated or revised project ontology content, workload questions, observable trajectories, project/run/checkpoint/evaluation records, derived causal knowledge, and host-mediated Schema/Tool revisions; their shipped file/in-memory representations, retrieval and reminder routes. Static unrevised plugin instructions are method, not learned memory. External host/provider internals, benchmark task-agent runtimes and experimental reproduction are excluded. | files, in-memory [afforded] | pull, push [afforded] | coarse [wired] | yes [afforded] | enforcement, instruction, knowledge, ranking, routing, validation [afforded] | AAS-2026-09-25-evoontology-01 |
| fragility-grid | Files accumulated or replaced by the grid and shell: result records, model summaries, shared config legend, temporary work queue and per-model logs; their scheduler and CPU loader consumers. Excludes fixed datasets, templates and pretrained weights, transient inference buffers, dependency caches, and downstream statistical artifacts beyond the loader boundary. | files [wired] | pull [wired] | inapplicable | no [wired] | knowledge, routing [wired] | AAS-2026-09-25-fragility-grid-01 |
| GBrain | Retained objects GBrain accumulates or changes through use: brain pages (Markdown files plus DB rows, chunks, embeddings), facts rows and ## Facts fences, fact withdrawals, takes/take proposals/grade cache/calibration profiles, link and timeline edges, session_context_state, the raw transcript corpus (session-end, compaction-segment and writeback-turn files) and Stop live buffer, dream/think-generated pages, the bootstrap agent-workspace memory files (MEMORY.md, memory/ daily logs, USER.md, SOUL.md) as edited by the agent, brain-resident shared skill revisions, SkillOpt versions/best/proposed/history/rejected buffers and the SKILL.md files SkillOpt commits; plus access metadata (last_retrieved_at, context_volunteer_events). Routes: explicit remember/put_page/Markdown sync/workspace file edits, ambient Stop-hook writeback, compaction checkpoint harvest, session-end corpus plus serve sweep extraction, page-write facts backstop, dream-cycle phases, SkillOpt, shared-skill publish; maintenance by dedup, supersession, TTL, forget/withdrawal, decay-ranking, consolidation, purge; read-back through requested recall/search/query/entity/context_pack/delta/think/synthesize/get_skill and automatic hook push, MCP _meta hot memory, harness file imports and compile-context files. Statically shipped bundled skills, plugin trees and bootstrap templates are delivery artifacts and excluded; the Memorable procedure-extraction implementation is excluded. |
files, graph, rdbms, repo, vector [wired] | pull, push [wired] | coarse, identifier, inferred-lexical [wired] | yes [wired] | enforcement, instruction, knowledge, ranking, validation [wired] | AAS-2026-09-23-gbrain-01 |
| JEPA-Anything | Accumulated or revised memory and later-consumer routes within the shipped compiler skill and validator/generator; none established. Current-task products, static guidance, host conversation storage, and external training are excluded. | inapplicable | inapplicable | inapplicable | no [wired] | inapplicable | AAS-2026-09-25-jepa-anything-01 |
| LHTB | Bundled Harbor single-step continue-until-timeout cross-phase project files, AgentContext, verifier artifacts, derived feedback, artifact transfer and result retention; restart and same-conversation, shared and separate verifier branches, process rewards disabled. Concrete agent/model, provider and test internals excluded. | files, in-memory [wired] | pull, push [wired] | coarse, identifier [wired] | yes [wired] | not-determinable | AAS-2026-09-25-lhtb-01 |
| mem | Accumulated journal content and metadata; derived or caller-authored entity facts, active-body/index projections, checkpoints, dispositions, controls, publication receipts/intents; managed writeback; versioned task file; their acquisition, maintenance, and later-consumer routes. Static program/skill prompts and pretrained/provider model assets are components, not accumulated memory; transient search/HTTP result caches are delivery state. | files, repo [wired] | pull, push [afforded] | coarse, identifier, inferred-lexical [afforded] | yes [wired] | enforcement, instruction, knowledge, ranking, routing, validation [afforded] | AAS-2026-09-25-instinctual-memory-01 |
| MerchantBench | ReAct baseline retained dialogue and control metadata, optional current-run scratchpad and history, environment trace/usage records and consumed protocol cursors/idempotency; both tool-enabled and tool-denied branches. Excludes simulator domain databases and provider internals. | files, in-memory [wired] | pull, push [wired] | coarse [wired] | yes [afforded] | enforcement, instruction, knowledge, learning, routing [afforded] | AAS-2026-09-25-merchantbench-01 |
| Meta^n | Meta^n-owned use-accumulated task traces, candidate code/rationale and frozen solutions, archive selection/checkpoint state, optional imported seeds, and built-in per-task conversation summaries; includes repository external-agent injection/Trace conversion, excludes SDK-internal memory, provider weights, static prompts/configs, datasets and diagnostic-only opaque log payloads. | files, in-memory [wired] | pull, push [afforded] | coarse, identifier, inferred-lexical [wired] | yes [wired] | enforcement, instruction, knowledge, learning, ranking, routing, validation [wired] | AAS-2026-09-24-meta-n-01 |
| ModularRSI | Accumulated ModularRSI solver chat/continuations and guard/observation state; trace-derived findings/proposals, imported and evolved module generations, archive/status/access metadata, editor-memory capability, task-history/review state and composition replay cache. Named subsystem and necessary interfaces only; static doctrine, unrelated Harbor agents, external providers and benchmark internals excluded. | files, in-memory [wired] | pull, push [wired] | coarse, identifier, inferred-judgment [wired] | yes [wired] | enforcement, instruction, knowledge, ranking, routing, validation [wired] | AAS-2026-09-25-modularrsi-01 |
| OpenViking | Session message capture, archive/commit, cumulative Working Memory and partial-Turn checkpoints; user-stage memory extraction/merge/write and later continuation, extraction prefetch, direct find, context-aware search and bounded user-memory recall. Excludes agent-evolution/training, session skills, resource/wiki compilation, plugin/bot loops and arbitrary host stores. | uninspected | pull, push [afforded] | coarse, identifier, inferred-embedding, inferred-judgment [wired] | yes [wired] | knowledge, ranking, routing [wired] | AAS-2026-09-25-openviking-01 |
| pi-posthorse | Extension-owned notes and migration, consumed native transcript and context-edit/image payloads, handoff/recovery records and reminder fingerprints, and their extension write, selection and later-consumer routes. Host persistence and fresh-window assembly are documented external contracts; host-internal summary generation and the Codex prototype are excluded. | files, in-memory [claimed] | pull, push [claimed] | coarse, identifier [wired] | yes [claimed] | enforcement, instruction, knowledge, ranking, routing [claimed] | AAS-2026-09-25-pi-posthorse-01 |
| PrimeScientist | Source-owned retained plans, priors, packets/statistics, logs and diagnostic reductions, inherited sandbox artifacts, imported insight files, baseline history and best workspace, and the reflector session access handle across search and vanilla with AutoLab/FIRE Codex interfaces. External Codex/provider context storage and transformations are excluded; their resume boundary remains explicitly opaque. Static task datasets/instructions and unrelated vendored utilities are not memory merely because stored. | files, in-memory [wired] | pull, push [afforded] | coarse, identifier, inferred-lexical [wired] | yes [wired] | enforcement, instruction, knowledge, ranking, routing [afforded] | AAS-2026-09-25-primescientist-01 |
| Reflexion | HotPotQA accumulated attempt text, reflection text, formatted delivery views and their file exports; CoT with/without context, notebook ReAct alternatives and the separate environment-based ReAct implementation. Static prompts/data, provider parameters and non-memory payloads inside opaque saved agents are outside this operative memory profile. Saved-agent contents remain uninspected; no archive-to-agent consuming route is established. | files, in-memory [wired] | push [wired] | coarse [wired] | yes [wired] | instruction, knowledge [wired] | AAS-2026-09-25-reflexion-02 |
| RSIAgent | Actor-owned durable memory and its access metadata; accumulated Actor, Curriculum and Verifier context; retained worklogs, diagnoses, continuation projections and their later consumers in the shipped OSWorld and ALE paths. Excludes static instructions, model-provider internals, generic event telemetry, benchmark assets and VM implementation. | files, in-memory [wired] | pull, push [wired] | coarse, identifier [wired] | yes [wired] | not-determinable | AAS-2026-09-24-rsiagent-01 |
| SkillLift | Retained task-local portfolio content, rubric versions, plans, scalar evaluation history, candidate patches and resume state accumulated by the flagship portfolio coordinator, including its superproject-owned WildClawBench and SkillsBench delivery paths. Static seed authoring, excluded benchmark internals, remote model internals and reuse after final export are outside. | files, in-memory [wired] | pull, push [afforded] | coarse, identifier, inferred-judgment [wired] | yes [wired] | instruction, knowledge, learning, ranking, routing, validation [afforded] | AAS-2026-09-25-skilllift-01 |
| SoL-Pi | Accumulated SoL-Pi observation and diagnostic archives, derived receipts, projection access metadata, retained plan/progress and scheduling statistics, and the native compaction/continuation boundary when the optional mechanisms are enabled. Includes host session retention and replay only as source-visible call contracts; excludes Pi internals, provider weights, ordinary edited project files and shipped static configuration/instructions. | files, in-memory [wired] | pull, push [afforded] | not-determinable | yes [wired] | instruction, knowledge, routing, validation [wired] | AAS-2026-09-24-sol-pi-01 |
| Supermemory | withSupermemory local formatted-context cache and automatic injection, imported profile/search service objects, and conversation upload payloads including images and optional tools; remote transformation/storage internals and caller template/model implementations excluded. | in-memory, service-object [wired] | pull, push [wired] | not-determinable | not-determinable | knowledge, routing [wired] | AAS-2026-09-25-supermemory-02 |
| SwarmWorld | SwarmWorld explicit within-episode private records, empirical knowledge, research-state and feedback derivatives, shared publications/messages, recipes and program/skill lineage, adaptive access metadata, queued plans, and files insofar as recorded model/action traces reconstruct these objects. Includes both disabled/default and enabled configuration branches, including optional turnover. Excludes provider internals, model weights, static scenario/configuration doctrine, presentation-only histories and external study datasets. | files, in-memory [wired] | pull, push [wired] | coarse, identifier, inferred-lexical [wired] | yes [wired] | enforcement, instruction, knowledge, ranking, validation [wired] | AAS-2026-09-25-swarmworld-01 |
| Tardigrade | Accumulated event history and derived runtime control, compaction checkpoints, output rejection/repair feedback, referenced attachment bytes and provider continuation consumed by the embedded infer/compact subsystem; service interfaces and transient projections included, physical stores and external producer implementations excluded. | in-memory, service-object [wired] | pull, push [wired] | coarse, identifier [wired] | yes [wired] | enforcement, instruction, knowledge, routing, validation [wired] | AAS-2026-09-25-tardigrade-01 |
| WikiSkill (Stahl-G) | Product Wiki patterns, human feedback, carried/candidate skills, training outcome/trace artifacts and gate history; research Wiki/index/logs, impacts, skills, raw and compacted trajectories; retained installation backups and their withdrawal route. Static prompts are mechanism evidence, not accumulated memory. Host runtime internals and external datasets/services are excluded. Includes isolated Spreadsheet study Wiki snapshots, candidate/final skill and purpose mappings, compacted traces and TRAIN reference feedback; its single round has no deployed next-round consumer. | files [wired] | pull, push [afforded] | coarse, identifier [wired] | yes [wired] | enforcement, instruction, knowledge, routing, validation [wired] | AAS-2026-09-17-wikiskill-stahl-g-01 |
doc-grounded (3)
| System | Compared boundary | Storage | Read-back | Push selection | Trace learning | Authority | Evidence |
|---|---|---|---|---|---|---|---|
| AIDE2 | Paper-described retained candidate code, execution outputs, extracted scores and feedback, bounded history/error context, outer candidate code/grade history and accepted harness versions from AIDE0 through AIDE85, including outer-reviewer evaluation artifacts. Static seed instructions, provider/model internals, task/scorer implementations and opaque AIDEhuman internals are excluded. Transient summary persistence is unresolved. | not-determinable | not-determinable | not-determinable | yes [claimed] | instruction, knowledge, ranking, routing, validation [claimed] | AAS-2026-09-25-aide2-01 |
| Prove2Me | Accumulated local project files and access/environment state; documented platform mathematical objects, attempts, feedback, discovery metadata and drafts; extracted facts and prescribed upload checkpoints. Static skill text supplies routes but is not itself learned memory. Remote storage internals, host internals and Lean internals are excluded. | files, repo, service-object [afforded] | pull, push [afforded] | identifier [afforded] | yes [afforded] | enforcement, instruction, knowledge, ranking, routing, validation [afforded] | AAS-2026-09-25-prove2me-01 |
| WikiSkill | Paper-described retained raw training traces, wiki patterns/catalog/evolution log/impact history, SKILL.md and PURPOSE.md, and their evolution, validation, later inference and cross-model reuse routes. Include the inference-wiki-access ablation as an alternative with unspecified delivery. Include proposer outcome summaries as a delivery derivative, without assuming separate durable storage. Exclude static benchmark prompts, underlying tool/model implementations and ordinary within-task action-history assembly. | files [claimed] | pull, push [claimed] | not-determinable | yes [claimed] | not-determinable | AAS-2026-09-25-wikiskill-01 |
Input identities
kb/agentic-systems/reviews/aide2.md:9fc73159645dc0cdebbcd56c1b9eec79ee2a02d4e45c0da59eb6b2e2a6a2e4bf; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-25-aide2-01/result.md:fa6898ba490739f176c00f5f515ce05ac179963ea0049c7b27ef800f0fbf2a8f; sourcehttps://arxiv.org/abs/2609.26457atsha256:86d1b094380280450bf2281b1e38945a4f0c2528f2b88aebdf9ea8b9a1e8da8e.kb/agentic-systems/reviews/apache-maka.md:19512eb05ec7741249fed04ca0ec88707fd953293556d4aec6bba9e6bc537d08; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-05-apache-maka-06/result.md:f19ebc73bb527919051b211f43a6ae584d78e2de7aaa1abe195ed9f1ed0ff1a8; sourcehttps://github.com/apache/makaat02f97c16d76e644d5b565889701958293ff7b5fb.kb/agentic-systems/reviews/arex-skill.md:b5c954b78399f85823288518246a2cc9a1084ba7e9815e9f89361493cf82ad28; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-25-arex-skill-01/result.md:973dc7c3679aaadff574cb9b4cc9d40602a8a9c120190430fa48bf8fdd356d47; sourcehttps://github.com/VectorSpaceLab/AREX-Skillatac3fe1afa80fb9a09775ecfb2b6cc3ba850a2db6.kb/agentic-systems/reviews/arsumbris.md:15c4b499c5a5f95798ebd610f9f82a1576d69ad7e4077d79ab6e859272222595; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-23-arsumbris-02/result.md:2745f83c243adeaeb224e17a2b8733be833977274156d8b855acebc8b4aed54a; sourcehttps://github.com/arsumbris/arsumbrisatarsumbris-0.0.1-alpha-release-bundle-2026-09-23.kb/agentic-systems/reviews/contextpilot.md:f30b879fd314655a95879674d3b9c5744de8a5c1af837c5bc9fd8ab5bc7ddb32; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-25-contextpilot-01/result.md:8cc0c60126af6b160e9e9496e5b3de7323e61bd6623f178e40d7a35d480ee2c3; sourcehttps://github.com/Tencent/ContextPilotat782cbb6611fb610c4cf6fafda6022b7e89cae191.kb/agentic-systems/reviews/dualgraph.md:8265827f6d562ecf726dac39e7a6d0918aa61b06cfe13fe0b80ef583027003e4; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-25-dualgraph-01/result.md:4f399f7c21431928fb2cf79da0fef75339874937f0d82bac219b5eb0f1cdb412; sourcehttps://github.com/microsoft/DKI_LLMatc090b7d57e3996c8900e86fecd16085b27e07f29.kb/agentic-systems/reviews/eal-bench.md:3ba65f54478b650422fe32664e912b94a667f9a125bbc301ec3618db7dbd2fcc; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-25-eal-bench-01/result.md:8dc0d5ebb29c47255e7a063dc66397a6baf94128595aed5c08657a5c95f22efe; sourcehttps://github.com/tommasocerruti/eal-benchat51648690bc52d7a9c7ac080a2c67a784fe9d56cb.kb/agentic-systems/reviews/ecdysis.md:ba6e339101903d11177c230eb25b284f075b798928ad15af6788576b1caaf952; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-25-ecdysis-01/result.md:fc2ef3c6a009d4d93c1da5b9aa61d3bf112f95bf50a9c88736d4e336f8979533; sourcehttps://github.com/cuiyu-ai/Ecdysisatcf93866d545b0974dbb0bc83b39c31fbbdeeecb8.kb/agentic-systems/reviews/enoch.md:4a832f0a47c050826e2f72f67c697af4276183cac6ecf5a2046179205ac2b1a5; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-25-enoch-01/result.md:00798b76561a0249d41b723e62d503d7d5464c8c92bf17b8dc6cd36d08807df3; sourcehttps://github.com/our-ark/enochat81000d502e776a7fd2ff39904f71f14084c9b4a8.kb/agentic-systems/reviews/evoontology.md:4c08ec36d5e97f4c135b50967d02fe711a70f017676106a7d64247a404b174da; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-25-evoontology-01/result.md:aa9766294d0d76b819168cc4d430077f219929da7689d5b5effeedb374ce105d; sourcehttps://github.com/ruc-datalab/EvoOntologyatddbb1c991de5a33e27eb32bc86a4212a7537e01e.kb/agentic-systems/reviews/fragility-grid.md:605a66d240aa5d748809d51a4a260e6e55793be3e7e99cc7f7ae2873e60ed906; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-25-fragility-grid-01/result.md:70d8208bdae0dedbb7f16efd1e23d0c699c5f3e3ec5d5bbad313553daf88ec9f; sourcehttps://github.com/NikolaTesla-007/fragility-gridat3f51444ead009d8351de1b6b19bf901c4da3d420.kb/agentic-systems/reviews/gbrain-garrytan.md:a683334dff5207d21b2508c5ea1598dc635e521b4e5246fc505b582076245604; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-23-gbrain-01/result.md:e07b4133703a2594672dd77ac8b6906e6d1173f83b658e0165b42f4f1885f2b1; sourcehttps://github.com/garrytan/gbrainat6040075c6cb95be5881cc2e1b76ef7d71f4e5d29.kb/agentic-systems/reviews/jepa-anything.md:0cdfbab20d2fd0d32f287109a3d345b994964dc7397bd9ba7ee3d0f3143463e6; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-25-jepa-anything-01/result.md:61c8f37fd062024ef63441a578523566765fafc22a0fb6dbf968ea76e5df11b0; sourcehttps://github.com/Gen-Verse/JEPA-Anythingatc6e6c88f3ef75a4ce7acd660d6fa5779d995512c.kb/agentic-systems/reviews/lhtb.md:f2a3c02037c284339042f279ffa9893ba8c5a1223eeb6499ae59532b0305e025; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-25-lhtb-01/result.md:615b32ab30f9072f535b674f92b9fd95609cb3f0a9d893c6abc3c7ba51a58af9; sourcehttps://github.com/zli12321/LHTBatd78f5eb52ad754c5ee9154741af73130a85a65b8.kb/agentic-systems/reviews/instinctual-memory.md:0e47c8f8c3e465ab7aea7beec355279d885373a920e7f358feae4d0df4b36e51; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-25-instinctual-memory-01/result.md:e84ebd9e68f376b0f3bb91d87492bc51239b76fe0ee796b2665bf1d77cb9d105; sourcehttps://github.com/jasonkneen/instinctual-memoryat6acb13dc35765bf5ccfc87e445dd09c480f1c28a.kb/agentic-systems/reviews/merchantbench.md:519b6eac200b35ba7935b6da7b63ece75b8eb391cb82485e863f7a32c6c8712c; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-25-merchantbench-01/result.md:615a4b3b68f419eb12331d03f36166f166e384ace782fc2743ab7e180b74215b; sourcehttps://github.com/KhanCold/merchantbenchatf44ce969aeccfd65d1eef6afe50f69868e510946.kb/agentic-systems/reviews/meta-n.md:64c839db3bad2fc4a2c04430bb69509caeb90a20a6eb136463df49e57d5f8bfb; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-24-meta-n-01/result.md:fb252da4610374093d302a11bb4e7ca6cb9a521dbd9c27ef89822ba58386fefa; sourcehttps://github.com/minnesotanlp/meta-natb7081843d3c7b0e0f418ca10aaf2ccbff856e7f8.kb/agentic-systems/reviews/modularrsi.md:d8f606a70a6a9aa522dc7dc4889af27fda56a80390530f5eb3bfdc0800d46e3f; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-25-modularrsi-01/result.md:1a1508b4847328bc2746c7f71397814f98386abc1e5da2c69b5367a7865955ac; sourcehttps://github.com/IQuestLab/ModularRSIatb5c72c36b0d08ff93f00ee202a8fbdebe849dfb9.kb/agentic-systems/reviews/openviking.md:88e0158486e301db7643adcf007af948457a7d5112a91eecadd5bdcab73c1aaa; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-25-openviking-01/result.md:1e8ba47630396c1649ccfa715649a31185f4c74fa14446e52226a189993a63af; sourcehttps://github.com/volcengine/OpenVikingat4edc30b068934893bc94a4e1b8e87bab2100bce5.kb/agentic-systems/reviews/pi-posthorse.md:e16939d9a6a035b1bbfa5a0fe70c893d09fcaca57da13667a7c339b6d8a6fcf7; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-25-pi-posthorse-01/result.md:19d203c7135f34278b17016d93a90c7637adee2484ca1f2d7810f59a0a2a0782; sourcehttps://github.com/fitchmultz/pi-posthorseat5bdba3536e186a7be845844d0398c3485a7f8e2b.kb/agentic-systems/reviews/primescientist.md:10b590868f7a37a85a2627ab35312f106ea19450d0a44e0a3dbc52998d16b100; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-25-primescientist-01/result.md:b04d0a39fdf389ca522784541573e85905762029469d2d890ea38cd8ba7ee36b; sourcehttps://github.com/Henri-XYu02/PrimeScientistat29971beac6f4f4b41309b1326762e1b83ceece98.kb/agentic-systems/reviews/prove2me.md:65b6b550cce4d116ef21cdc884d8f94e6e7775981eaa1f4b76da12c76c6528f3; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-25-prove2me-01/result.md:20273c57e7ee775353a5afafc1a6e601182f5d84624f97641fe8b7845e54ecd8; sourcehttps://github.com/prove2me/prove2me_workspaceat326b972580e0640b1f3739ec7b12d2b34d8d3527.kb/agentic-systems/reviews/reflexion.md:941686d5ecd5e425288adf674982fce5d0fd516cf93db26d0930b02f40428a20; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-25-reflexion-02/result.md:93285ee41e4ffa28444c2b5b7818fd771c7878cd25fd7923e5f17938c267219e; sourcehttps://github.com/noahshinn/reflexionat218cf0ef1df84b05ce379dd4a8e47f17766733a0.kb/agentic-systems/reviews/rsiagent.md:79bb86f7d2baca0a82637d7b33388fcdcca546312b39da86815747db58e474b9; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-24-rsiagent-01/result.md:08ec7733d485341e2768725babe475c29aa5f0f33ffa6ac7c41a76532a6cd966; sourcehttps://github.com/AetherLabsAI/RSIAgentata9e56263f6deaa493496ad6b155fe24bf131bc12.kb/agentic-systems/reviews/skilllift.md:d942a0720e7072ee47d3e04679fae27a10e07eb51f4596443b9d75a8f44d6d3f; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-25-skilllift-01/result.md:5582d6a34c759515681c476a2cef7a5d7592b3fcb39d5c29f8e56f44f31982b4; sourcehttps://github.com/WalteR-MittY-pro/SkillLiftat599358b4d4c4a27c0e004df8228ab93026600653.kb/agentic-systems/reviews/sol-pi.md:8dad8b9313f475c58824e309370ab3addf286aa63910fe77451d00653381870a; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-24-sol-pi-01/result.md:aff4937d35f2287af3204726aef7ea934b134209c987953de3ab8a09ffffaa16; sourcehttps://github.com/NVlabs/SoL-Piat1559b5cb12c72da4a485bc50fe326586b216fb19.kb/agentic-systems/reviews/supermemory.md:1d16df257062c800b42f3bbab41f2fabf0b02c7ab57f0f17c3f024d20e074860; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-25-supermemory-02/result.md:e17dce575e775ef57c49b700dcaf640d3f332a19d5096072d1e8c4b28c05c408; sourcehttps://github.com/supermemoryai/supermemoryat0e12f0b3a65af1cf7b03f48561f20ddf4369bc3f.kb/agentic-systems/reviews/swarmworld.md:cbf05b92ffa6ca0c9bba691f654547e5bc9d00e03631e9ce0d672993b953fa1c; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-25-swarmworld-01/result.md:4a9d0bc1dbd7479ace7f980b34a827ff0bb5c22c13cd39cc4388999ef546aaaf; sourcehttps://github.com/lamm-mit/SwarmWorldat6af7ae9fa36d98b07b0492cf139658e8af1f6eab.kb/agentic-systems/reviews/tardigrade.md:c49055b5c826fc375151466e094e73aa7f28cf0c55bd6c6f163a66bde7377e05; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-25-tardigrade-01/result.md:a47b08d9087eced85e18424df28ae9fdfe363f5482e51e630d3df9803904e802; sourcehttps://github.com/clavia-labs/tardigradeat1c4f4efaab2aaeec0bc482bc38e8ddf3be6f8267.kb/agentic-systems/reviews/wikiskill.md:2ccbdd966dc5d6ede5bed3221cce141ca68be4b59d0a07171c87c06e40275c38; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-25-wikiskill-01/result.md:1b27b3f32961a43c1df6a9bd0fc742d07785d9ab2ba0b1951b4d37195175f82b; sourcehttps://arxiv.org/abs/2608.27454atsha256:c093e240375e9780468edca9aa91aefecb8714ebdbef40810316644d9b2822d0.kb/agentic-systems/reviews/wikiskill-stahl-g.md:5102bd052f6473075d75644cd9c56a5adbcc6dcd6b0d67845a68c539719167ca; exact resultkb/reports/retained/agentic-system-analysis/AAS-2026-09-17-wikiskill-stahl-g-01/result.md:740c450f3edefc8d13f81300e3cd05596fbf59278bf886a3576e51686314f2a8; sourcehttps://github.com/Stahl-G/wikiskillat9df975b2145a0e924f344f2a3d116e11d3f025ac.