{"id":"https://openalex.org/W7140280410","doi":"https://doi.org/10.48550/arxiv.2603.23013","title":"Knowledge Access Beats Model Size: Memory Augmented Routing for Persistent AI Agents","display_name":"Knowledge Access Beats Model Size: Memory Augmented Routing for Persistent AI Agents","publication_year":2026,"publication_date":"2026-03-24","ids":{"openalex":"https://openalex.org/W7140280410","doi":"https://doi.org/10.48550/arxiv.2603.23013"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2603.23013","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.23013","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2603.23013","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5030019120","display_name":"X Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Xunzhuo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130617562","display_name":"Bowei He","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"He, Bowei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130627866","display_name":"Xue Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Xue","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129563769","display_name":"Andy Luo","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Luo, Andy","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129552855","display_name":"Haichen Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Haichen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5033129678","display_name":"H Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Huamin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.272599995136261,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.272599995136261,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11714","display_name":"Multimodal Machine Learning Applications","score":0.1459999978542328,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10286","display_name":"Information Retrieval and Search Behavior","score":0.1371999979019165,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/correctness","display_name":"Correctness","score":0.6632000207901001},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.6549000144004822},{"id":"https://openalex.org/keywords/redundancy","display_name":"Redundancy (engineering)","score":0.6489999890327454},{"id":"https://openalex.org/keywords/context","display_name":"Context (archaeology)","score":0.49059998989105225},{"id":"https://openalex.org/keywords/routing","display_name":"Routing (electronic design automation)","score":0.4212000072002411},{"id":"https://openalex.org/keywords/recall","display_name":"Recall","score":0.4036000072956085}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8364999890327454},{"id":"https://openalex.org/C55439883","wikidata":"https://www.wikidata.org/wiki/Q360812","display_name":"Correctness","level":2,"score":0.6632000207901001},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6549000144004822},{"id":"https://openalex.org/C152124472","wikidata":"https://www.wikidata.org/wiki/Q1204361","display_name":"Redundancy (engineering)","level":2,"score":0.6489999890327454},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.49059998989105225},{"id":"https://openalex.org/C74172769","wikidata":"https://www.wikidata.org/wiki/Q1446839","display_name":"Routing (electronic design automation)","level":2,"score":0.4212000072002411},{"id":"https://openalex.org/C100660578","wikidata":"https://www.wikidata.org/wiki/Q18733","display_name":"Recall","level":2,"score":0.4036000072956085},{"id":"https://openalex.org/C2779530757","wikidata":"https://www.wikidata.org/wiki/Q1207505","display_name":"Quality (philosophy)","level":2,"score":0.39959999918937683},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.39160001277923584},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.34549999237060547},{"id":"https://openalex.org/C7545210","wikidata":"https://www.wikidata.org/wiki/Q838123","display_name":"Data redundancy","level":2,"score":0.3296000063419342},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.3003000020980835},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.2944999933242798},{"id":"https://openalex.org/C81669768","wikidata":"https://www.wikidata.org/wiki/Q2359161","display_name":"Precision and recall","level":2,"score":0.27799999713897705},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.2535000145435333},{"id":"https://openalex.org/C3018263672","wikidata":"https://www.wikidata.org/wiki/Q1296251","display_name":"Efficient algorithm","level":2,"score":0.25130000710487366}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2603.23013","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.23013","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2603.23013","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.23013","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Production":[0],"AI":[1,256],"agents":[2],"frequently":[3],"receive":[4],"user-specific":[5,125,184],"queries":[6,72,153],"that":[7,34,123,228,240],"are":[8],"highly":[9],"repetitive,":[10],"with":[11,27],"up":[12],"to":[13,18,69,128,154,165],"47\\%":[14],"being":[15],"semantically":[16],"similar":[17],"prior":[19],"interactions,":[20],"yet":[21],"each":[22],"query":[23],"is":[24,246],"typically":[25],"processed":[26],"the":[28,93,116,137,155,199,247],"same":[29],"computational":[30],"cost.":[31],"We":[32,53,134,202],"argue":[33],"this":[35,85],"redundancy":[36],"can":[37],"be":[38],"exploited":[39],"through":[40],"conversational":[41,67,187],"memory,":[42,241],"transforming":[43],"repetition":[44],"from":[45],"a":[46,55,61,74,96,107],"cost":[47,103],"burden":[48],"into":[49],"an":[50,223],"efficiency":[51,253],"advantage.":[52],"propose":[54],"memory-augmented":[56],"inference":[57,76],"framework":[58],"in":[59,182,254],"which":[60],"lightweight":[62],"8B-parameter":[63],"model":[64,99,109,119,132,244],"leverages":[65],"retrieved":[66,183],"context":[68],"answer":[70],"all":[71],"via":[73],"low-cost":[75],"path.":[77],"Without":[78],"any":[79],"additional":[80,224],"training":[81],"or":[82],"labeled":[83],"data,":[84],"approach":[86],"achieves":[87],"30.5\\%":[88],"F1,":[89,226],"recovering":[90],"69\\%":[91],"of":[92,95,139,152,193,250],"performance":[94,200,221],"full-context":[97],"235B":[98,108],"while":[100],"reducing":[101],"effective":[102],"by":[104,179,222],"96\\%.":[105],"Notably,":[106],"without":[110],"memory":[111,188],"(13.7\\%":[112],"F1)":[113,163],"underperforms":[114],"even":[115],"standalone":[117],"8B":[118],"(15.4\\%":[120],"F1),":[121],"indicating":[122],"for":[124],"queries,":[126],"access":[127],"relevant":[129],"knowledge":[130],"outweighs":[131],"scale.":[133],"further":[135,197],"analyze":[136],"role":[138],"routing":[140,147,173],"and":[141,209,252],"confidence.":[142],"At":[143],"practical":[144],"confidence":[145],"thresholds,":[146],"alone":[148],"already":[149],"directs":[150],"96\\%":[151],"small":[156],"model,":[157],"but":[158],"yields":[159],"poor":[160],"accuracy":[161,251],"(13.0\\%":[162],"due":[164],"confident":[166],"hallucinations.":[167],"Memory":[168],"does":[169],"not":[170],"substantially":[171],"alter":[172],"decisions;":[174],"instead,":[175],"it":[176],"improves":[177,220],"correctness":[178],"grounding":[180],"responses":[181],"information.":[185],"As":[186],"accumulates":[189],"over":[190],"time,":[191],"coverage":[192],"recurring":[194],"topics":[195],"increases,":[196],"narrowing":[198],"gap.":[201],"evaluate":[203],"on":[204],"152":[205],"LoCoMo":[206],"questions":[207],"(Qwen3-8B/235B)":[208],"500":[210],"LongMemEval":[211],"questions.":[212],"Incorporating":[213],"hybrid":[214],"retrieval":[215,229],"(BM25":[216],"+":[217],"cosine":[218],"similarity)":[219],"+7.7":[225],"demonstrating":[227],"quality":[230],"directly":[231],"enhances":[232],"end-to-end":[233],"system":[234],"performance.":[235],"Overall,":[236],"our":[237],"results":[238],"highlight":[239],"rather":[242],"than":[243],"size,":[245],"primary":[248],"driver":[249],"persistent":[255],"agents.":[257]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-03-26T00:00:00"}