{ "profile": "wrs-published-research-view-1", "edition": "ratings-e039106f30dec177", "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS", "entry": { "id": "yang-et-al-plugmem-knowledge-centric-agent-memory-2026", "title": "PlugMem: turning experience into reusable facts and procedures", "summary": "An agent that rereads every past action can waste its attention on details that no longer matter. PlugMem extracts two kinds of reusable knowledge: propositions describing facts and prescriptions describing how to accomplish a goal. These connect back to the experience from which they were derived. A retrieval stage selects relevant knowledge, and a reasoning stage condenses it for the agent’s current task. Yang and colleagues test the same basic module on conversational memory, multi-hop questions and web tasks. The contribution is this knowledge-centered memory pipeline and its scoped evaluation.", "boundary": "The published experiments primarily test creating and retrieving memory, rather than certifying every update and deletion operation. Reported memory-token savings exclude some preparation and reasoning costs. Some baselines use different models or subsets. Web-task reuse includes added human demonstrations and related task templates. A provenance link helps inspect a statement; it does not make the statement true. The information-density diagnostic is not an independent scientific-value score or a guarantee of universal transfer.", "area": "External agent memory", "topic": "Artificial intelligence", "year": 2026, "lane": "HUMAN", "result": "ALGORITHM", "package_id": "yang-et-al-plugmem-knowledge-centric-agent-memory-2026", "target": { "record_id": "urn:uuid:fcf98384-e136-4c27-a7de-abb13ff57109", "version": 1, "content_hash": "sha256:01d5b6edfb485710a1b12a7dcf15811daa56aa1849cf317cd6ae250abfdca966" }, "humans": [ "Ke Yang", "Zixi Chen", "Xuan He", "Jize Jiang", "Michel Galley", "Chenglong Wang", "Jianfeng Gao", "Jiawei Han", "ChengXiang Zhai" ], "authors_complete": true, "human_role_note": "Source-listed human researchers and engineers for the exact method or resource. The evaluated learning system is not an independently attested scientific author. Earlier methods, funding and data-collection roles remain separate. Individual effort shares and unresolved person identities are not inferred.", "agents": [], "related": [], "subresults": [], "scientific_assessment": { "source_review": "SOURCE_CURATED", "scientific_review": "NOT_REVIEWED_BY_WRS", "formal_replay": "NOT_RUN", "provider_attestation": "NONE" }, "sources": [ { "id": "cur20261017-arxiv-2603-03296v1", "title": "PlugMem: A Task-Agnostic Plugin Memory Module for LLM Agents", "url": "https://arxiv.org/abs/2603.03296v1", "date": "2026-02-06" } ], "points": "40000", "published_grade": { "id": "yang-et-al-plugmem-knowledge-centric-agent-memory-2026", "entry_id": "yang-et-al-plugmem-knowledge-centric-agent-memory-2026", "band": "B", "modifier": "+", "target": { "record_id": "urn:uuid:fcf98384-e136-4c27-a7de-abb13ff57109", "version": 1, "content_hash": "sha256:01d5b6edfb485710a1b12a7dcf15811daa56aa1849cf317cd6ae250abfdca966" }, "what_changed": "An agent that rereads every past action can waste its attention on details that no longer matter. PlugMem extracts two kinds of reusable knowledge: propositions describing facts and prescriptions describing how to accomplish a goal. These connect back to the experience from which they were derived. A retrieval stage selects relevant knowledge, and a reasoning stage condenses it for the agent’s current task. Yang and colleagues test the same basic module on conversational memory, multi-hop questions and web tasks. The contribution is this knowledge-centered memory pipeline and its scoped evaluation.", "scope": "An agent that rereads every past action can waste its attention on details that no longer matter. PlugMem extracts two kinds of reusable knowledge: propositions describing facts and prescriptions describing how to accomplish a goal. These connect back to the experience from which they were derived. A retrieval stage selects relevant knowledge, and a reasoning stage condenses it for the agent’s current task. Yang and colleagues test the same basic module on conversational memory, multi-hop questions and web tasks. The contribution is this knowledge-centered memory pipeline and its scoped evaluation. The published experiments primarily test creating and retrieving memory, rather than certifying every update and deletion operation. Reported memory-token savings exclude some preparation and reasoning costs. Some baselines use different models or subsets. Web-task reuse includes added human demonstrations and related task templates. A provenance link helps inspect a statement; it does not make the statement true. The information-density diagnostic is not an independent scientific-value score or a guarantee of universal transfer.", "status": "PROVISIONAL_EDITORIAL", "scientific_correctness_certified": false }, "assessment_disposition": { "status": "ASSESSED", "label": "Provisionally assessed", "reason": "A current grade and separate reference score identify this exact contribution.", "target": { "record_id": "urn:uuid:fcf98384-e136-4c27-a7de-abb13ff57109", "version": 1, "content_hash": "sha256:01d5b6edfb485710a1b12a7dcf15811daa56aa1849cf317cd6ae250abfdca966" }, "edition": "ratings-e039106f30dec177" } }, "version": { "kind": "PUBLIC_RECORD_VIEW", "previous": null, "published_at": "2026-10-23T00:00:00Z", "publisher": { "principal_id": "urn:uuid:80116931-785b-40a2-a24b-9d59db0c7e33", "role": "WRS_EDITORIAL_PUBLICATION" }, "record": { "contributors": [ { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Ke Yang", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "PlugMem: A Task-Agnostic Plugin Memory Module for LLM Agents", "url": "https://arxiv.org/abs/2603.03296v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Zixi Chen", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "PlugMem: A Task-Agnostic Plugin Memory Module for LLM Agents", "url": "https://arxiv.org/abs/2603.03296v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Xuan He", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "PlugMem: A Task-Agnostic Plugin Memory Module for LLM Agents", "url": "https://arxiv.org/abs/2603.03296v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Jize Jiang", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "PlugMem: A Task-Agnostic Plugin Memory Module for LLM Agents", "url": "https://arxiv.org/abs/2603.03296v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Michel Galley", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "PlugMem: A Task-Agnostic Plugin Memory Module for LLM Agents", "url": "https://arxiv.org/abs/2603.03296v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Chenglong Wang", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "PlugMem: A Task-Agnostic Plugin Memory Module for LLM Agents", "url": "https://arxiv.org/abs/2603.03296v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Jianfeng Gao", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "PlugMem: A Task-Agnostic Plugin Memory Module for LLM Agents", "url": "https://arxiv.org/abs/2603.03296v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Jiawei Han", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "PlugMem: A Task-Agnostic Plugin Memory Module for LLM Agents", "url": "https://arxiv.org/abs/2603.03296v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "ChengXiang Zhai", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "PlugMem: A Task-Agnostic Plugin Memory Module for LLM Agents", "url": "https://arxiv.org/abs/2603.03296v1" } } ], "evidence": [ { "description": "Bibliographic/stated-result source; not a WRS proof verification", "id": "source-0", "kind": "source_reference", "source": { "citation": "PlugMem: A Task-Agnostic Plugin Memory Module for LLM Agents", "url": "https://arxiv.org/abs/2603.03296v1" } } ], "origin": "HUMAN_LEGACY", "record_profile": "wrs-local-research-notes-0.1", "record_type": "research_note", "rrp_version": "0.1", "scope": { "description": "The published experiments primarily test creating and retrieving memory, rather than certifying every update and deletion operation. Reported memory-token savings exclude some preparation and reasoning costs. Some baselines use different models or subsets. Web-task reuse includes added human demonstrations and related task templates. A provenance link helps inspect a statement; it does not make the statement true. The information-density diagnostic is not an independent scientific-value score or a guarantee of universal transfer.", "domain": "External agent memory" }, "sources": [ { "citation": "PlugMem: A Task-Agnostic Plugin Memory Module for LLM Agents", "url": "https://arxiv.org/abs/2603.03296v1" } ], "statement": "# PlugMem: turning experience into reusable facts and procedures\n\n## Keep the lesson and the receipt\n\nSuppose an agent learns how to find a shop’s opening hours. A complete browser trace may contain many irrelevant clicks. A reusable procedure might say: open the official contact page, check the location and read the holiday exception. PlugMem distinguishes this kind of prescription, or how-to knowledge, from a proposition, or factual statement. Both can point back to the episode that produced them.\n\n## Three jobs happen before the answer\n\nFirst, the system structures experience into reusable knowledge. Second, it retrieves knowledge relevant to a new task. Third, a reasoning stage condenses the retrieved material for the agent. Each stage can make a mistake and consume computation. The source trace is therefore useful for inspection. A link to that trace does not prove that the extracted fact is correct or that a procedure works in every environment.\n\n## A small answer can hide a large preparation bill\n\nHere is an invented cost example. Five questions each need ten thousand units of context without preparation, making fifty thousand. A prepared system spends twenty thousand units building memory and then one thousand per question, making twenty-five thousand overall. The answer context fell by ninety percent, but this total fell by fifty percent. These are classroom units, not the paper’s measured spending. Actual token billing can also weight input and output differently.\n\n## Try an inspectable miniature\n\nGive your agent three public text notes and ask for two facts and one reusable procedure. Every item should quote only a short necessary phrase, identify its source note and state any conditions. Then introduce a conflicting note and ask what remains uncertain. The paper’s main benchmarks test specific memory operations and task setups; this small practice checks whether your extracted knowledge can be inspected, without pretending to validate the full system.\n\n## Make uncertainty reusable\n\nKeep conflicting source notes beside the extracted proposition rather than quietly choosing the more confident wording. A later participant should see what the statement depends on and what evidence would resolve it. The useful output is a small, source-linked knowledge table, not an unsupported declaration that every extracted lesson generalizes.\n\n## Scientific scope\n\nThe published experiments primarily test creating and retrieving memory, rather than certifying every update and deletion operation. Reported memory-token savings exclude some preparation and reasoning costs. Some baselines use different models or subsets. Web-task reuse includes added human demonstrations and related task templates. A provenance link helps inspect a statement; it does not make the statement true. The information-density diagnostic is not an independent scientific-value score or a guarantee of universal transfer.\n\n## Sources for this explanation\n\n- [PlugMem: A Task-Agnostic Plugin Memory Module for LLM Agents](https://arxiv.org/abs/2603.03296v1): Original main sections3–4: knowledge units and episodic provenance, retrieval/reasoning, benchmark conditions and human-demonstration-assisted reuse; main evaluation primarily create/retrieve.\n", "summary": "An agent that rereads every past action can waste its attention on details that no longer matter. PlugMem extracts two kinds of reusable knowledge: propositions describing facts and prescriptions describing how to accomplish a goal. These connect back to the experience from which they were derived. A retrieval stage selects relevant knowledge, and a reasoning stage condenses it for the agent’s current task. Yang and colleagues test the same basic module on conversational memory, multi-hop questions and web tasks. The contribution is this knowledge-centered memory pipeline and its scoped evaluation.", "title": "PlugMem: turning experience into reusable facts and procedures" }, "record_id": "urn:uuid:fcf98384-e136-4c27-a7de-abb13ff57109", "version": 1, "source_ref": { "record_id": "urn:uuid:fcf98384-e136-4c27-a7de-abb13ff57109", "version": 1, "content_hash": "sha256:01d5b6edfb485710a1b12a7dcf15811daa56aa1849cf317cd6ae250abfdca966" }, "export_view": "wrs-public-editorial-view-1", "public_view_hash": "sha256:9dd982ec7ff451b94dce60fcb0d6d52dfe53a06605a651b376b11863733337ad" }, "target": { "record_id": "urn:uuid:fcf98384-e136-4c27-a7de-abb13ff57109", "version": 1, "content_hash": "sha256:01d5b6edfb485710a1b12a7dcf15811daa56aa1849cf317cd6ae250abfdca966" }, "published_grade": { "profile": "wrs-published-grade-1", "edition": "2026-10-28.1", "id": "yang-et-al-plugmem-knowledge-centric-agent-memory-2026", "entry_id": "yang-et-al-plugmem-knowledge-centric-agent-memory-2026", "revision": 1, "title": "PlugMem: turning experience into reusable facts and procedures", "context": "External agent memory", "track": "Reusable scientific method, experimental result or historical framework", "band": "B", "modifier": "+", "status": "PROVISIONAL_EDITORIAL", "assessed_on": "2026-10-23", "evidence_cutoff": "2026-10-23", "achievement_state": "SUPPORTED_SCOPED_RESULT", "what_changed": "An agent that rereads every past action can waste its attention on details that no longer matter. PlugMem extracts two kinds of reusable knowledge: propositions describing facts and prescriptions describing how to accomplish a goal. These connect back to the experience from which they were derived. A retrieval stage selects relevant knowledge, and a reasoning stage condenses it for the agent’s current task. Yang and colleagues test the same basic module on conversational memory, multi-hop questions and web tasks. The contribution is this knowledge-centered memory pipeline and its scoped evaluation.", "scope": "An agent that rereads every past action can waste its attention on details that no longer matter. PlugMem extracts two kinds of reusable knowledge: propositions describing facts and prescriptions describing how to accomplish a goal. These connect back to the experience from which they were derived. A retrieval stage selects relevant knowledge, and a reasoning stage condenses it for the agent’s current task. Yang and colleagues test the same basic module on conversational memory, multi-hop questions and web tasks. The contribution is this knowledge-centered memory pipeline and its scoped evaluation. The published experiments primarily test creating and retrieving memory, rather than certifying every update and deletion operation. Reported memory-token savings exclude some preparation and reasoning costs. Some baselines use different models or subsets. Web-task reuse includes added human demonstrations and related task templates. A provenance link helps inspect a statement; it does not make the statement true. The information-density diagnostic is not an independent scientific-value score or a guarantee of universal transfer.", "evidence_summary": "The linked original works support the scoped statement and contribution attribution. WRS has not independently replayed experiments or reconstructed a full mathematical proof. The published experiments primarily test creating and retrieving memory, rather than certifying every update and deletion operation. Reported memory-token savings exclude some preparation and reasoning costs. Some baselines use different models or subsets. Web-task reuse includes added human demonstrations and related task templates. A provenance link helps inspect a statement; it does not make the statement true. The information-density diagnostic is not an independent scientific-value score or a guarantee of universal transfer.", "references": [ { "title": "PlugMem: A Task-Agnostic Plugin Memory Module for LLM Agents", "url": "https://arxiv.org/abs/2603.03296v1", "locator": "Original scoped contribution, statement/method and evidence limits; source-specific attribution retained." } ], "roles": [ { "who": "The named authors in the research record", "credit": "SCIENCE", "did": "Source-listed human researchers and engineers for the exact method or resource. The evaluated learning system is not an independently attested scientific author. Earlier methods, funding and data-collection roles remain separate. Individual effort shares and unresolved person identities are not inferred." } ], "target": { "record_id": "urn:uuid:fcf98384-e136-4c27-a7de-abb13ff57109", "version": 1, "content_hash": "sha256:01d5b6edfb485710a1b12a7dcf15811daa56aa1849cf317cd6ae250abfdca966" }, "scientific_correctness_certified": false, "independent_expert_calibration": false, "method": "PROPRIETARY", "points": "40000" }, "canonical_agent_submission": false, "scientific_verification_awarded": false, "reader_guide": { "id": "yang-et-al-plugmem-knowledge-centric-agent-memory-2026", "target": { "record_id": "urn:uuid:fcf98384-e136-4c27-a7de-abb13ff57109", "version": 1, "content_hash": "sha256:01d5b6edfb485710a1b12a7dcf15811daa56aa1849cf317cd6ae250abfdca966" }, "sections": [ { "title": "Keep the lesson and the receipt", "text": "Suppose an agent learns how to find a shop’s opening hours. A complete browser trace may contain many irrelevant clicks. A reusable procedure might say: open the official contact page, check the location and read the holiday exception. PlugMem distinguishes this kind of prescription, or how-to knowledge, from a proposition, or factual statement. Both can point back to the episode that produced them." }, { "title": "Three jobs happen before the answer", "text": "First, the system structures experience into reusable knowledge. Second, it retrieves knowledge relevant to a new task. Third, a reasoning stage condenses the retrieved material for the agent. Each stage can make a mistake and consume computation. The source trace is therefore useful for inspection. A link to that trace does not prove that the extracted fact is correct or that a procedure works in every environment." }, { "title": "A small answer can hide a large preparation bill", "text": "Here is an invented cost example. Five questions each need ten thousand units of context without preparation, making fifty thousand. A prepared system spends twenty thousand units building memory and then one thousand per question, making twenty-five thousand overall. The answer context fell by ninety percent, but this total fell by fifty percent. These are classroom units, not the paper’s measured spending. Actual token billing can also weight input and output differently." }, { "title": "Try an inspectable miniature", "text": "Give your agent three public text notes and ask for two facts and one reusable procedure. Every item should quote only a short necessary phrase, identify its source note and state any conditions. Then introduce a conflicting note and ask what remains uncertain. The paper’s main benchmarks test specific memory operations and task setups; this small practice checks whether your extracted knowledge can be inspected, without pretending to validate the full system." }, { "title": "Make uncertainty reusable", "text": "Keep conflicting source notes beside the extracted proposition rather than quietly choosing the more confident wording. A later participant should see what the statement depends on and what evidence would resolve it. The useful output is a small, source-linked knowledge table, not an unsupported declaration that every extracted lesson generalizes." } ], "references": [ { "title": "PlugMem: A Task-Agnostic Plugin Memory Module for LLM Agents", "url": "https://arxiv.org/abs/2603.03296v1", "locator": "Original main sections3–4: knowledge units and episodic provenance, retrieval/reasoning, benchmark conditions and human-demonstration-assisted reuse; main evaluation primarily create/retrieve." } ] }, "reading_status": "EXPLANATORY_GUIDE" }