{"paper":{"title":"MEMTIER: Tiered Memory Architecture and Retrieval Bottleneck Analysis for Long-Running Autonomous AI Agents","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"MEMTIER's tiered memory architecture improves long-running AI agent accuracy from 5% to 38% on the LongMemEval-S benchmark using only a 6GB consumer GPU.","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Bronislav Sidik, Lior Rokach","submitted_at":"2026-05-05T12:14:10Z","abstract_excerpt":"Long-running autonomous AI agents suffer from a well-documented memory coherence problem: tool-execution success rates degrade 14 percentage points over 72-hour operation windows due to four compounding failure modes in existing flat-file memory systems. We present MEMTIER, a tripartite memory architecture for the OpenClaw agent runtime that introduces a structured episodic JSONL store, a five-signal weighted retrieval engine, an attention-attributed cognitive weight update loop, an asynchronous consolidation daemon promoting episodic facts to a semantic tier, and a PPO-based policy framework "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"On the full 500-question LongMemEval-S benchmark, MEMTIER achieves Acc=0.382, F1=0.412 with Qwen2.5-7B on a consumer 6GB GPU - a +33 percentage point improvement over the full-context baseline (0.050 -> 0.382). With DeepSeek-V4-Flash fact pre-population, single-session recall reaches 0.686-0.714, exceeding the paper's RAG BM25 GPT-4o baseline (0.560).","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the observed accuracy and recall lifts are caused by the tripartite architecture, five-signal engine, consolidation daemon, and PPO weight adaptation rather than by benchmark-specific choices, model selection, or the external DeepSeek pre-population step, and that the infrastructure-validated components will deliver the stated performance gains once the camera-ready version is complete.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"MEMTIER delivers 38% accuracy on the 500-question LongMemEval-S benchmark with a 7B model on 6GB GPU, a 33-point gain over full-context baselines, via structured episodic memory, five-signal retrieval, and semantic consolidation.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"MEMTIER's tiered memory architecture improves long-running AI agent accuracy from 5% to 38% on the LongMemEval-S benchmark using only a 6GB consumer GPU.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"a5e5a513d98df722cd73bec43eee1b78e19830c93120abb57971f2142be57ed8"},"source":{"id":"2605.03675","kind":"arxiv","version":2},"verdict":{"id":"9fadb931-20d1-41b6-a851-4e7569a400e7","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-07T04:09:29.186448Z","strongest_claim":"On the full 500-question LongMemEval-S benchmark, MEMTIER achieves Acc=0.382, F1=0.412 with Qwen2.5-7B on a consumer 6GB GPU - a +33 percentage point improvement over the full-context baseline (0.050 -> 0.382). With DeepSeek-V4-Flash fact pre-population, single-session recall reaches 0.686-0.714, exceeding the paper's RAG BM25 GPT-4o baseline (0.560).","one_line_summary":"MEMTIER delivers 38% accuracy on the 500-question LongMemEval-S benchmark with a 7B model on 6GB GPU, a 33-point gain over full-context baselines, via structured episodic memory, five-signal retrieval, and semantic consolidation.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the observed accuracy and recall lifts are caused by the tripartite architecture, five-signal engine, consolidation daemon, and PPO weight adaptation rather than by benchmark-specific choices, model selection, or the external DeepSeek pre-population step, and that the infrastructure-validated components will deliver the stated performance gains once the camera-ready version is complete.","pith_extraction_headline":"MEMTIER's tiered memory architecture improves long-running AI agent accuracy from 5% to 38% on the LongMemEval-S benchmark using only a 6GB consumer GPU."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.03675/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-20T13:36:31.370937Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-20T00:31:21.602262Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T15:06:15.583763Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"db0cecd5c591a117ede0a6fef72eab3ace0338a22e009a96bc11416adec816bd"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}