Human-Inspired Memory Architecture for LLM Agents

TL;DR

This paper introduces a biologically-inspired memory architecture for LLM agents, enhancing long-term memory management through six cognitive mechanisms and a synthetic calibration method, evaluated on two benchmarks.

Contribution

It presents a novel, biologically-grounded memory system with a calibration approach that improves memory retention and efficiency without benchmark data exposure.

Findings

Achieved 97.2% retention precision with 58% store reduction on issue-tracking data.

Matched retrieval accuracy at 70.1% with a 200K-token context budget.

Improved preference recall by 13.3 percentage points at S-tier scale.

Abstract

Current LLM agents lack principled mechanisms for managing persistent memory across long interaction horizons. We present a biologically-grounded memory architecture comprising six cognitive mechanisms: (1) sleep-phase consolidation, (2) interference-based forgetting, (3) engram maturation, (4) reconsolidation upon retrieval, (5) entity knowledge graphs, and (6) hybrid multi-cue retrieval. Each mechanism addresses a specific failure mode of naive memory accumulation. We introduce a synthetic calibration methodology that derives all pipeline thresholds without benchmark data exposure, eliminating a common source of evaluation leakage. We evaluate on two benchmarks. First, a VSCode issue-tracking dataset (13K issues, 120K events) where deduplication-based consolidation achieves 97.2% retention precision with 58% store reduction (+21.8 pp over baseline). Second, the LongMemEval personal-chat benchmark where we conduct the first streaming M-tier evaluation (475 sessions, ~540K unique turns). At a 200K-token context budget, our pipeline matches raw retrieval accuracy (70.1% vs. 71.2%, overlapping 95% CI) while exposing a tunable accuracy/store-size operating curve. At S-tier scale (50 sessions), dedup-based consolidation yields a +13.3 pp improvement in preference recall.