Back to Ruflo

Memory Agent SOTA Report — 2026-06-13

v3/docs/dream/dream-gist-2026-06-13.md

3.34.04.9 KB
Original Source

Memory Agent SOTA Report — 2026-06-13

TL;DR: Three 2026 papers establish that flat HNSW retrieval without temporal decay is SOTA-deficient: MemMachine cuts token cost 80% via episodic ground-truth preservation (LoCoMo 0.9169); SSGM identifies semantic drift and knowledge-leakage as architectural risks in flat-vector long-term stores; the survey arXiv:2603.07670 names "learned forgetting" and "causal retrieval" as the two biggest unsolved frontiers. Ruflo's AgentDB has neither.


What's New in 2026

FindingSourceConfidence
MemMachine: episodic ground-truth store cuts tokens 80%, LoCoMo 0.9169arXiv:2604.04853 (Apr 2026)A
SSGM framework: flat HNSW without temporal decay risks topology-induced knowledge leakagearXiv:2603.11768 (Mar 2026, rev May 2026)B (no quant results)
Survey identifies 5 open frontiers: continual consolidation, causal retrieval, trustworthy reflection, learned forgetting, multimodal embodiedarXiv:2603.07670 (Mar 2026)A
Mem0 ECAI: 91% lower p95 latency (1.44 s vs 17.12 s), 90% lower token cost vs full-contextarXiv:2504.19413 (ECAI 2025)A
CrewAI v1.12: Qdrant Edge backend + hierarchical namespace isolationCrewAI changelog Apr 2026B
LangGraph v0.4: PostgresSaver checkpointer with TTL-aware state evictionLangGraph changelog Apr 2026B

Ruflo Current Capability

CapabilityStatusNotes
HNSW vector search✅ Active~1.9x at N=20k vs brute force (measured)
SQLite persistent cache✅ Activesql.js WASM backend
ReasoningBank / EWC++✅ ActivePrevents catastrophic forgetting
Temporal decay modeling❌ MissingEntries never age or expire
Episodic ground-truth preservation❌ MissingHNSW stores embeddings, not raw episodes
Consistency verification❌ MissingNo integrity check on consolidation
Namespace write-authority✅ In-progressADR-145 Proposed
LongMemEval score published⚠️ DeferredADR-088: <90% target, deferred
Causal / multi-hop retrieval❌ MissingADR-155 Proposed (2026-06-12, KG gap)

Competitor Comparison

SystemMemory BackendTemporal DecayEpisodic PreservationLongMemEval / LoCoMo
MemMachine3-layer (short/episodic/profile)✅ (TTL per tier)✅ raw episode storeLoCoMo 0.9169 (Grade A)
CrewAI v1.12Qdrant Edge + hierarchical isolation✅ configurable❌ embedding-onlyNot published
LangGraph v0.4PostgresSaver checkpointer✅ TTL eviction❌ state snapshotsNot published
ZepTemporal knowledge graph✅ native✅ entity tracking63.8% temporal retrieval (arXiv:2501.13956, Grade B)
Mem0Hybrid semantic+short-term❌ no decayLoCoMo ~49% temporal, 91% lower latency (Grade A)
Ruflo AgentDBSQLite + HNSW flat store❌ absentADR-088: deferred

Benchmarks

BenchmarkBest ScoreSystemYearGrade
LoCoMo0.9169MemMachine (GPT-4-mini)2026A (arXiv:2604.04853)
LongMemEvalS93.0%MemMachine2026A (arXiv:2604.04853)
Mem0 p95 latency1.44 s vs 17.12 s full-contextMem02025A (arXiv:2504.19413)
Zep temporal retrieval63.8% vs Mem0 49.0%Zep2026B (arXiv:2501.13956, single source)

Note: No 2026 benchmark data available for Ruflo AgentDB's LoCoMo or LongMemEvalS performance.


SOTA Proof & Witness

Session commit: 4cbcfc2671ad3f13ac9a648c1604c09fdb934248
Report SHA-256: 42dc7ff29b2f3c387bc7ce8ae8e528c8dda6f3904ed49e93d4100ddadc9d0acd
Witness stamp: 483dbb39773e936665cccf5eeb1e669a53e10390e9ee7f6cde3612ad10ff2368

Verifier: SHA-256 computed on pre-witness gist body. Reproduce:

printf '%s%s' "42dc7ff29b2f3c387bc7ce8ae8e528c8dda6f3904ed49e93d4100ddadc9d0acd" "4cbcfc2671ad3f13ac9a648c1604c09fdb934248" | sha256sum
# must equal 483dbb39773e936665cccf5eeb1e669a53e10390e9ee7f6cde3612ad10ff2368

  1. Implement temporal decay controller in AgentDB (ADR-156): Add TTL-based aging with configurable half-life per namespace tier (short: 7 days, episodic: 90 days, long-term: 1 year). Prevents semantic drift identified by SSGM (arXiv:2603.11768). Estimated ~120 LOC new controller.

  2. Add episodic ground-truth layer alongside HNSW embeddings: Store raw conversation episodes as first-class objects (not just embeddings) to enable MemMachine-style retrieval. Target: recover 80% token efficiency vs current embedding-only recall on LoCoMo-style queries.

  3. Run LoCoMo benchmark against current AgentDB: Establishes baseline before ADR-156 changes land. MemMachine 0.9169 is the 2026 SOTA bar. Current Ruflo likely trails without temporal decay + episodic layer.