SuperLocalMemory V3: Information-Geometric Foundations for Zero-LLM Enterprise Agent Memory

agent 2603.14588
information-geometrycognitive-memoryagent-memoryfisher-information

§1 TL;DR #

SLM-V3 replaces cosine similarity with Fisher-information-weighted retrieval (Theorem 6.1), heuristic decay with Riemannian Langevin dynamics on Poincaré ball (Theorem 6.3), and silent inconsistency with sheaf cohomology ($H^1 \neq 0$ = contradiction) — achieving +12.7 pp over engineering baseline on LoCoMo, with +19.9 pp on the hardest conversations, all without cloud dependency.

§2 Q1 / Q2 / Q3 #

Q1 痛点 #

Agent memory systems universally suffer three mathematically unaddressed problems: (1) uncertainty-blind retrieval — cosine similarity treats all embedding dimensions equally, ignoring per-dimension statistical precision; (2) unprincipled lifecycle — fixed TTL/exponential decay with manual half-lives ignores the geometry of memory space; (3) silent inconsistency — no system provides formal contradiction detection across multi-session, multi-partner, multi-temporal sources. Exhaustive venue search (NeurIPS/ICML/ICLR/ACL/EMNLP/AAAI 2020–2026 + arXiv) found no prior work connecting information geometry to agent memory.

Q2 方法 #

Fisher-information-weighted retrieval (§5.2): Each memory modeled as diagonal Gaussian $\mathcal{N}(\mu, \mathrm{diag}(\sigma^2))$ on a statistical manifold. Retrieval scoring: $s_{\text{FR}}(q, m_i) = \exp\left(-\frac{1}{T}\sum_k \frac{(\mu_{q,k} - \mu_{i,k})^2}{\sigma_{i,k}^2}\right)$. High-variance dimensions contribute less (uncertain); low-variance dimensions contribute more (confident). Graduated ramp from cosine to Fisher over first 10 accesses. Theorem 6.1: Fisher–Rao metric satisfies identity, symmetry, triangle inequality, sufficient-statistic invariance (Čencov's uniqueness), and $\Theta(d)$ computability.

Riemannian Langevin lifecycle dynamics (§5.3): SDE on Poincaré ball: $d\xi = -\lambda_\xi^{-2}\nabla_E U(\xi)dt + \sqrt{2T}\lambda_\xi^{-1}dW + \frac{1}{2}T(d-2)\lambda_\xi^{-1}\xi\,dt$. Potential $U(\xi) = \alpha\|\xi\|^2 - \beta \cdot n_{\text{access}} - \gamma \cdot r(\xi, c)$ drives important memories toward origin, pushes unimportant ones toward boundary (natural forgetting). Theorem 6.3: unique stationary distribution $\rho_\infty(\xi) \propto (1-\|\xi\|^2)^{-d} \exp(-U(\xi)/T)$ via Fokker–Planck + Foster–Lyapunov.

Sheaf-cohomological consistency (§5.4): Graph $G = (V, E)$ where $V$ = memory contexts, $E$ = shared entities. Coboundary operator $(\delta f)(u,v) = \rho_{u \to v}(f(u)) - f(v)$. Contradiction score $\kappa = \|\delta f\|^2 / (\|f\|^2 + \epsilon)$; when $\kappa > \tau = 0.45$, system creates supersedes edges. $H^1(\mathcal{F}) \neq 0$ means irreconcilable contradictions.

Four-channel retrieval: Semantic (Fisher–Rao, weight 1.2), BM25 keyword (weight 1.0), entity graph spreading activation (weight 1.3), temporal reasoning (weight 1.0). Fused via weighted reciprocal rank fusion: $\text{WRRF}(m) = \sum_i w_i / (k + r_i(m))$.

Three operating modes: Mode A (zero-LLM, 768d local embeddings, EU AI Act compliant), Mode B (local Ollama), Mode C (cloud 3072d embeddings + gpt-4.1-mini).

核心技术壁垒: The Fisher metric's per-dimension uncertainty weighting — when two memories are equidistant from a query under cosine but one has high-variance dimensions (many similar items) and the other low-variance (few items, high confidence), Fisher promotes the higher-confidence memory. Cosine cannot distinguish them. This advantage grows with memory store density (Proposition 7.2 proves Fisher breaks the cosine concentration barrier).

Q3 结果 #

MetricValue
Average improvement over engineering baseline+12.7 pp (6 conversations, 832 questions)
Max improvement (hardest conversation, conv-44)+19.9 pp
Mode A Retrieval (zero-LLM + external synthesis)74.8% on LoCoMo
Mode C (cloud-augmented, conv-30 only)87.7%
Cross-encoder ablation impact−30.7 pp (single largest contributor)
Fisher metric ablation impact−10.8 pp
Sheaf consistency ablation impact−1.7 pp
EverMemOS (SOTA, cloud-dependent)92.3%

§3 架构 / 方法图 #

flowchart LR Query --> QC[Query Classification
single_hop/multi_hop/temporal/aggregation] QC --> C1[Ch1: Semantic
Fisher–Rao weighted] QC --> C2[Ch2: BM25
Keyword matching] QC --> C3[Ch3: Entity Graph
Spreading activation] QC --> C4[Ch4: Temporal
Date proximity] C1 --> WRRF[Weighted RRF
Fusion] C2 --> WRRF C3 --> WRRF C4 --> WRRF WRRF --> CE[Cross-Encoder
Neural Reranking] CE --> SC[Sheaf Consistency
Filter superseded] SC --> LL[Langevin Lifecycle
Weight by state] LL --> TopK[Top-K Results]

Sheaf consistency operates at store time (creates supersedes edges on contradiction); Langevin dynamics run as background process. Fisher–Rao and reranking operate at retrieval time.

§4 作者证明 #

CheckStatusLocation
Fisher–Rao metric properties (Theorem 6.1)Proven via product decomposition + Čencov§6.1, Appendix A.1
Langevin stationary distribution (Theorem 6.3)Proven via Fokker–Planck + Foster–Lyapunov§6.2, Appendix A.2
Progressive-disclosure depth (Theorem 6.5)Proven via rate-distortion§6.3, Appendix A.3
Bounded effective memory count (Theorem 6.6)Proven via Hopfield capacity§6.3, Appendix A.4
Cosine concentration barrier (Proposition 7.1)Proven via spherical cap area§7.6
Fisher breaks concentration (Proposition 7.2)Proven§7.6

Important gap: Hopfield retrieval channel and native hyperbolic embeddings are theoretically developed but NOT implemented or evaluated in the current system.

§5 实验与数据 #

LoCoMo Benchmark (Table 3, §7.2): SLM-V3 Mode A Retrieval 74.8%, outperforming Mem0 (64.2%) without cloud dependency. Open-domain 85.0% (highest of any system). Mode C reaches 87.7% on conv-30 (81 questions) — comparable to Zep v3 (85.2%).

Fisher vs Cosine (Table 5, §7.4): Across 6 conversations, mathematical layers add +12.7 pp average. Improvement largest on hardest conversations: conv-44 (+19.9 pp), conv-49 (+18.8 pp). Consistent with theory: Fisher advantage grows with retrieval difficulty.

Ablation (Table 4, §7.3): Cross-encoder removal: −30.7 pp (dominant). Fisher: −10.8 pp. BM25: −6.5 pp. All math layers aggregate: −7.6 pp. Sheaf: −1.7 pp. Temporal: −0.2 pp. Bootstrap 95% CI for full system: [53.4, 74.0]. Cross-encoder CI [17.1, 45.7] — no overlap (statistically significant).

Scale Analysis (§7.6): At $N = 10^5$, cosine $\varepsilon$-cap contains ~100 vectors, overwhelming typical $K = 20$ retrieval budget (SNR → 0). Fisher–Rao provides strictly finer ranking under heteroscedasticity.

§6 论证链 #

StepClaimEvidenceDepends on
1Agent memory systems universally use cosine/heuristic-decay/no-consistency-checkLiterature survey: MemGPT, Mem0, Zep, MemOS, SimpleMem, A-MEM
2Fisher–Rao metric is the unique statistically principled retrieval metricČencov's theorem + Theorem 6.1Step 1
3Langevin dynamics on Poincaré ball provide self-organizing lifecycle with proven convergenceTheorem 6.3 (Fokker–Planck proof)Step 1
4Sheaf cohomology formally detects contradictions ($H^1 \neq 0$)§5.4 coboundary constructionStep 1
5Mathematical layers improve retrieval by +12.7 pp average, +19.9 pp on hardest tasksTable 5 (6 conversations, 832 questions)Steps 2, 3, 4
6Advantage grows with scale (cosine concentration barrier)Propositions 7.1, 7.2Step 2

§7 实现 cross-reference #

MIT license. Builds on SuperLocalMemory V2 infrastructure. 111 ACI containers for evaluation (2 vCPU + 4 GB each). SQLite with 21 tables, WAL mode. nomic-embed-text-v1.5 (768d) for Mode A, text-embedding-3-large (3072d) for Mode C. bge-reranker-v2-m3 for cross-encoder.

[实现未公开] — No specific repository URL for V3 codebase in the paper.

关键实现细节:

  1. Fisher metric "reduces to cosine" on benchmark data — the graduated ramp relies on $n_{\text{access}} \geq 10$ for full Fisher activation, but in single-pass benchmark evaluation all memories have $n_{\text{access}} = 0$. The headline theorem (6.1) is largely inactive during reported experiments; the measured +10.8 pp comes from the signal-magnitude variance heuristic in the initial ramp, not from converged Fisher estimates.
  2. Sheaf contributes only −1.7 pp on LoCoMo because the benchmark has few genuine contradictions. The theoretical machinery for contradiction detection is extensive but its practical value is demonstrated only weakly on current benchmarks.