NEWOne memory for every agent, bot and chat windowOne memory for every agent and botWhat’s new →
The SuperLocalMemory mascot examining the research
Open Access Research

SuperLocalMemory research. AI agent memory, open to scrutiny.

The governed-memory paper is publicly available on arXiv with a companion Zenodo archive, alongside three earlier arXiv preprints. These are preprints, not venue-reviewed publications, and product-release claims require separate release evidence.

4
Public research records
Zenodo + arXiv
Open-access archives
2026
Latest preprint
00
arXiv:2608.08253 Zenodo 21853302 2026 August 2026 · V4.0

SuperLocalMemory 4.0: The Governed Memory Operating System for AI Agents

Varun Pratap Bhardwaj, Garima Singh, and Arun Pratap Bhardwaj

This SuperLocalMemory 4.0 preprint describes a governed, local-first memory control plane for AI agents: policy-authorized writes, durable projection obligations, completion manifests, multi-scope memory, auditable operations, and the CLI, MCP, dashboard, and integration surfaces around the runtime.

Release evidence
  • The revised v2 preprint reports 2,199 of 2,200 scoped component properties upheld across eleven fault-injection scenarios. It discloses negative results and withdraws the earlier write-envelope overhead comparison.
  • Claims are scoped to component and mechanism measurements; they are not a generic end-to-end retrieval-accuracy benchmark.
  • The public preprint is available on arXiv:2608.08253, with Zenodo as the companion citable archive.
Public paper
Read the V4 paper on arXiv Companion archive on Zenodo Citation DOI: 10.5281/zenodo.21853302
01
arXiv:2603.14588 2026 March 2026

Information-Geometric Foundations for Zero-LLM Enterprise Agent Memory

Varun Pratap Bhardwaj

This preprint introduces information-geometric principles for agent memory and reports the published V3 LoCoMo evidence carried into V4.0. In the current product pipeline, dense candidate generation still uses cosine similarity while Fisher-derived terms inform later scoring and lifecycle behavior.

Key Contributions
  • Fisher-informed scoring layered onto candidate retrieval
  • Research evaluation of information-geometric retrieval and lifecycle techniques
  • 74.8% Mode A Retrieval across 10 conversations / 1,276 questions, with GPT-4.1-mini answer synthesis disclosed
  • 87.7% Mode C on Conv-30 / 81 questions, with cloud embeddings and GPT-4.1-mini disclosed
Full Paper
Read on arXiv 
@article{bhardwaj2026informationgeometric,
  title={Information-Geometric Foundations for Zero-LLM
    Enterprise Agent Memory},
  author={Bhardwaj, Varun Pratap},
  journal={arXiv preprint arXiv:2603.14588},
  year={2026}
}
02
arXiv:2603.02240 2026 March 2026

Privacy-Preserving Multi-Agent Memory with Bayesian Trust Defense Against Memory Poisoning

Varun Pratap Bhardwaj

This preprint evaluates privacy, Bayesian trust, search, graph, and multi-agent defenses against memory poisoning. It is not a lifecycle, compression, or million-memory capacity proof for the current release.

Key Contributions
  • Bayesian trust defense against low-trust memory writes
  • Privacy and multi-agent threat model
  • Historical search and graph experiments
  • Versioned research evidence, separate from current release proof
Full Paper
Read on arXiv 
@article{bhardwaj2026privacypreserving,
  title={Privacy-Preserving Multi-Agent Memory with Bayesian
    Trust Defense Against Memory Poisoning},
  author={Bhardwaj, Varun Pratap},
  journal={arXiv preprint arXiv:2603.02240},
  year={2026}
}
03
arXiv:2604.04514 2026 April 2026

The Living Brain: Biologically-Inspired Forgetting, Cognitive Quantization, and Multi-Channel Retrieval

Varun Pratap Bhardwaj

This preprint reports historical experiments on biologically inspired forgetting, cognitive quantization, and multi-channel retrieval. It does not prove a generic prompt-compression ratio or current V4.0 capacity.

Key Contributions
  • Biologically inspired forgetting experiments
  • Cognitive quantization research
  • Historical multi-channel retrieval evaluation
  • Release behavior requires separate current-artifact verification
Full Paper
Read on arXiv 
@article{bhardwaj2026livingbrain,
  title={The Living Brain: Biologically-Inspired Forgetting,
    Cognitive Quantization, and Multi-Channel Retrieval},
  author={Bhardwaj, Varun Pratap},
  journal={arXiv preprint arXiv:2604.04514},
  year={2026}
}
Evidence
LoCoMo · arXiv:2603.14588 Published V3 protocol carried into V4.0

Protocol-scoped LoCoMo results

The architecture evaluated in the V3 paper is carried into V4.0. These results retain their original models, answer-construction, question count, and conversation scope; they are not relabeled as a newly rerun 3.7 package benchmark.

  • 60.4% — Mode A Raw: 10 conversations, 1,276 questions, local embeddings/retrieval and zero-LLM answer construction.
  • 74.8% — Mode A Retrieval: 10 conversations, 1,276 questions, local retrieval with GPT-4.1-mini answer synthesis.
  • 87.7% — Mode C: Conv-30, 81 questions, text-embedding-3-large plus GPT-4.1-mini answer generation and judge.
  • +12.7pp — information-geometric layers across six conversations: 71.7% with mathematics versus 58.9% without.
Category detail

Mode A Retrieval: 72.0% single-hop, 70.3% multi-hop, 80.0% temporal, and 85.0% open-domain. Mode C reports 64.0%, 100.0%, and 86.0% respectively for single-hop, multi-hop, and open-domain in its Conv-30 scope.

SuperLocalMemory 4.0 — Zenodo 21853302 2603.14588 — V3 SLM: Information-Geometric Agent Memory 2603.02240 — Privacy-Preserving Multi-Agent Memory 2604.04514 — The Living Brain
BibTeX

Cite this work

The SuperLocalMemory 4.0 preprint is publicly archived on Zenodo; three prior papers are open access on arXiv. The current release citation is below.

Zenodo 21853302 — SuperLocalMemory 4.0 Preprint 
@misc{bhardwaj2026slmv4,
  title={SuperLocalMemory 4.0: The Governed Memory Operating System for AI Agents},
  author={Bhardwaj, Varun Pratap and Singh, Garima and Bhardwaj, Arun Pratap},
  year={2026},
  doi={10.5281/zenodo.21853302},
  url={https://zenodo.org/records/21853302}
}
Methodology

Why mathematics?

SLM combines inspectable heuristics with mathematical components. Component properties do not guarantee end-to-end recall, so the release keeps research and product evidence separate.

01 · The problem with cosine

Cosine and confidence answer different questions.

Cosine similarity is used for dense candidate generation. Stored-memory confidence is a separate assertion field. Later score enhancement may combine additional signals, but neither value is answer probability.

02 · Fisher-Rao respects probability

Component math has a bounded role.

Fisher-derived terms can inform later scoring when required state is available. Their mathematical properties do not establish retrieval correctness, calibration, or safety for the complete system.

See the mathematics and runtime boundary →
03 · Protocol disclosure matters

Scope before score.

The papers are public arXiv preprints, not venue-reviewed publications. Historical results are kept separate from current V4.0 release claims and must disclose answer construction, sample scope, and cloud assistance.

Read the preprint →
Mathematical foundation · Geometry, not guesswork

Mathematical foundations

Four public arXiv preprints describe versioned research experiments. In the shipped path, dense candidates use cosine similarity; Fisher-derived terms can inform later scoring. Product behavior is governed by release code, tests, and measured artifacts rather than a component theorem.

The boundary

Where information geometry fits.

Cosine similarity is the shipped dense candidate metric. It measures embedding direction and does not, by itself, encode the confidence stored with an assertion. That limitation does not make cosine invalid; it defines what later scoring stages may add.

Score meanings stay separate. Query-relative relevance, internal ranking utility, stored-memory confidence, trust, and answer confidence are different values. Score Contract v2 preserves those distinctions.

Fisher-derived research terms can influence post-candidate scoring when their required state is available. They do not replace every cosine path, and they do not turn rank into answer probability. Until a held-out calibrator is frozen, answer_confidence remains null.

Field Meaning V4 status
relevance_score query-relative relevance bounded 0..1
ranking_score internal ordering utility not a probability
memory_confidence stored assertion field separate from relevance
answer_confidence calibrated answer probability null / uncalibrated
Canonical response: score_contract_version: "2"
Calibration status is explicit. V4 does not present retrieval ordering as answer probability.
01 — Retrieval Metric
Fisher-Rao Distance · arXiv:2603.14588
dFR(p, q) = arccos( Σ √(pi · qi) )

Where p and q are probability distributions over memory confidence scores. The sum Σ is taken over all memory dimensions i. The arccos maps the result to angular distance on the unit sphere — the natural geometry of probability distributions under the Fisher information metric.

This is the geodesic distance on the statistical manifold — the shortest path between two probability distributions, measured along the curved surface they inhabit. Not the Euclidean shortcut through the void. The manifold curvature is determined by the Fisher information matrix, which encodes how much information each dimension carries. High-confidence dimensions curve the space more steeply. Low-confidence dimensions barely curve it at all.
  • Symmetric dFR(p, q) = dFR(q, p)
  • Triangle inequality dFR(p, r) ≤ dFR(p, q) + dFR(q, r)
  • Degenerate at identical distributions dFR(p, p) = 0
  • Bounded 0 ≤ dFR(p, q) ≤ π / 2
2603.14588 — V3 SLM: Information-Geometric Agent Memory →
02 — Lifecycle Model
Riemannian lifecycle research · arXiv:2604.04514
∇γ ∂tγ = 0     ·     Expp(v) = γ(1)

The preprint studies geometry-inspired forgetting, consolidation, and cognitive quantization. The equations describe the research model evaluated there; they are not a complete specification of every current runtime transition.

The released lifecycle also contains explicit policy, usage, recency, thresholds, and maintenance code. Those inspectable mechanisms—not this equation alone—define shipped behavior.

A mathematical property of one component does not guarantee end-to-end retention or recall. V4 lifecycle claims require update, supersede, archive, restore, forget, deletion, and stale-representation tests across the frozen artifact.
2604.04514 — The Living Brain →
03 — Compression Theory
Compression evidence boundary
H(X|Y) ≤ H(X)     ·     I(X;Y) ≥ 0

These are Shannon's fundamental information inequalities. H(X|Y) is the conditional entropy of prompt X given prior context Y — the irreducible information content that cannot be compressed away without information loss. I(X;Y) is mutual information — the quantity of information that X and Y share, which is always non-negative.

These inequalities describe theoretical limits; they do not prove a compression ratio, fidelity result, or provider saving for the shipped implementation. Product claims require a declared corpus, mode, tokenizer, round-trip test, latency measurement, and raw artifact.

Safe mode preserves JSON and code and may produce no reduction. Aggressive prose compression is separately opt-in and lossy. Shannon inequalities cannot be used to turn either behavior into a generic product-wide percentage.
  • Structured payloads (JSON, code) Measure reduction and round trip per corpus
  • Unstructured prose Aggressive path is opt-in and lossy
  • Byte-exact reversible on structured paths Verify byte invariants in release tests
compression — live example
# Check prompt compression on a structured payload $ slm compress --status --payload schema.json ↳ record input and output bytes/tokens for the supplied corpus safe path: verify structured-data invariants and byte round trip provider cache: measure separately from compression
See the shipped compression boundaries →
The field

Why information geometry?

Background · Amari & Rao, 1945–1985

A geometry built for probability.

Information geometry, developed by Shun-ichi Amari and C. R. Rao, is the study of probability distributions as geometric objects. The key insight: probability distributions do not live on a flat plane. They live on a curved manifold — the statistical manifold — where the natural notion of distance is the Fisher information metric, not Euclidean distance. Amari's 1985 monograph unified differential geometry and statistics into a single framework that lets you reason about distributions the same way classical geometry reasons about shapes.

Why it fits agent memory

Memories are distributions, not points.

The research asks whether probability-aware geometry can add useful signals beyond embedding similarity. In the current implementation, dense candidates still use cosine similarity and Fisher-derived terms may modify later scoring. Whether that improves retrieval is an empirical question for a frozen corpus and trace, not an automatic consequence of the theory.

Public Research

Four arXiv preprints.

The research claims on this page are traceable to public arXiv preprints. Read the methods, reproduce the results, and cite the work.

01 · Retrieval metric
dFR(p, q) = arccos( Σ √(pi · qi) )

Fisher-Rao Distance

Historical information-geometric retrieval experiments. Current dense candidate generation uses cosine; Fisher-derived terms may inform later scoring.

arXiv:2603.14588 →
02 · Trust and privacy
∇γ ∂tγ = 0  ·  Expp(v)

Bayesian Trust Defense

Historical evaluation of trust, privacy, search, graph, and multi-agent memory-poisoning defenses.

arXiv:2603.02240 →
03 · Forgetting and quantization
H(X|Y) ≤ H(X)  ·  I(X;Y) ≥ 0

The Living Brain

Historical forgetting, cognitive quantization, and multi-channel retrieval experiments. Not a generic prompt-compression proof.

arXiv:2604.04514 →
Explore the architecture →

Public preprints. SuperLocalMemory source under AGPL-3.0-or-later.

Every listed preprint is freely available on arXiv. The SuperLocalMemory source and packages are published on GitHub under AGPL-3.0-or-later. Optional models, providers, and third-party components retain their own licenses and terms.

GitHub All Papers on arXiv

Read the guidesBenchmark method and ablations · Cite SLM

★ Star on GitHub