Skip to content
BiVelio joins the NVIDIA Inception ProgramNVIDIA Inception Program BiVelio is part of the ElevenLabs Grants Program

The Research library

Everything we publish, in one place.

Papers with their PDF, the algorithms already running in production and the articles that explain them. Search by title, topic or author.

Papers
4
Algorithms
3
Articles
16

Papers

Papers and preprints

The full papers as PDF, each with a page that sums it up: authors, date, results and how to cite it.

PreprintOctober 2026PDF

Caleta: An Evidence-Oriented Trust Layer for Reproducible Local Large Language Model Inference

Joan-Marc Fisa

Technical preprint by Joan-Marc Fisa (BiVelio Inc, October 2026): Caleta, an evidence layer around llama.cpp that makes explicit which artifact ran, with which configuration and how it was measured. Not peer reviewed. On Zenodo, CC BY 4.0.

Summary and results Read the paper (opens in a new tab) Download the PDF (opens in a new tab)
How to cite it

Joan-Marc Fisa (2026). Caleta: An Evidence-Oriented Trust Layer for Reproducible Local Large Language Model Inference. BiVelio Inc, October 2026. https://doi.org/10.5281/zenodo.23763881

Engineering draftOctober 2026PDF

XMRMesh: Offline Monero Payments over Bluetooth Low Energy Mesh Networks

BiVelio Inc

BiVelio Inc preprint BV-XMRM (version 1.7, October 2026): a proposed, unvalidated engineering draft for relaying signed Monero transactions over a Bluetooth Low Energy mesh until a connected peer submits them. On Zenodo, CC0.

The paper behind BV-XMRM · XMRMesh

Summary and results Read the paper (opens in a new tab) Download the PDF (opens in a new tab)
How to cite it

BiVelio Inc (2026). XMRMesh: Offline Monero Payments over Bluetooth Low Energy Mesh Networks. BiVelio Inc, October 2026. https://doi.org/10.5281/zenodo.23762733

Revised preprintOctober 2026PDF

BiVelio Shield: Architecture, Deployment Boundaries, and Reproducible Detection Benchmarks

BiVelio Privacy Research Team

Corrected BiVelio Shield technical report: architecture and deployment boundaries, 339 catalogue entries including 204 credential types, and reproducible cooperative, adversarial, and NLP benchmarks. Company preprint, not peer reviewed.

Related reading for BV-PRGA · BiVelio Privacy Gateway

Summary and results Download the PDF
How to cite it

BiVelio Privacy Research Team (2026). BiVelio Shield: Architecture, Deployment Boundaries, and Reproducible Detection Benchmarks. BiVelio, October 2026. https://bivelio.com/en/research/papers/bivelio-shield

PaperAugust 2026PDF

BV-SALA Savings Layer: Verifiable Measurement of Cost Savings in Large Language Model Applications

Adrián Vidal

BiVelio paper (August 2026): cost-savings evaluation using provider-reported usage, versioned list prices, four ledgers and cluster-bootstrap intervals, distinguishing live double-runs from counterfactual replay.

The paper behind BV-SALA · BiVelio Savings Layer

Summary and results Download the PDF
How to cite it

Adrián Vidal (2026). BV-SALA Savings Layer: Verifiable Measurement of Cost Savings in Large Language Model Applications. BiVelio Inc, August 2026. https://bivelio.com/en/research/papers/bv-sala-savings-layer

Algorithms in production

Optimization before the model, not inside it

We do not replace OpenAI, Anthropic, DeepSeek or Gemini. We sit in front of them: we better prepare what you send them so you spend less, expose less and retrieve what matters most. Two algorithms, one shared principle. Each solves a different problem in the layer that connects your company with AI models.

In research

Engineering drafts

Proposed designs with partial evidence that are not yet validated. We publish them so they can be reviewed and replicated, not as products.

BV-XMRM · Engineering draft

XMRMesh

A proposed, unvalidated design for an already-signed Monero transaction to travel over a Bluetooth Low Energy mesh until someone with connectivity submits it.

How it works

Articles

Research articles

Every piece stands on its own: context, intuition, formalization and results.

Optimization BV-PRGA: inside the privacy gateway 7 min Optimization BV-SALA: inside the LLM cost layer 11 min Systems 40% of agentic AI dies without an Autonomy Rate 12 min Agents 95% of AI pilots fail: it's not the model 12 min Systems A single agent fails: governed AI in 3 layers 9 min Foundations Adding a copilot is not an operating model 11 min Systems AI agents: it's not the model, it's the architecture 9 min Foundations AI Workforce OS: not a suite, but an operating model 10 min Systems Buying more software won't make you an autonomous company 11 min Agents How to govern AI agents: authority, not output 10 min Foundations Human-in-the-loop: 5 components, not an approval button 11 min Foundations RPA automates tasks. A process operating system governs operations 10 min Agents RPA vs workflow automation vs AI agents: layers, not rivals 11 min Foundations Your ungoverned AI agent is a liability, not a milestone 11 min Agents Why your RAG fails and the graph recovers double 8 min Retrieval Your RAG's problem isn't the model, it's the order 8 min