Hanzo AI

Research Papers

77 papers

Hanzo AIResearch Papers

Owned, self-hostable AI and post-quantum cryptography — sovereign infrastructure for regulated industries, critical infrastructure, and government. Technical papers, measured case studies, and formal proofs.

Hanzo AI for Security

Automated, Agentic Security Auditing at Estate Scale

How an agentic stack discovers and remediates security debt across sprawling microservice estates. Case study: engaged for a narrow scope, the team uncovered a 200+ microservice estate, surfaced 93 findings (22 critical) across 127+ repositories plus 780+ dependency vulnerabilities, and remediated them to zero actionable while consolidating to 3 binaries — with security built into the OSS stack via KMS, IAM, post-quantum identity, signed actions, and an immutable on-chain audit trail.

Zach Kelling
Sovereign AISecurityAgentic AIAudit
PDFSource

Hanzo AI for Compliance

An OSS Cloud Built to Pass the Audit

The sovereign OSS stack produces audit evidence by construction — centralized IAM, KMS secrets custody, an immutable signed on-chain audit trail, post-quantum identity, and machine-checked proofs — mapping natively to SOC 2 Trust Services Criteria, NIST 800-53 control families, FINRA/SEC recordkeeping, and the defense ATO/RMF process, so compliance becomes evidence collection rather than months of manual work.

Zach Kelling
Sovereign AIComplianceSOC 2ATO
PDFSource

hanzo.ai vs. claude.ai

Owned, self-hostable AI versus a centralized cloud chat

Owned, self-hostable AI versus a metered cloud chat — on deployment and control, not benchmark scores. Includes a measured productivity case: on a regulated build where every engineer had the same frontier model (Claude), the client's own audit found one engineer on Hanzo's agentic stack shipped more than three contractor pods combined at ~250x-900x lower cost per line — the differentiator being the owned agentic stack, not the model. Covers data residency, air-gapped operation, provenance, post-quantum security, and cost at scale.

Zach Kelling
Sovereign AIPost-QuantumOpen Weights
PDFSource

Agentic Engineering at Regulated Scale

Building a Post-Quantum, FHE-Native Securities Platform

Founding CTO Zach Kelling was the #1 Anthropic user by token count worldwide in 2025 — and that throughput built a real FINRA/SEC-approved, SOC 2 securities exchange on post-quantum cryptography, a native-GPU blockchain, and fully-homomorphic encryption. The build consolidated 200+ microservices into 3 compiled binaries (3.07M → 1.06M lines) with a small team, and the FHE layer — computing on data you are not permitted to see — applies across regulated industries, from healthcare to finance to government.

Zach Kelling
Sovereign AIAgentic AIFHEPost-Quantum
PDFSource

Cloud Economics of a Sovereign OSS Stack

A Regulated Capital-Markets Migration Case Study

Cut cloud spend ~90% — from ~$35K/mo to ~$3.5K/mo (~$380,000/yr) — migrating a regulated capital-markets platform off a sprawling public-cloud footprint (200+ microservices, 185 projects on GCP) onto a consolidated, self-hostable open-source stack on owned hardware. Decomposes where the savings come from (no metered model markup, no hyperscaler margin, efficient transport, consolidation) and shows the same mechanism applies to any organization carrying a metered-API plus hyperscaler-sprawl cost structure.

Zach Kelling
Sovereign AICloud EconomicsSovereign StackOSS
PDFSource

Confidential Computing by Default

An FHE-Native Chain and Agent Runtime

A design for making fully homomorphic encryption the default mode of computation across the two layers that normally demand plaintext: a blockchain's state and rules, and an autonomous agent's actions. State stays encrypted on-chain, policies are evaluated over ciphertext, and agents act on data they are never able to read, with a candid assessment of what is production today versus an active research frontier.

Zach Kelling
Sovereign AIFHEConfidential ComputingPost-Quantum
PDFSource

ZAP: Zero-Copy Application Protocol

For Contested, Edge, and Bandwidth-Limited Military Environments

A high-performance binary serialization and RPC framework achieving 2.2 ns parse latency with zero allocations, 11.5M transactions per second, and 91% memory reduction versus JSON-RPC. Includes native post-quantum transport security (ML-KEM-768 + ML-DSA-65), decentralized identity via W3C DIDs, threshold agent consensus, and mesh discovery, analyzed for contested electromagnetic environments and tactical links.

Zach Kelling
Sovereign AIProtocolPost-QuantumEdge
PDFSource

Full-Spectrum Cryptographic Architecture

FIPS 203/204/205 Compliant Design for Information Dominance

A cryptographic architecture implementing all three NIST post-quantum standards (ML-DSA, ML-KEM, SLH-DSA) integrated into a production blockchain consensus system with hybrid classical/quantum security and sub-second finality. Details six threshold MPC schemes over a zero-copy wire protocol, GPU-accelerated operations, an FHE virtual machine, and a key management system with Shamir secret sharing, backed by 17 formal cryptographic proofs.

Zach Kelling
Sovereign AIPost-QuantumFIPSCryptography
PDFSource

Machine-Checked Proofs for Post-Quantum Systems

Automated Verification for Distributed Cryptographic Systems

A formal verification suite for a post-quantum distributed system spanning consensus, threshold cryptography, FHE, and cross-chain messaging: 50 machine-checked Lean 4 proofs, 18 cryptographic security reductions, 12 protocol-correctness proofs, and 7 DeFi invariant proofs, plus symbolic execution, fuzz targets, and NIST KAT validation. Identifies the critical gaps and presents a phased roadmap for closing them.

Zach Kelling
Sovereign AIFormal VerificationLean 4Post-Quantum
PDFSource

Agentic AI for Autonomous Cyber Operations

Multi-Domain Intelligence Fusion and Tactical Decision Support

An agentic AI framework of 83 specialized agents and 15 orchestrators for autonomous cyber operations, intelligence fusion, and tactical decision support across a control plane, an agent SDK with native MCP integration, and a ZAP-based consensus protocol with post-quantum authentication. Agents execute 200--300 sequential tool calls and run fully disconnected on edge hardware using quantized Zen models.

Zach Kelling
Sovereign AIAgentic AICyber OperationsIntelligence
PDFSource

Edge-Deployable AI/ML Framework

For Disconnected Tactical Environments

An edge-deployable AI/ML framework for disconnected, degraded, and bandwidth-limited environments, combining a Rust-native inference engine for 15+ text and 16+ vision architectures with aggressive 2--8 bit quantization across hardware from laptops to embedded ARM. A family of 18 models provides capability tiers from nano-scale edge inference to frontier reasoning, operating fully offline with sandboxed execution and no cloud dependencies.

Zach Kelling
Sovereign AIEdge AIQuantizationOffline
PDFSource

Multi-Level Security and Cross-Domain Solutions

Homomorphic Policy Evaluation and Zero-Trust Fabric

A cross-domain solution where data-transfer policies are evaluated on encrypted content using FHE so the guard never observes plaintext, integrating a zero-trust overlay with identity-first micro-segmentation, an FHE policy engine, Shamir-based key management, and org-scoped identity. Cross-domain transfers are cryptographically mediated and produce immutable audit records anchored to a post-quantum blockchain.

Zach Kelling
Sovereign AICross-DomainFHEZero-Trust
PDFSource

Post-Quantum Resilient Mesh Networking

For Maritime and Tactical Edge Operations

A post-quantum resilient networking architecture for maritime, tactical edge, and DDIL environments, integrating the Reticulum Network Stack as a native consensus transport over LoRa, packet radio, serial, and mesh topologies without centralized infrastructure or DNS. All sessions use hybrid post-quantum key exchange and authentication, with a zero-trust overlay and sub-second consensus finality over mesh networks.

Zach Kelling
Sovereign AIMesh NetworkingPost-QuantumMaritime
PDFSource

Quantum-Resistant Counter-Exploitation

Confidential Compute and Coercion-Resistant Communications

A suite of counter-exploitation technologies providing quantum-resistant secure messaging at 526,000 messages per second with forward secrecy, FHE with threshold decryption, post-quantum anonymous ring signatures with linkability, zero-knowledge privacy pools, and a multi-agent adversarial framework. Together these prevent adversary exploitation of cryptographic, network, and AI systems in contested environments.

Zach Kelling
Sovereign AICounter-ExploitationPost-QuantumFHE
PDFSource

Quantum Key Distribution and Quantum Sensing Integration

For Post-Quantum Blockchain Consensus and Threshold Cryptography

An architecture integrating quantum key distribution hardware and quantum sensors into a production post-quantum blockchain stack. A triple-hybrid key derivation combines classical Diffie-Hellman, lattice encapsulation, and physics-layer QKD secrets so link security holds if any one of three independent foundations holds. Quantum sensors feed cryptographically attested measurements on-chain via a sensor oracle with dual threshold signatures.

Zach Kelling
Sovereign AIQKDQuantum SensingPost-Quantum
PDFSource

Automated Verification and Validation Framework

With Digital Twin Environments

An integrated verification and validation framework combining continuous observability, behavioral analytics, ML pipeline orchestration, digital twin environments, and formal cryptographic proofs. Telemetry across 100+ AI providers feeds an OLAP backend, workflows orchestrate LLM evaluation, and a network simulation environment with 38 virtual machines enables mission rehearsal and system qualification with statistical decision support.

Zach Kelling
Sovereign AITest & EvaluationDigital TwinObservability
PDFSource

One Native Stack for Private, Continuously-Learning AI

Inference at the Bandwidth Wall, On-Device QLoRA, One Engine

A synthesis paper: train and serve collapse into a single Rust engine with one tensor type, one quantization core, and one device backend — eliminating the format-break tax of training in PyTorch, exporting to GGUF/ONNX, and serving on a third engine. Formalizes a unified bandwidth-budget theorem and QLoRA bit-exactness, with every claim sourced to a measured ROCm campaign on consumer hardware. Scrupulously scoped: claims no speed win over NVIDIA and no zero-leakage guarantee.

Zach Kelling
EngineInferenceOn-DevicePrivacy
PDFSource

Native ROCm Inference on a Consumer RDNA3.5 APU

Reaching llama.cpp Decode Parity via a Unified 1-bit-to-Full Quant Core

A measurement-driven engineering campaign that took native-ROCm LLM inference on a consumer AMD Ryzen AI Max+ 395 "Strix Halo" (Radeon 8060S, gfx1151) from the slowest of five backends to decode parity (0.98×) and 0.86× prefill versus llama.cpp on the same GPU. A lane-strided decode kernel reaches 245.8 GB/s — 97% of the memory-bandwidth wall. Documents negative results in detail and derives the physical ~25 tok/s ceiling that bounds both engines.

Zach Kelling
EngineROCmInferenceBenchmarks
PDFSource

Edge LLM Inference Across Commodity Accelerators

Measured AMD, NVIDIA, and Apple Performance

Measured single-GPU inference for one 8B dense transformer at Q8_0 across three commodity edge accelerators — AMD "Strix Halo", NVIDIA GB10 "Grace Blackwell", and Apple M4 Max — comparing the Hanzo engine to llama.cpp on identical model and metric. Two findings: decode parity on all three backends, and the fastest box flips by workload (NVIDIA wins prefill, Apple wins decode ~2.5×). Includes a full reproducibility appendix with exact hardware, commands, and git SHA.

Zach Kelling
EngineInferenceEdgeBenchmarks
PDFSource

Native Training in the Hanzo Engine

One Engine, One Quant Format, One Backend

The architecture for training language models natively inside the Hanzo Rust inference stack, eliminating the conversion boundary between a Python training stack and a separate serving engine. Shows the QLoRA primitive already exists in the engine's reverse-mode autograd, and formalizes QLoRA parameter/memory accounting plus additive LoRA-soup composition. Unusually candid: an entire section names the stack's own non-functional stubs to replace.

Zach Kelling
EngineTrainingQLoRAOn-Device
PDFSource

Continuously-Learning Private AI

Self-Improvement and Privacy in the Hanzo Native Stack

A private continuously-learning model as three loops: an inference loop that serves and captures its own interactions, a curation loop that filters/redacts/de-duplicates into safe training data, and a training loop that folds it back into cheap swappable adapters. Formalizes a privacy invariant (no PII token enters the training set under a sound gate) and bounded-forgetting (frozen base → zero base drift), and is honest about model-collapse and delta-leakage failure modes and which components are still research.

Zach Kelling
EnginePrivacyContinual LearningOn-Device
PDFSource

Economical Serving and the hanzo.ai Platform

A Bandwidth-Optimal Unified Train-and-Serve Runtime

The serving economics, platform architecture, and on-device thesis that follow from a native Rust train-and-inference engine whose decode kernel runs at the memory-bandwidth wall. Derives a cost law (tokens/$/s ∝ B / (q·N_active)) showing quantization — not bigger accelerators — is the primary cost lever, and proves paged-attention tenant concurrency and a structural egress bound (raw data never crosses the wire). Honestly marks train/personalize/federate as stub/roadmap.

Zach Kelling
EngineEconomicsPlatformOn-Device
PDFSource

Decentralized Semantic Optimization (DSO)

Byzantine-Robust Prior Aggregation

Decentralized learning protocol with Byzantine fault tolerance and three-layer semantic storage architecture for multi-agent coordination.

Hanzo AI Research
DSOByzantine RobustDecentralized Learning
PDFSource

Hamiltonian Market Maker (HMM)

Oracle-Minimal Pricing for AI Compute Marketplace

A novel AMM design for heterogeneous compute resources with provably-stable pricing via Hamiltonian invariants. Enables oracle-minimal economic settlement for AI inference and training workloads.

Hanzo AI Research
EconomicsDeFiCompute MarketplaceAMM
PDFSource
HIP-004 SpecificationHanzo Network

Hanzo Network Whitepaper

L1 Blockchain for Decentralized AI Compute

Comprehensive whitepaper describing Hanzo Network's Layer-1 blockchain infrastructure, integrating HMM for compute pricing and supporting PoAI consensus for quality verification.

Hanzo AI Team
BlockchainInfrastructureNetworkArchitecture
PDFSource

Hanzo Network Architecture

Decentralized AI Compute with Post-Quantum Security

Network architecture with post-quantum cryptography and physics-inspired LLM routing for decentralized AI compute.

Hanzo AI Team
ArchitecturePost-QuantumLLM Routing
PDFSource

Jin: Unified Multimodal AI Architecture

Vision-Language, Audio, and Cross-Modal Understanding

Unified architecture for vision-language, audio, and cross-modal understanding in a single multimodal framework.

Hanzo AI Research
MultimodalVision-LanguageAudioArchitecture
PDFSource

Model Context Protocol Server Architecture

Context Management and Tool Execution for AI Systems

Context management and tool execution for AI systems via the Model Context Protocol with 260+ integrated tools.

Hanzo AI Research
MCPContext ManagementTool Use
PDFSource

Hanzo Agent SDK

Orchestration, Tool Use, and Memory Systems

Building AI agents with multi-agent coordination, tool use, planning, and persistent memory systems.

Hanzo AI Research
AgentsSDKTool UseMemory
PDFSource

Hanzo ML: Production ML Framework

Model Serving and A/B Testing

Enterprise ML operations with model versioning, feature stores, and inference infrastructure for production deployment.

Hanzo AI Research
MLOpsModel ServingA/B Testing
PDFSource

Identity NFTs for AI Agents

Capability Tokens for Autonomous Agents

Token standard for autonomous AI agents with economic agency, identity, and capability delegation.

Hanzo AI Research
NFTAI IdentityCapability Tokens
PDFSource

Hanzo API Gateway

Unified Commerce API Architecture

API gateway architecture for unified commerce operations and integrations with 100+ provider support.

Hanzo AI Team
APIGatewayCommerce
PDFSource

Hanzo Checkout

Frictionless Payment Processing

Modern payment processing architecture for e-commerce with frictionless checkout flows.

Hanzo AI Team
PaymentsCheckoutCommerce
PDFSource

The Hanzo SDK Ecosystem

Multi-Language Commerce Integration

TypeScript, Python, Go, and Rust SDK architecture for commerce integration.

Hanzo AI Team
SDKMulti-LanguageIntegration
PDFSource

Hanzo LLM Gateway

Unified Proxy for 100+ AI Providers

Unified LLM proxy supporting 100+ providers with intelligent routing, cost optimization, and observability.

Hanzo AI Research
LLMGatewayAI ProvidersRouting
PDFSource

Hanzo Operative: Computer Use for AI

Browser and Desktop Automation for AI Agents

Computer use framework enabling AI agents to operate browsers and desktops autonomously with vision-based interaction.

Hanzo AI Research
Computer UseAutomationAI AgentsVision
PDFSource

Hanzo Chat: Multi-Model AI Interface

14 Zen Models with MCP Tool Integration

AI chat platform supporting 14 Zen models and 100+ third-party providers with MCP tool integration.

Hanzo AI Research
ChatMulti-ModelMCPInterface
PDFSource

Hanzo Flow: Visual Workflow Builder

Visual AI Pipeline Orchestration

Visual workflow builder for AI pipeline orchestration with drag-and-drop composition.

Hanzo AI Research
WorkflowVisualOrchestrationPipeline
PDFSource

Hanzo Platform: PaaS Infrastructure

Platform-as-a-Service for AI Workloads

Platform-as-a-Service infrastructure for deploying and managing AI workloads at scale.

Hanzo AI Research
PaaSInfrastructureDeploymentCloud
PDFSource

Hanzo Candle: Rust ML Framework

High-Performance Tensor Operations in Rust

Rust-based ML framework with tensor operations, neural network layers, and GPU acceleration.

Hanzo AI Research
RustML FrameworkTensorGPU
PDFSource

Hanzo ACI: AI Chain Infrastructure

Blockchain for Decentralized AI Operations

Blockchain infrastructure for decentralized AI compute, model verification, and inference consensus.

Hanzo AI Research
BlockchainAIDecentralized ComputeConsensus
PDFSource

The Zen AI Model Family

Democratizing AI Through Efficient Architecture

Technical overview of the full Zen model family and architecture whitepaper covering 600M to 1T+ parameters.

Hanzo AI Inc · Zoo Labs Foundation
ZenOverviewArchitectureModel Family
PDFSource
Model HubDocumentation

Zen-Omni: Multimodal Text Model

Multimodal Text Understanding and Generation

Multimodal text understanding and generation across modalities in a unified architecture.

Hanzo AI Inc · Zoo Labs Foundation
ZenOmniMultimodalText
PDFSource

Zen-Next: Flagship Model

Next-Generation Frontier Model

Next-generation flagship model technical report covering architecture innovations and benchmark results.

Hanzo AI Inc · Zoo Labs Foundation
ZenFlagshipNext-GenModel
PDFSource

Zen-Nano: Mobile and IoT Intelligence

Ultra-Efficient Models for Edge Deployment

Ultra-efficient models for mobile and IoT deployment with minimal resource requirements.

Hanzo AI Inc · Zoo Labs Foundation
ZenNanoMobileIoT
PDFSource

Zen-Eco: Consumer Hardware Models

Optimized for Consumer-Grade Hardware

Models optimized for consumer-grade hardware with excellent quality-to-compute ratio.

Hanzo AI Inc · Zoo Labs Foundation
ZenEcoConsumerEfficient
PDFSource

Zen-Artist: Text-to-Image Generation

High-Quality Image Synthesis

Text-to-image generation model with high-quality synthesis and artistic control.

Hanzo AI Inc · Zoo Labs Foundation
ZenArtistText-to-ImageGeneration
PDFSource

Zen-Designer-Thinking: Visual Reasoning

Visual Reasoning and Analysis

Visual reasoning and analysis for design tasks with chain-of-thought design understanding.

Hanzo AI Inc · Zoo Labs Foundation
ZenDesignerVisual ReasoningAnalysis
PDFSource

Zen-Reranker: 7680-Dimensional Embeddings

Semantic Optimization with BitDelta Compression

High-dimensional embeddings with BitDelta compression for semantic search and retrieval optimization.

Hanzo AI Inc · Zoo Labs Foundation
ZenRerankerEmbeddings7680-Dim
PDFSource

Zen-3D: 3D Scene Generation

Text-to-3D and Image-to-3D Synthesis

3D scene generation from text and image inputs with high-fidelity synthesis.

Hanzo AI Inc · Zoo Labs Foundation
Zen3DGenerationSynthesis
PDFSource

Zen-World: World Model

Predictive World Simulation

Predictive world model for simulation and planning in complex environments.

Hanzo AI Inc · Zoo Labs Foundation
ZenWorld ModelSimulationPrediction
PDFSource

Zen-Foley: Sound Effects Generation

AI-Powered Foley and Sound Design

AI-powered foley and sound effects generation for film, games, and media production.

Hanzo AI Inc · Zoo Labs Foundation
ZenAudioSound EffectsFoley
PDFSource

Zen-Musician: Music Generation

Composition and Arrangement AI

AI music generation model for composition, arrangement, and production.

Hanzo AI Inc · Zoo Labs Foundation
ZenMusicGenerationComposition
PDFSource

Zen-Voyager: Embodied AI Navigation

Autonomous Exploration and Navigation

Embodied AI model for autonomous exploration and navigation in physical and virtual environments.

Hanzo AI Inc · Zoo Labs Foundation
ZenEmbodied AINavigationExploration
PDFSource

Zen Technical Architecture

MoDE Architecture and Training Methodology

Technical architecture report covering the MoDE (Mixture of Diverse Experts) design and training methodology.

Hanzo AI Inc · Zoo Labs Foundation
ZenArchitectureMoDETechnical Report
PDFSource

Zen-Reasoning: Chain-of-Thought Models

Advanced Mathematical and Logical Reasoning

Chain-of-thought reasoning model for advanced mathematical and logical problem solving.

Hanzo AI Inc · Zoo Labs Foundation
ZenReasoningCoTMathematics
PDFSource

Zen-Medical: Clinical AI Models

Medical Diagnosis and Clinical Decision Support

Clinical AI model for medical diagnosis and clinical decision support systems.

Hanzo AI Inc · Zoo Labs Foundation
ZenMedicalClinicalDiagnosis
PDFSource

Zen-Voice-Clone: Voice Synthesis

Zero-Shot Voice Cloning and TTS

Zero-shot voice cloning and text-to-speech synthesis with high fidelity.

Hanzo AI Inc · Zoo Labs Foundation
ZenVoiceTTSCloning
PDFSource

Zen-Video: Video Generation

Text-to-Video and Video Editing

Text-to-video generation and video editing model with temporal consistency.

Hanzo AI Inc · Zoo Labs Foundation
ZenVideoGenerationEditing
PDFSource

Zen-Director: AI Film Production

Scene Composition and Cinematography AI

AI-powered film production model for scene composition and cinematography.

Hanzo AI Inc · Zoo Labs Foundation
ZenFilmDirectorCinematography
PDFSource