# Chasing the Pareto Frontier of AI — Hanzo AI — Hanzo AI

> From SKULLY and open-source commerce, through Techstars '17, to 1,600+ open repositories and the sovereign AI cloud today — advancing AI safety, performance, equitable access, privacy, and ownership.

Techstars ’17

# Chasing the Pareto Frontier of AI

From SKULLY and open-source commerce, through Techstars ’17, to 1600+ open repositories and the sovereign AI cloud today — advancing intelligence, speed, deterministic safety, and equitable cost together.

[Get Started](https://hanzo.ai/pricing)[Our History](https://hanzo.ai/history)[Open Source](https://hanzo.ai/open-source)

### Antje Worring

Chief Executive Officer and co-founder

Leads Hanzo’s strategy, enterprise partnerships and sovereign AI deployments.

[Website ↗](https://antjeworr.ing)[X ↗](https://x.com/antjeworring)[GitHub ↗](https://github.com/zooqueen)[LinkedIn ↗](https://www.linkedin.com/in/antje-worring/)

### Zach Kelling

Chief Technology Officer and founder

Builds Hanzo’s AI cloud, its engine runtimes and its agent frameworks. A long-time open-source author.

[Website ↗](https://zeekay.io)[X ↗](https://x.com/zeekay)[GitHub ↗](https://github.com/zeekay)[LinkedIn ↗](https://www.linkedin.com/in/zeekay/)

Why We Build

## Moving the frontier without trading away safety, privacy, or ownership.

Most AI platforms ask you to pick two: raw capability, low latency, affordable cost, or control over your data. Chasing the Pareto frontier means advancing every axis at once through better systems engineering and open research.

### Safety in AI

Deterministic verification over guesswork

Generative models drift and hallucinate when left unchecked. We pair frontier reasoning with Kai calibrated discrete decision models, formal policy gates, and sandboxed execution so autonomous agents stay bounded in production.

[Safety research](https://hanzo.ai/research)

### Frontier Performance

Bare-metal C, CUDA, and Metal runtimes

Speed is an architectural choice. From Hanzo Engine’s zero-Python inference kernels and Kai’s decisions, which write no text, to Enso’s learned multi-model routing, every layer is engineered to maximize intelligence per millisecond.

[Live benchmarks](https://hanzo.ai/benchmarks)

### Equitable Access

Frontier capability without hyperscaler tax

Breakthrough AI cannot belong only to trillion-dollar incumbents. By routing each task to the most efficient model, publishing open weights, and pricing compute fairly, we put startups, researchers, and builders on equal footing.

[Transparent pricing](https://hanzo.ai/pricing)

### Open Source

1600+ public repositories since 2014

Our engines, agent SDKs, MCP servers, UI kits, and Hanzo Improvement Proposals (HIPs) are developed in the open. Anyone can inspect the code, verify how decisions are made, fork the stack, and contribute.

[Browse repositories](https://hanzo.ai/open-source)

### Privacy by Architecture

No training without your say

Business and API accounts are never used to train shared models unless an administrator or a written agreement says so, and a personal account only if you turn it on. Organizations sit in isolated tenant databases with scoped HKDF-SHA256 keys, local-first CLI workflows, and strict gateway identity stripping.

[Security & trust](https://hanzo.ai/trust)

### Ownership & Sovereignty

Cloud, private VPC, on-prem, or air-gapped

If you cannot move your weights, memory, and runtime onto your own hardware, you are renting your intelligence. Hanzo runs identically on our global cloud, inside your VPC, or completely air-gapped on your own GPUs.

[Explore platform](https://hanzo.ai/platform)

Proven Track Record

## Built with founders, scaled to public markets.

From SKULLY and open-source commerce through Techstars ’17 to Damon Motors and Triller listing on Nasdaq in 2024, Hanzo has engineered sovereign commerce and AI systems for hundreds of venture-backed companies. Read the complete 2014–2026 archive on our history page, or start building today.

100+Startups built & funded

$1B+Client revenue generated

2Went public on Nasdaq (2024)

[Start Building](https://hanzo.ai/pricing)[Read Our Full History](https://hanzo.ai/history)[Customer Stories](https://hanzo.ai/customers)

Operating Principles

## The Zen of Hanzo

The principles the company argues from when a decision is close.

### Enable the Underdog

Enable others to win. Democratize technology so small businesses and startups can compete with industry giants.

### Data-Driven Everything

Let data be your guide. Embrace analytics and AI as the foundation of decision-making.

### Open Innovation

Be open, share often. Openness accelerates growth—both for the company and its community.

### Adaptability

Pivot with purpose. There is no failure, only feedback. Move quickly and strike in new directions when opportunity calls.

### Customer-Centric Growth

Treat each client as a long-term partner and guide them to success.

### First Principles Thinking

See the invisible, do the impossible. Boil problems down to their essence and solve them in fundamentally better ways.

[Explore the full Zen manifesto](https://hanzo.ai/philosophy)

Open Source & Deterministic Safety

## Verifiable code. Calibrated decisions.

Opaque black-box platforms ask you to trust what you cannot inspect. Hanzo pairs public source code and open specifications with calibrated decision bounds and full infrastructure sovereignty.

### 1600+ Open Repositories

Auditable infrastructure from kernel to interface

Since open-sourcing our commerce architecture in 2014, we have published over 1600 public repositories. From C/CUDA/Metal inference engines to agent compilers, MCP tools, and cross-platform UI kits, our stack is built to be read, audited, and extended.

Zero Proprietary Lock-In: Export and run models, agents, and state on your own hardware without rewriting application code.

HIP Standards: Core protocols and architectures published openly through Hanzo Improvement Proposals.

Shared Ecosystem: MCP servers, skills, UI primitives, and inference runtimes built in public with contributors worldwide.

[Explore Open Source Ecosystem](https://hanzo.ai/open-source)

### Deterministic Safety Bounds

Calibrated decisions, not guesswork

Unconstrained token generators drift in high-stakes workflows. Hanzo puts calibrated decisions (Kai), policy checks and isolated sandboxes in front of any autonomous agent that touches production data or billing.

Policy Governance: Policy checks stop a tool call or a change your policy refuses before it is sent.

Sovereign & Air-Gapped: Run inside your own VPC or offline cluster with zero external telemetry or third-party relays.

Calibrated Confidence: Kai returns calibrated probability distributions and defers cleanly when evidence is insufficient.

[Read Our Safety Research](https://hanzo.ai/research)

Proof of Work: Recent Commits by Our Core TeamPublic Activity

1600+ repositories[github.com/hanzoai](https://github.com/hanzoai)

hanzo-community/daz_synapse-10[Initial commit](https://git.hanzo.ai/hanzo-community/daz_synapse-10/commit/b45b5c3b9639378b882a88e7ea8925f1a8bee682)

b45b5c3Nov 2, 21:38 UTC

hanzo-community/vshanbhag_analytics-dashboard-3[Initial commit](https://git.hanzo.ai/hanzo-community/vshanbhag_analytics-dashboard-3/commit/c711a8e7c7788a60a8036b85993386b291c4d257)

c711a8eNov 2, 21:30 UTC

hanzoai/commerce[billing: one wire settles one invoice, and a wire plan ends what it replaces](https://github.com/hanzoai/commerce/commit/6ea958112b733f85e937f4d0c1966f9cd8a4b4e4)

6ea9581Nov 2, 20:27 UTC

hanzoai/iam[users: rule who makes an org admin, and make a revoke take effect now](https://github.com/hanzoai/iam/commit/e48704c976233f5b2748d16f556f220c42e71c31)

e48704cNov 2, 20:07 UTC

hanzoai/commerce[limit: a bound staff set on an org or a plan is stored beside the plans](https://github.com/hanzoai/commerce/commit/218534f5924524af47eecdbf7ab28659ca546c21)

218534fNov 2, 20:06 UTC

hanzoai/social[Add sleep-900 test ok to SMOKE.txt](https://github.com/hanzoai/social/commit/3131c4e04325bffff10c915551b3ea637ab4fd35)

3131c4eNov 2, 19:35 UTC

Every line written, reviewed, and shipped in the open by Hanzo engineers.[View Open Source Dashboard](https://hanzo.ai/open-source)

Frontier Models & Research

## Original architectures, engines, and published research.

Rather than wrapping third-party endpoints, Hanzo designs decision models, semantic routers, bare-metal inference engines, and open research on verifiable agent safety.

### Models & Runtimes

[View Model Catalog](https://hanzo.ai/models)

#### Kai

Decision Intelligence

Calibrated discrete decision model for autonomous agent loops. Factorized discrete choice, calibrated probability distributions, and deterministic policy governance.

No output tokens · Calibrated confidence[Details →](https://hanzo.ai/kai)

#### Zen Multimodal

Foundation Models

Multimodal reasoning models across text, code, vision, speech, and structured output, built for both low-power edge devices and sovereign GPU clusters.

High token throughput · Edge & Cloud[Details →](https://hanzo.ai/zen)

#### Hanzo Engine

Inference Runtime

Bare-metal C, CUDA, and Metal inference engine with speculative decoding and NVFP4/FP8 kernels, removing Python runtime overhead on production hardware.

Zero Python runtime · C / CUDA / Metal[Details →](https://hanzo.ai/engine)

#### Enso Router

Semantic Routing

Learned semantic router and multi-model coordinator (HIP-0510) that sends each prompt to a model by accuracy and cost.

Learned arbitration · Accuracy & cost[Details →](https://hanzo.ai/enso)

### Selected Papers & Technical Writing

[Papers](https://hanzo.ai/research)[Blog](https://hanzo.ai/blog)

[Research PaperOct 2026 #### Kai: Calibrated Discrete Decision Intelligence for Autonomous Agent Loops Factorized discrete choice, calibrated distributions, and deterministic policy governance for mission-critical autonomy.Read publication](https://papers.hanzo.ai/kai-decision-model/)

[Research PaperSep 2026 #### Linkage, Not Similarity: What Long-Term Agent Memory Has to Retrieve Why cosine vector similarity falls short on multi-session recall and how linkage graphs preserve causal context over time.Read publication](https://papers.hanzo.ai/hanzo-context-memory/)

[Technical BlogOct 2026 #### Your AI is Planning To Kill You: Deterministic Safety Bounds Why unconstrained generative loops fail under adversarial pressure, and how Kai enforces mathematical safety bounds.Read publication](https://hanzo.blog/2026-10-16-your-ai-is-planning-to-kill-you/)

[Technical BlogOct 2026 #### Optimizing for Headless Commerce in the Era of Agentic LLMs What it takes for infrastructure to be readable, verifiable, and payable by autonomous agents with clear human authorization.Read publication](https://hanzo.blog/headless-commerce-agentic-llms/)

Careers at Hanzo

## Push the Pareto frontier with us.

We hire systems programmers, AI safety researchers, and product engineers who care about open source, equitable access, privacy, and sovereign ownership. High autonomy, zero bureaucracy, and code that ships in public.

### Distributed Systems & Inference

Scaling bare-metal C/CUDA/Metal kernels, multi-node GPU clusters, and sovereign inference clouds.

### AI Safety & Formal Verification

Advancing calibrated discrete decision models (Kai), verification graphs, and deterministic policy bounds.

### Open Agent Compilers & Tools

Building MCP infrastructure, linkage memory, and local-first developer tools used across the ecosystem.

[Explore Open Roles](https://hanzo.ai/careers)[Meet the Team](https://hanzo.ai/team)[Browse Code on GitHub](https://hanzo.ai/open-source)