The AI cloud you can run yourself. hanzo.network is hanzod nodes that spawn the same unified cloud binary we run in production — OSS AI cloud in one binary, BYO GPU or K8s, mine on any device, free for public and open-source workloads.

The AI cloud you can run yourself

Hanzo is one Go binary. The cloud binary — CGO_ENABLED=0, statically linked — embeds the whole platform: all 67 capabilities across the eight movements, mounted as subsystems, served behind one contract at /v1/<capability>. The same binary that answers api.hanzo.ai in production is the binary you run on your laptop with one command.

hanzo.network is what happens when you point that binary at a chain instead of a single host. It is a network of hanzod nodes that spawn and power that same unified cloud binary. The network is the cloud — not a hosted service you rent, but a substrate anyone can join and run.

curl hanzo.sh | sh

This is the story specified in HIP-0127 — V8 · Open Edition: one binary, one contract, a network that is the cloud. Nothing here is a hosted tier you are locked into — every capability is the same open-source binary, whether it runs on our nodes or yours.

One binary, everywhere

The naive way to ship a cloud is dozens of services, each its own image, each its own deploy. Hanzo rejects that. The application cloud is one static Go binary with no cgo, no cross-toolchain build, no runtime dependencies — which is exactly what makes "curl | sh, run the whole cloud anywhere" true.

On your machine

One binary, no dependencies. Runs the entire cloud locally for development or a single-tenant deployment.

On your GPU or K8s

Bring your own accelerator or cluster. The binary uses what you give it — no vendor hardware lock-in.

On the network

hanzod nodes spawn the same binary and coordinate over the chain. Join as a provider or a consumer.

Because the surface is one uniform contract — /v1/<capability> over HTTP + ZAP — a consumer never needs to know or care which node, which process, or which language answered. The route is the identity.

GitHub for compute

Public and open-source workloads run free on public nodes. Think of it as GitHub for compute: your open work is hosted at no cost on the shared network, the same way open source is hosted for free today. Private and metered workloads are market-priced.

Public + OSS → free

Open-source and public workloads run at no cost on public hanzod nodes. Openness is the pricing tier.

Private → market-priced

Private workloads are metered and priced by the market of providers — you pay for what you use, to whoever runs it.

Mine on any device

Any device can be a provider. A hanzod node contributes its compute to the network and is paid at market price for the work it does. Confidential workloads run inside NVIDIA TEE / Confidential Compute enclaves, so a provider serves work it cannot see and a consumer runs work a host cannot inspect.

Any device

Laptop, workstation, server, or datacenter GPU — if it runs the binary, it can mine.

Confidential by default

NVIDIA TEE/CC enclaves keep provider and consumer mutually blind. Trust the math, not the operator.

Market-priced

Providers are paid the going rate for the compute they contribute. The network clears the price.

The language seam is a wire, never FFI

Hanzo is Go and Rust: the cloud is Go, while the engine, the dev agent, and the ML kernels are Rust. They compose over ZAP — a network or local socket — never over FFI. Each stays a clean single-language binary; they exchange ZAP frames, they never link each other's code.

This is deliberate. FFI (cgo/uniffi) is easy — one process — but it complects two toolchains and destroys the single static binary. The wire seam is simpler: independent, composable, language-free at the boundary. Don't fuse — bridge. See Architecture & Philosophy for the full reasoning.

Run it yourself

Install the binary

curl hanzo.sh | sh

One command installs the hanzo CLI — a ~15 MB Rust binary that speaks to any live cloud, whether it is production, your laptop, or a self-host.

Log in

hanzo login

Authenticate with Hanzo IAM — the same identity works against every cloud, everywhere.

Use every capability

hanzo chat "Explain the network"     # Intelligence
hanzo deploy ./my-app                # Platform
hanzo secrets set API_KEY=…          # Identity & Trust

The CLI is a client, never the server — it never imports the cloud. It talks ZAP (or its SDK) to a live cloud. Point it at prod, your laptop binary, or a customer self-host; the commands are identical.

The Network