Your infrastructure
Run Hanzo in your cloud, on your own servers, or fully offline.
Models, agents and one shared workspace for everyone in your company, in your cloud, on your servers or fully offline.
One platform, four places to run it. Choose one, or start in our cloud and move when you are ready.
Deployment
We run it for you. Start today. On Hanzo Base, your organization's data is its own database.
Where it runs01 / 04
Inside your VPC. Hanzo runs in your own cloud account, on your network.
Where it runs02 / 04
On premises. On hardware in your data center, beside the systems you already run.
Where it runs03 / 04
Fully offline. Sealed inside your network, with models served on machines you control.
Where it runs04 / 04
Our models' scores come from research runs anyone can open. Customer results come from the case studies of launches Hanzo ran.
Proof
Your own infrastructure, dedicated compute, and engineering support around the clock.
What you get
Run Hanzo in your cloud, on your own servers, or fully offline.
Train and host your own models, on your own data.
Engineering support around the clock, and hands-on onboarding to get your teams started.
Compute set aside for your organization alone.
An uptime guarantee and a security review, both in your contract.
Unlimited members, single sign-on, roles and an audit log.
A security review is part of every Enterprise contract. We answer your questionnaire against the code, and the engineers who wrote it walk your reviewers through it.
Security
Sign in through your own OpenID Connect provider. SCIM 2.0 creates, updates and removes accounts from your directory.
On Hanzo Base, each organization's data is its own database file, kept apart from every other organization. Identity comes only from a verified token.
Every request is recorded: who acted, from where, what they called and the status it returned.
Grant access by organization, workspace and project. Every account can add an authenticator app, or a code by SMS or email.
Zen 6 runs on one DGX Spark. Zen 6 Max wants a rack of Blackwell servers. What we measured, and what a rack is made of.
Your own GPUs
Zen 6 on one DGX Spark, as Hanzo measured it
A cold prompt of about that many tokens on a quiet box, measured Sep 20, 2026: NVIDIA DGX Spark (GB10), 121 GiB unified memory, shared with the benchmark's other jobs; zen6, NVFP4 with FP8 attention, weights 21.6 GiB (serve-spark-zen6.sh).
Refer customers, resell Hanzo under your own brand, or have Hanzo’s agency build it with you.
Partners
$999 a month
Your own AI agency, under your brand, on your domain.
Ask about White-label →From $4,999 a month
Agency, Forward Deployed, Embedded Pod: Hanzo’s agency ships the work with you, from one workstream to an embedded team.
Hanzo Agency →Apply to refer organizations to Hanzo. The terms are agreed with each affiliate we approve.
The affiliate programme →Hanzo runs in your cloud and on these. They are the clouds and the silicon Hanzo deploys on, not partners or customers.
Share a few details about your company. We come back with a plan: where Hanzo runs, what it costs and how you start.