hyperlakeDiscuss a deployment ↗
FAQ

Questions about Hyperlake, answered.

What Hyperlake is, where it runs, and how it runs managed open-source stacks for your team and your agents.

The basics

What is Hyperlake?

Hyperlake is the AI cloud for your team and your agents, on your cloud. It runs managed open-source stacks (Kubernetes, ClickHouse, Postgres, Trino, Iceberg, open models and agent sandboxes) in your own AWS or Google Cloud account, with LakeGate securing every query to your databases and data engines. You get managed customer-cloud workloads in your own account. Control-plane data and stored credentials may be processed elsewhere, as described in section 5.8 of the Terms.

What does sovereign AI mean at Hyperlake?

It means the customer controls where the AI stack runs and who can reach it. Customer-cloud workloads run within infrastructure the customer or their client controls, with identity, policy and audit controls. Hyperlake's control-plane data and stored credentials may be processed elsewhere, as described in section 5.8 of the Terms.

Who is Hyperlake for?

Enterprise and platform teams that want to own their AI environment, and AI product companies and service providers that deploy into many client environments and need repeatable installs, client ownership and one operating approach.

The platform

Where can Hyperlake run?

On AWS and Google Cloud today. Azure, OVH, private cloud and on premises are coming soon. The foundation runs on Kubernetes, which keeps services portable and manageable from one environment to the next. Governance and operations stay the same wherever the stack lives.

What does the Hyperlake platform include?

Stacks, each installed in one run on Foundry, Hyperlake's managed Kubernetes, with LakeGate in front of their databases and data engines. They cover secure data access, federated SQL, real-time analytics, lakehouses, application databases, private models and retrieval, agent sandboxes, ML model serving and Physical AI pipelines, and new stacks are added every week. See every stack.

Which data engines and model tools does Hyperlake use?

Engine choice follows the workload: Iceberg and Trino for lakehouse analytics, PostgreSQL for applications, ClickHouse for fast analytics, MongoDB for documents, Redis for state, Kafka for streams, Qdrant or Milvus for vectors, and graph engines for relationships. Open models are served through KServe. Lifecycle support depends on the engine and deployment.

How do people and AI agents get access to data?

Through one identity path instead of shared, long-lived database passwords: sign-in through an OAuth/OIDC proxy connected to your identity provider, a validated JWT that carries scoped user or workload identity, OPA policy decisions, and enforcement and logging at integrated access points.

Can I manage AI infrastructure in natural language?

That is the experience Hyperlake is building toward. Infrastructure is callable through an application, a CLI and MCP-compatible tools: a team states an intent, inspects the proposed plan, then approves, deploys, observes and maintains it under the organization's policies. Specific integrations and automations depend on the deployment.

Costs and use cases

How does Hyperlake reduce costs?

You run open-source engines in your own cloud account, at your own cloud prices and discounts, instead of paying for their commercial editions. Through reuse, less integration work and capacity that follows the workload. We don't promise a fixed percentage. Teams deploy complete stacks from reusable patterns, eligible inference endpoints can scale down when idle, and Hyperlake adds no compute markup. Scale-to-zero depends on the serving mode and whether the underlying compute can also scale down.

Why build for AI agents as the main user?

Because agents don’t log off. Persistent agents run queries, model calls, code and jobs around the clock, so the work, the access requests and the bill grow with every action. Platforms priced per call or per seat charge for each of those actions, while capacity you own costs the same until you add more. That’s why Hyperlake runs the stack in your own account and puts LakeGate in front of your databases, so agents get the same governed access as your people.

When does owning AI capacity beat paying per use?

When agents keep calling models, querying data and running jobs across many users or clients. Per-use charges can rise with each action, while well-used owned capacity can make extra actions cost relatively little until more capacity is needed. The curves can cross either way, so Hyperlake recommends making the ROI case per workload.

What can teams build on Hyperlake?

Example deployments include a repeatable Physical AI development environment, a private agentic cloud, a governed data and intelligence foundation, and autonomous multi-agent systems with purpose, memory and limits. These are examples of workloads Hyperlake can enable, not a fixed product catalog. See the use cases.

Does Hyperlake support Physical AI and robotics?

Yes. Hyperlake can bring simulation, orchestration, training, data and model services together in the customer's cloud, integrating tools such as NVIDIA OSMO, Omniverse, Isaac Sim and Isaac Lab, subject to licensing and validated integration, so teams can repeat the environment across sites.

Getting started

How do I get started with Hyperlake?

Tell the Hyperlake team what you want to deploy, whose environment it must run in, and which data and policies it needs; the platform is shaped around the real workload. Existing users sign in at app.hyperlake.cloud. See what to bring and how to start.

Start with a workload. Build the environment around it.

What do you need to deploy? Whose environment does it have to run in, and what does it connect to? Tell us.