Skip to content

ZB COMPUTE

ZB Compute is the compute infrastructure that runs the models, the agents and your finished solution.

We are starting with 64 GPUs, in a cloud or inside a country's border.

ZB COMPUTE / EXECUTION LAYER
ORCHESTRATIONMaestroAssign tasks and check results
GPU inferenceRun open models
Agent computersRun code, tools, and tests
JOB RECORDOutputs, checks, and cost

RUNTIME

One AI project needs two kinds of compute, and ZB Compute runs both.

Open models run on GPUs sized to the work in front of them, and each agent works in its own computer, with its own files and tools.

MANAGED GPU INFERENCE

Open model inference.

Open models run on GPUs sized to the work in front of them.

Hosting
Open models on managed compute
Capacity
Model serving scales with demand
Return
Generated text or structured fields
MODEL EXECUTIONRUNTIME SCHEMATIC
INPUTModel requestInput from a Maestro task
GPU INFERENCE
OUTPUTModel responseText or structured output

ISOLATED AGENT COMPUTERS

A sealed computer for every agent.

Each agent works in its own computer, with its own files and tools.

Tools
Code, browser, files, and tests
Boundary
Separate runtime environment per agent
Return
Files, tool output, or test results
AGENT EXECUTIONRUNTIME SCHEMATIC
ENVIRONMENT AAgent computer A
Code executionFiles
Isolated environment
ENVIRONMENT BAgent computer B
BrowserTests
Isolated environment
↓
Job record Files, checks, and cost

DEPLOYMENT

You choose where your work runs.

On any cloud, in your private cloud, or inside your border.

MANAGED CLOUD

On any cloud

Compute on demand, in the regions you choose.

ZB COMPUTE ENVIRONMENT
GPU inferenceAgent execution

PRIVATE OR IN-COUNTRY

In your private cloud or inside your border

ZB Compute inside your own cloud account, or on compute that stays in your country.

YOUR DEPLOYMENT BOUNDARY
In-scope modelsAgents and job files
Explore enterprise deployment ↗

DEPLOYMENT PLANNING

Model placement
List model versions and which calls can go to hosted providers.
Agent tools
List the files, browsers, code, and external services the agent uses.
Data location
Identify where inputs, job files, and outputs need to stay.

QUESTIONS

ZB Compute FAQ.

What does serverless mean on ZB Compute?

ZB Compute manages the compute for inference and agent workloads. You bring the job, and ZB Compute provides the environment that runs it.

Can one job use both open and frontier models?

Yes. ZB Model can coordinate open-model inference on ZB Compute with calls to frontier models through the providers that host them. Each task can use the model suited to the work.

What actually improves after launch?

ZB Model can test a change to the workflow’s instructions, tools, or handling of exceptions. It can check the revised workflow against the missed case and replay cases that already work. Separately, ZB Model learns how to assign and check future work from completed task outcomes.

What does the job record capture?

Task commits in a Git-compatible repository capture what the task produced, the models it called, which checks passed, and what it cost.

Can I run a workload in my cloud or country?

We scope private and in-country deployments with enterprise teams. Bring the intended environment, the data that must stay there, and the models and agents your job uses.

GET STARTED

Bring your next AI project to ZB Compute.

Tell us what you want to build and where it needs to run.