Deployment Options

Qoro ships one stack in three deployment models. Code written against Divi, Qiskit, or PennyLane runs unchanged on all three. The differences are where the compute runs, who operates it, and which components are present. Commercial details are on enterprise.qoroquantum.net; this page covers the technical side.

	Solo	Maestro Enterprise	Dedicato
Components	Divi → Composer → Maestro fleet / QPU Interfaces	Maestro engine and GPU plugin	Divi → Composer → Maestro fleet / QPU Interfaces, plus the Composer distributed-QC toolkit
Compute	Qoro’s shared CPU and GPU simulators, open to every account	Your CPUs and NVIDIA GPUs	Dedicated CPU and GPU capacity in a single-tenant VPC
Operated by	Qoro	Your HPC or IT team	Qoro
Entry point	Composer API (QoroService in Divi) and the dashboard	Python, C/C++, CLI, or your scheduler	Composer API and dashboard, on a private endpoint
QPU access	Bring your own IBM, IQM, or Braket credentials	Simulation only	Your own credentials, or vendor access managed by Qoro with budget caps
Simulator selection	Maestro Auto (automatic per circuit) or a fixed method	You choose the method	Maestro Auto or a fixed method

Solo: Qoro’s multi-tenant cloud

The full cloud path described in System Architecture, running on infrastructure Qoro shares between customers.

• Access. Users sign up at dash.qoroquantum.net, create an API token, and point Divi’s QoroService at it.
• Simulators, CPU and GPU. Every account can use both Qoro simulator clusters: a CPU cluster (16 vCPUs, 62 GB RAM) and a GPU cluster (NVIDIA T4 and A10). A job chooses one by name with JobConfig(simulator_cluster=...), and the dashboard’s Clusters page lists the names. GPU runs need no Maestro GPU licence, because the licence is part of the platform. Workers scale with queue load. By default the platform chooses a simulation method for each circuit; a job can fix the method through a Maestro configuration instead.
• QPUs. QPU access is also open to every account. Users register a QPU system with their own vendor credentials (IBM, IQM, or AWS Braket) in the dashboard, then target it by name with JobConfig(qpu_system=...). Credentials are stored encrypted and passed to the interface worker only for that job. The vendor bills hardware time to the user’s own account.
• Billing. Credits are charged per second of run time at the target’s rate. A job is checked against the user’s balance before it is accepted.

Maestro Enterprise: on-premises simulation

The Maestro engine without the cloud platform, installed on the customer’s hardware.

• Form factors. Python package, C/C++ library, maestro command-line tool, a JSON request API, and a local server daemon for multi-user sites. It is also available as a container.
• GPU. The GPU plugin adds CUDA/cuQuantum backends for statevector, MPS, tensor-network, Pauli-propagation, density-matrix, and MPO simulation. Distributed plugins spread one statevector across several GPUs in a node, or across nodes over CUDA-aware MPI.
• Scheduling. Jobs go through the site’s own scheduler. Slurm integration goes through QRMI and the Maestro local server (see HPC & Deployment).
• Data. Circuits and results never leave the site. The GPU plugin’s licence check is the only outbound call, and air-gapped licensing removes it. The open-source CPU engine has no licence check.

Dedicato: managed single-tenant platform

The Solo stack deployed for a single organisation in its own VPC, either in Qoro’s cloud or in the customer’s AWS, Azure, or Google Cloud account. Qoro deploys and operates it.

• Isolation. Compute runs in private subnets with no inbound internet access, and no infrastructure is shared with other customers.
• Capacity. Simulator tiers, GPU types, and concurrency limits are sized to the customer. Tiers can scale to zero when idle.
• QPU access. Customers either bring their own vendor credentials or use vendor access that Qoro manages for them. Spending caps are enforced by the platform.
• Distributed QC. Higher tiers add the Composer engine’s distributed quantum computing toolkit: circuit distribution across networked QPUs, circuit cutting, and resource estimation (see Composer).