Loading docs
Select quantum providers, inspect hardware fit, and keep credentials private while routing workflows.
Provider routing connects the workflow objective to available execution infrastructure. It should show fit, readiness, and provider context without exposing credentials.
The route decision belongs to the workflow record because it affects how results should be interpreted later.
Section 01
Open provider selection from the workflow surface when possible. That keeps the route decision close to the circuit and generated code.
If the provider is not ready, fix setup in the private provider area. Do not paste tokens into docs, comments, run notes, or screenshots.
Automatic target selection considers live discovery status, operation support, qubits, shots and queue depth. An unavailable discovery service prevents a fresh hardware recommendation. Organization hardware grants, access terms, member limits and concurrent run capacity are checked again when the run is admitted.
IBM gate runs use the current client-side Runtime Sampler/Executor. Canonical measurements retain their original classical-bit positions, including unused bits; the execution cap defaults to 60 seconds. Historical saved jobs remain readable. Backend capabilities and account entitlement are checked before submission.
For IBM expectation values, choose Circuit expectation values under Hardware program. Define named Hermitian observables as real sums of Pauli strings; the rightmost letter acts on logical qubit 0. QFlow compiles the current circuit and maps observables to the selected backend layout before using the client-side Runtime Estimator/Executor. Terminal measurements are excluded by the explicit readout contract; intermediate measurements, classical conditions, resets and simulation noise channels are rejected.
An Estimator shot setting is a total budget shared across measurement configurations, not a returned state count. Error mitigation and twirling are disabled for this bounded execution path; the explicit execution limit is at most 60 seconds. Results and CSV retain expectation values, standard errors and any provider-reported configuration shots or target precision. Access to a cataloged IBM device still requires current account entitlement.
GPU inventory is separate from verified GPU execution. Container visibility and CUDA device order must be verifiable before a GPU is shown as available; the CPU runner remains the supported execution path until a GPU execution profile has passed its own runtime proof.
Section 02
Marketing pages show recognizable providers and SDKs so teams understand the operating layer. The actual credentials and provider status remain in authenticated surfaces.
Keep this boundary clear in team documentation: public evidence can name a provider route, but credentials stay sealed.