What is a Quantum Mesh Network?

An AI‑driven routing layer that load‑balances across the providers we route to today, optimizes cost and latency, and automates failover.

How does NL→Quantum work?

Our compiler converts plain‑English goals into optimized circuits in under 500ms with validation and provider selection.

How are costs reduced?

Provider arbitrage routes each run to the cheapest machine that meets its fidelity and latency constraints, with auditable quotes.

Tracking Gate Fidelity Drift with Dynamic Calibration

QuantufAI Labs · 2025-01-10

Quantum hardware is not stationary. Gate fidelities drift between calibration cycles, qubits get retuned, and the device you benchmarked on Monday is not quite the device running your circuit on Friday. Any routing decision made against last week's numbers is a decision made against a device that no longer exists.

What QHTA reads

QHTA 2.0 — our hardware testing and attestation layer — consumes the calibration telemetry that hardware vendors publish for their devices: reported gate fidelities, coherence times, and readout errors, refreshed as vendors recalibrate. On top of that raw feed it maintains attestation records, so a routing decision can point at the exact calibration snapshot it was made against.

The word "reads" is doing deliberate work in that sentence. These are the vendors' own reported calibration figures, relayed with provenance — not fidelity measurements QuantufAI performed independently. What we add is the cross-vendor view: the same telemetry, normalized, timestamped, and tracked over time so that drift becomes visible instead of anecdotal.

Why drift tracking matters for routing

A router choosing between backends needs more than a spec sheet — it needs to know what each device looks like right now, and how stable that picture has been. A backend whose reported two-qubit fidelity oscillates between calibrations is a different risk than one that holds steady at the same average. Tracking the telemetry over time is what turns a snapshot into a signal.

That is also why attestation is bundled with telemetry in the same layer: when a workload's placement is questioned after the fact, the answer should be a record, not a recollection.

The honest boundary

We publish what the telemetry layer can substantiate: which vendors' devices we read calibration data from, when it was last refreshed, and what it said. We do not publish achieved-fidelity benchmarks of our own, because real-hardware execution is currently gated and we refuse to print numbers we have not measured.

Reflects integration and availability, not benchmarked results — job execution is currently gated. Access varies by provider: Quantinuum is currently plan-limited to its emulator.

— QuantufAI Labs