qb-compiler — free + open source

Start with a free check.Scale to governance.

QubitBoost is an open-core trust layer for quantum execution. It starts with qb-compiler, a free, open-source CLI in the IBM / Qiskit ecosystem that runs circuit preflight and ObservableGate, our QEC decoder-input correctness audit. Every check becomes a receipt: the check is free, the trusted receipt is the product.

Install the CLI today, catch a bad run before you pay for it, and grow into signed receipts and drift governance only when your workload needs them. Validated on real superconducting, trapped-ion and neutral-atom hardware data.

Dublin, Ireland · Stelor Evolution Ltd · IBM Quantum Ecosystem partner

Hardware-validated decoder adaptation. Click any card to read the deep-dive notebooks with reproduction code.

201Kper-shot labelled records
4validated vendors (IBM, Rigetti, IonQ, IQM)
8backends
38.8%SafetyGate protected-logical reduction, IBM Torino (postselection, ~71% throughput)

qb-compiler

Apache 2.0 · PyPI

Our calibration-aware compiler is in the IBM Quantum Ecosystem under Error Mitigation + Circuit Optimization.

pip install qb-compiler · routes around bad qubits at submit time, not after the run.

OPEN-CORE: START FREE, SCALE TO GOVERNANCE

Most teams meet QubitBoost as a free CLI. The same checks become signed receipts in Pro, and the full governance stack in Enterprise. You only pay when the trust matters.

FREE · OPEN SOURCE

qb-compiler

Circuit preflight + ObservableGate decoder-input audit. JSON receipts, CI exit codes. Apache 2.0, listed on the IBM Quantum Ecosystem.

  • pip install qb-compiler
  • Preflight: VIABLE / CAUTION / DO NOT RUN
  • qbc dem-audit (ObservableGate)
  • Local JSON receipts + CI exit codes
Get the CLI

QUBITBOOST PRO

The receipts

The check is free; the trusted receipt is the product. Signed and stored receipts, batch reports, shared dashboards, CI policy bundles.

  • Signed + stored receipts
  • Batch reports across runs
  • Shared dashboards
  • CI policy bundles
See Pro

ENTERPRISE

Governance

The full stack: DriftRadar, SafetyGate, Result Passport, ChemGate, ShotValidator and the decoder adaptor. Runtime integration, SLA, audit, indemnity.

  • DriftRadar + SafetyGate
  • Result Passport provenance
  • Decoder adaptor pilot
  • Runtime integration + SLA
Request a pilot
CAPABILITYFREEPROENTERPRISE
Circuit preflight / viability
ObservableGate DEM audit (qbc dem-audit)
JSON receipts (local)
CI exit codes
Signed receipts
Stored / shared receipts + dashboards
Batch reports
CI policy bundles
DriftRadar / SafetyGate / governance stack
Decoder adaptor (pilot)
Runtime integration + SLA + indemnity

Free forever. Pro from €29/user/mo. See pricing →

NEW · qb-compiler 0.8.0

OBSERVABLEGATE: AUDIT YOUR DECODER'S INPUT BEFORE YOU TRUST ITS OUTPUT

A stim Detector Error Model error mechanism carries both detector targets and logical-observable masks. If a DEM-to-matrix step merges mechanisms by detector signature alone, it can collapse detector-identical, logical-distinct mechanisms and erase the logical frame, inflating the reported logical error rate. ObservableGate flags it and offers an observable-preserving canonicalization. CI-safe: PASS / WARN / FAIL.

RUN THE AUDIT

$ qbc dem-audit model.dem
  ObservableGate: PASS
  groups: 482  mixed (H==, L!=): 0

$ qbc dem-canonicalize model.dem -o safe.dem
  merged exact (detector, observable) duplicates
  kept distinct logical masks separate

HONEST SCOPE

Standard surface, repetition and Bivariate-Bicycle / Gross codes audit PASS, the qLDPC codes teams actually run are observable-mask safe.

ObservableGate is a preflight invariant that catches the edge case at the DEM boundary before it reaches your decoder. It is not a claim that any decoder or library is broken.

THE PROBLEM

Fault tolerance is gated on one number: the decoder's logical error rate. Every decoder reads an error model that is wrong from the moment calibration ends, and a wrong model means more physical qubits per logical qubit.

Decoders assume the error model handed to them is correct. It is not: the device drifts the moment calibration ends, so the model the decoder consumes is stale and the logical error rate it reports is worse than it needs to be. Everyone is racing to build better decoders. Nobody is fixing the model those decoders read before they run. That is the gap we fill.

Stale error model

We correct the model your decoder reads before it decodes.

High qubit overhead

Lower logical error rate means fewer physical qubits per logical qubit.

Risk of regression

A built-in guard means we never make your decoder worse.

WHY NOW

Three events in the last six months turned quantum from research into billed infrastructure.

NOV 2025

IBM Nighthawk ships, $96/min billing.

120 qubits on Utility tier. Quantum compute is now a line item on an enterprise invoice. Nobody is checking the answers.

FEB 2026

Software stack consolidates.

Classiq closes $110M Series C, Riverlane $75M. The compiler and decoder layers are claimed. The performance layer that improves those decoders, by correcting the model they read, is still open.

APR 2026

qb-compiler in IBM Quantum Ecosystem.

Our calibration-aware compiler is listed alongside the other algorithm vendors. Apache 2.0 telemetry surface, paid SDK on top. We are in the stack, not adjacent to it.

THE QUANTUM ERROR CORRECTION PERFORMANCE LAYER

QubitBoost lowers your decoder's logical error rate. The same platform improves runtime efficiency and result assurance across the run, so the whole job costs less and stands up to audit. Every number below is from real hardware unless marked simulation.

DECODING PERFORMANCE

lower logical error rate, any decoder

We correct the error model your decoder reads, before it decodes, so the decoder you already run reports a lower logical error rate. No decoder swap, no retraining.

A built-in guard means your decoder is never made worse.

2 platforms

Validated on real neutral-atom and superconducting hardware data

MWPM + BP+OSD

Improves the decoders teams already run

No swap

Slots in front of the decoder you already have

Never worse

Built-in no-harm guard

RUNTIME EFFICIENCY

Spend fewer shots and fewer evaluations for the same answer.

44.2% fewer VQE evaluations on IBM Fez caffeine (n=8, deterministic), reaching Hartree-Fock energy.

RESULT ASSURANCE

Know when to trust a run, and prove it afterward.

A signed Result Passport on every run: drift state, trust score, which gates fired, what the policy decided.

SEE THE PASSPORT FOR YOURSELF

One JSON document per QPU run. Trust score, drift state, which gates fired, what the policy decided.

KEEP YOUR DECODERWE LOWER ITSERROR RATE.

Every run also ships a signed, machine-readable record of what we did to the error model and how the decoder fared, so you can audit the gain on real hardware data.

RESULT PASSPORTv1.0
ACCEPTSLA: GOLD
0.000Trust Score
Backend Mood
FRESH
Drift State
STABLE
Gate Scores
DriftRadarObservatory,
LiveGateDoom abort,
SafetyGateQEC trust,
ChemGateEval savings,
ShotValidatorSyndrome check,
Provenance
Passport IDtpp-a3f8c2e1...9e02
Backendibm.fez (IBM)
Calibration Age1h 10m
Circuit Depth47
Qubits7
Total Shots4,096
Noise Exposure0.31
Decoherence Ratio0.22
Passport Excerpt
signature: hmac-sha256:9f3a...c7e1qubitboost.io

“Does this result have a passport?”

Cut wasted quantum computeCertify the results you keepValidated on real Google Willow

HOW IT WORKS

01

SUBMIT

Point us at your QEC job and the decoder you already run. No swap, no rewrite.

02

ADAPT

We correct the error model your decoder reads, before it decodes. Your decoder stays exactly as it is.

03

DECODE

Your decoder runs on the corrected model and reports a lower logical error rate. A built-in guard means we never make it worse.

Four SLA tiers (Bronze, Silver, Gold, Platinum) match the trust level your workload demands. See pricing →

FOR DEVELOPERS

PYTHON SDK

# pip install qubitboost-sdk
from qubitboost import QubitBoost

qb = QubitBoost(api_key="...")

# run your job, keep your decoder and circuits as they are
result = qb.submit(circuit, backend="ibm_fez", shots=4000).wait()

print(result.status)
print(result.counts)

The Decoder Adaptor runs as a private pilot on your own data. Request access.

CLI

$ pip install qb-compiler
$ qbc preflight circuit.qasm -b ibm_fez
$ qbc analyze circuit.qasm -b ibm_fez
$ qbc compile circuit.qasm -b ibm_fez --receipt

WHERE WE SIT

We are not a decoder, not an SDK, not a control stack. We are the vendor-neutral performance layer that lowers the logical error rate of the decoder you already run, on any of them.

 RIVERLANEQ-CTRLCLASSIQQUBITBOOST
What they sellHardware decoderControl + suppressionSoftware + compilerGovernance + trust
ModelFPGA integrationVendor SDKIDEVendor-neutral SaaS
The open layerOwns the decoderOwns controlOwns the IDEOwns the model the decoder reads
Funding$75M Series C$113M total$110M Series CPre-seed, capital-efficient

201K SHOTS. FOUR VENDORS. EIGHT BACKENDS.

The optimization and governance numbers come from real superconducting and trapped-ion hardware (IBM Heron, Rigetti Ankaa, IonQ, IQM), drift-stratified, with the methodology file on record. The decoder result is validated separately on real neutral-atom and superconducting data.

Five earlier claims were inflated. We retracted them publicly in Q1 2026. The discipline that made us retract is the discipline we sell.

IBM Fez

156-qubit Heron

Rigetti Ankaa-3

QEC at distance d=7

4 vendors, 8 backends

Cross-platform validation

IonQ Aria, Quantinuum H2

Trapped-ion, signed Result Passport

ONE API, EVERY VENDOR

IBM Quantum

LIVE

AWS Braket

ADDING LIVE DATA

Rigetti

ADDING LIVE DATA

IonQ

ADDING LIVE DATA

IQM

ADDING LIVE DATA

Quantinuum

ADDING LIVE DATA

ACADEMIC ACCESS

Free for universities and research labs.

$500/month of real hardware time. All backends. Publish your results, cite us if the numbers helped, argue with us if they did not. See pricing.

REQUEST PILOT ACCESS

Two weeks on your workloads, your backends. Tell us which QPU and we will point the pilot at it. No commitment, no credit card, no deck.

Fill in name and email first

No spam. We'll reach out to discuss your use case and set up the pilot.