QEC Decoding for Python
One library for many decoder families on the same check matrix: Union-Find, exact MWPM, BP-OSD and Relay-BP for qLDPC codes, on CPU and GPU, with Stim and Sinter integration. Every correction is checked against its syndrome, and commercial licences verify offline, so it runs air-gapped.
Built for hardware teams, qLDPC researchers and academic labs. Free for academic and non-commercial research; commercial plans from $79/month with a 14-day money-back guarantee.
pip install qector-decoder-v3==1.0.1
python -c "from qector_decoder_v3 import BlossomDecoder; print('QECTOR OK')"QECTOR publishes contracts, methodology, and artifacts rather than portable speed numbers. Read the evidence policy before generating a local measurement.
A Clear Path From Code to Evidence
The public release, validation contract, and commercial delivery path are explicit by design.
qector-decoder-v3 1.0.1
The live PyPI release supports Python 3.9 through 3.14 with 18 binary wheels across Windows, Linux, and macOS, each with build provenance.
Correctness before speed
Every correction is checked against H c = s (mod 2). Measurements stay workload-scoped, artifact-hashed, and separate from universal performance claims.
Signed, traceable licensing
Commercial fulfillment requires a live Stripe event, issues a v2 Ed25519 token, and records independent billing and license delivery states.
Read the license activation procedure →Works with your existing stack
pip install stimpip install sinterpip install qiskit-qecExecutable Verification for the 1.0 Line
Hashed-lock reproducible install, CERT_STRICT gate, and Lean/Coq kernel evidence - verifiable against Reference Manual v1.0.0.
Hashed-lock & CERT_STRICT
Proof runs are hashed-locked and gated by CERT_STRICT - no universal benchmark figures, only workload-scoped artifacts.
DOI 10.5281/zenodo.21941046
Normative manual with syndrome-faithfulness foundations, every decoder backend, and API stability tiers. All proofs trace to this document.
Kernel-checked proofs
Executable suite checks GF(2) invariants and logical cosets with Lean/Coq kernels - evidence, not claims.
From Installation to Validation
QECTOR integrates directly into your existing QEC workflow with minimal friction.
Install
Install the 1.0.1 wheel in a supported Python 3.9-3.14 environment alongside your existing Stim or PyMatching workflow.
pip install --upgrade "qector-decoder-v3>=1.0.1,<2"Decode
Access 17 concrete Rust-accelerated decoder configurations through a unified Python API. Workbench also exposes AutoDecoder and Auto Router as two orchestration kinds, for 19 named kinds in total.
from qector_decoder_v3 import BlossomDecoder, BpOsdDecoderValidate
Claims are scoped to a declared workload and backed by SHA-256 sealed artifacts. Run qector-doctor before relying on optional paths and preserve the report with the result.
evidence policy · SHA-256 sealedMore Than Just MWPM
QEC decoding is not one-size-fits-all. Surface codes, qLDPC codes, optional batch paths, and simulation workloads each demand different decoders. QECTOR documents the supported contracts and the evidence needed to evaluate a selected workload.
Blossom MWPM (Exact)
Weight-optimal exact minimum-weight perfect matching for graph-like codes. The reference decoder for surface codes.
qLDPC Support
Matching decoders are restricted to graphlike structures. BP-OSD handles reachable arbitrary GF(2) check matrices, including LDPC and qLDPC workloads.
Workload-Scoped Execution
Union-Find and optional GPU batch paths are available for supported workloads. Availability and performance must be verified on the target environment.
One Library, Pluggable
Drop QECTOR into any existing Stim or PyMatching workflow. Same API surface, swappable backend. No vendor lock-in.
Decoding Algorithms
Union-Find, Fast Union-Find, Blossom MWPM, Sparse Blossom, Belief-Matching, BP-OSD, GPU Batch (CUDA/OpenCL), Hybrid, Predecoded, two-stage, ambiguity-cluster, colour-code, space-time, and other concrete paths are exposed through a consistent Python API. Workbench adds the two orchestration kinds, AutoDecoder and Auto Router, to make 19 named kinds.
Interactive QEC Simulator
Planar stabilizer code (d=3). Click edges to toggle physical errors.
Measure on Your Own Hardware
No universal benchmark figures are published on this site, because results depend on your hardware, drivers, and workloads. The qector bench harness ships with the package so you can measure logical error rates and throughput on your own machines.
GitHub Artifacts →Enterprise Compliance & Licensing
License Clarity
The package is source-available under PolyForm Noncommercial 1.0.0 for community use. Commercial use, hosted services, and redistribution require the written scope described on the license and pricing pages.
Learn about licensing →Commercial Licences, Air-Gapped
Pro from $79/month or $790/year with GPU batch decoding, Lab for distance 63 and an SLA. Licences are verified offline, with no licence server, and every plan has a 14-day money-back guarantee.
View Pricing & Tiers →QECTOR Pro
Air-gapped licence · Cancel any time · 14-day money-back guarantee
$79/mo
or $790/year (2 months free) · Production use
Need Lab, perpetual or OEM? View all tiers →