Neura Parse

The quantum substrate of the Neura Parse stack.

One IR. Every modality. Provenance by default. Fault-tolerant ready. qmesh is the missing layer above Qiskit, Cirq and PennyLane — a single intermediate representation that crosses gate, neutral-atom, photonic-CV and pulse modalities, with an AI-augmented compilation path and runtime where every result is signed, chained, and audit-ready by default.

qmesh / pipeline

qiskit · cirq · pennylane · pasqal · sf · bloqade

QPUPrimitive · ClassicalTask · ChannelOp · DAG

surface · BB qLDPC · cultivation

Aer · Stim · QutipEmulator · OpenPulse · IBM · IonQ

Quantinuum H2 reached Microsoft Level 2 Resilient with logical error rates 800× below physical. IBM Kookaburra (4,158 qubits) is on the 2026 roadmap. IonQ committed to 80,000 logical qubits via modular trapped-ion interconnects.

Photonic interconnects between trapped-ion and neutral-atom systems shipped commercially in April 2026 (IonQ, Cisco UQS). Neutral-atom platforms are Spectrum's “2026 big leap.”

Qiskit, Cirq, PennyLane and friends remain provider-centric. Cross-modality programs require bespoke glue. Auditability is an afterthought.

A single typed IR (qmesh.ir) for gate, neutral-atom, photonic-CV and pulse programs. Deterministic content-hash. Cross-frontend semantic equivalence — write once, run on any backend the program is compatible with.

Surface, repetition, BB qLDPC and cultivation-aware lattice surgery — promotion is a single config flag. PyMatching, sliding-window, BP+OSD and neural decoders are interchangeable. Resource estimates in the Microsoft RE shape, signed end-to-end.

Pasqal Pulser, Strawberry Fields, MrMustard, Bloqade frontends. QutipEmulator, Aer, Stim, OpenPulse backends. First-class ChannelOp lets a single program cross modalities — gate → Rydberg → CV in one signed run.

Typed DAG nodes (QPUPrimitive, ClassicalTask, MCMRegion, Barrier, Fanout, EntanglementBarrier). Topological executor with thread-pool parallelism. SLURM / PBS connectors. Crash-safe resume via signed parent_manifest_hash lineage.

Intent compiler for common patterns. LLM provider abstraction (Ollama / Anthropic / OpenAI / Mock). Token-level grammar masking with real-time GrammarMaskedProvider. Neural decoder training pipeline (Stim → PyTorch → signed weights). QuanBench evaluation suite.

Every run produces an ed25519-signed manifest. Tamper-evident compliance packs with offline-verifiable hash chains. Cert-chain trust model with TrustStore. Metriq export for public benchmarking. Designed for regulated industries.

Click any stage to see the IR primitives, the matching CLI surface, and the active flag at that point in compilation.

Typed, content-hashed

Every program lowers to a single typed intermediate representation with deterministic content-hash. Two semantically equivalent programs produce identical hashes — across frontends, across modalities.

  • Typed nodes: QPUPrimitive, ClassicalTask, ChannelOp, MCMRegion
  • Cross-frontend semantic equivalence
  • Deterministic SHA-256 content-hash
  • First-class ChannelOp lets gate → Rydberg → CV in one program
qmesh / ir
ir.hash()
# 'ab1c4f...e8d2'  (deterministic across frontends)

ir=qmesh.ir.v1

Click any stage to inspect its inputs, primitives, and the qmesh CLI surface at that point in the pipeline.

The compatibility matrix below shows which SDK frontend or backend connector is active for each hardware vendor. Filter by modality to focus.

9 backends
IBM Heron
Superconducting (gate)
Primary QANTIS hardware target
Google Willow
Superconducting (gate)
Cirq native frontend
Quantinuum Helios
Trapped-ion (gate)
Microsoft RE shape supported
IonQ
Trapped-ion (gate)
Modular interconnect roadmap
QuEra
Neutral-atom
Analog Rydberg + ChannelOp
Pasqal
Neutral-atom
Pulser-native pipeline
Atom Computing
Neutral-atom
Beta connector
Xanadu
Photonic-CV
Strawberry Fields native
PsiQuantum
Photonic-CV
MrMustard interop, planned backend
Frontend & backendFrontend onlyBackend onlyPlannedNot supported
DefenseCert-chain signed runs, tamper-evident audit packs, classical+quantum DAG with SLURM / PBS.
AerospaceQuantum navigation (QANTIS) compiled through qmesh; reproducible, manifest-signed flight pipelines.
IndustrialHybrid optimisation pipelines with HPC connectors; Mitiq ZNE mitigation for noisy-era runs.
HealthcareCompliance pack with offline-verifiable hash chain; cert-chain trust model satisfies regulatory provenance.
AutomotiveSensor-fusion + quantum-accelerated planning via cross-modality DAGs.

We don't want to replace Qiskit, Cirq, or PennyLane. We want to give you a single program — written once, signed always, runnable on any modality — that those frontends can compile into. Bring the SDK you already know.

QiskitCirqPennyLaneqBraidOpenQASM 3StimPyMatchingMitiqPasqal PulserStrawberry FieldsMrMustardBloqade
$ pip install -e ".[all]"

# 1 — Hello, quantum
$ python3 examples/hello_bell_crossbackend.py

# 2 — A real algorithm
$ python3 examples/vqe_h2.py                    # E = −1.859 vs FCI −1.857

# 3 — Fault-tolerant memory
$ python3 examples/ft_logical_memory.py         # Λ = 23 across d=3 → d=5

# 4 — Cross-modality, single signed run
$ python3 examples/dag_cross_modality.py        # gate → Rydberg → CV

# 5 — Compliance pack for auditors
$ python3 examples/compliance_export.py         # signed chain, offline-verifiable

# Service mode
$ uvicorn qmesh.service.app:app --port 8788

# CLI (16 commands)
$ qmesh submit circuit.qasm --shots 4096
$ qmesh check ledger/2026-05-02/<hash>.json
$ qmesh compliance-pack ./ledger --since 2026-05-01