OPEN SOURCE PYTHON SDK · BY SUPERAGENTIC AI
An open software layer across quantum SDKs, with a classical baseline for every result.See whether the quantum part earns its keep.
A REAL SAVED RESULT
LOCAL SIMULATORTwo qubits. One linked result.
This Bell test uses two simulated quantum bits. Across 1,024 measurements, both bits always had the same value.
| Both bits | Count | Share |
|---|---|---|
| 00 | 511 | 49.9% |
| 11 | 513 | 50.1% |
Saved with SDK 0.3.0 and seed 42. This is a computer model, not a live cloud test. It does not show that quantum methods beat standard ones.
Download the saved result (JSON)DEMO
Watch SuperQuantX 0.3.0
A short silent walkthrough with captions on screen: one interface across SDK backends on local simulators, VQE and QAOA with classical checks, then the local MCP server with all six tools.
SuperQuantX 0.3.0 · Local simulation · No quantum hardware used
NEW TO QUANTUM AI?
A shared workbench for quantum experiments
SuperQuantX is an open source toolkit that makes quantum AI experiments easier to run and compare. You use one Python interface across different quantum frameworks, with classical methods alongside, so you can see what actually helps.
Think of it as a shared experimental workbench: different quantum engines underneath, one interface on top, and a classical benchmark beside the results.
Today: the validated workflows run on local quantum simulators. Running on real hardware, and any useful quantum advantage, need separate evidence.
THE PROBLEM
A quantum result means little without a baseline
Quantum SDKs each have their own API, and most examples stop at the quantum number. SuperQuantX runs the same circuits and algorithms on several SDKs and reports a classical reference beside each result, so you can tell when a quantum model does not beat a classical one.
SUPPORTED BACKENDS
One API across quantum SDKs
Six gate-model backends are tested in the SDK unit tests on their local simulators. D-Wave Ocean runs QUBO and Ising problems. Hardware goes through each SDK with your own credentials and is not tested.
Azure Quantum
Rigetti Forest
cuQuantum
TensorFlow QuantumLogos and product names are trademarks of their respective owners. Their use identifies the SDKs SuperQuantX integrates with and does not imply endorsement. Sources are listed in the third-party notices.
ALGORITHMS
Each one reports its classical check
Exact ground energy, H2 STO-3G built in
Brute-force optimum and approximation ratio
Tuned RBF SVM and logistic regression via compare_to_classical
compare_to_classical works with any sklearn-style classifier
Brute-force optimal selection and a Markowitz reference
compare_to_classical verdict and p-value
import numpy as npimport superquantx as sqx vqe = sqx.create_vqe_for_molecule("H2", backend="simulator")print(vqe.find_ground_state(), vqe.exact_ground_energy())# both about -1.857275 Ha W = np.array([[0, 1, 1, 0, 1], [1, 0, 1, 1, 0], [1, 1, 0, 1, 0], [0, 1, 1, 0, 1], [1, 0, 0, 1, 0]], dtype=float)r = sqx.QAOA(backend="simulator", p=2, seed=7).solve(W)print(r["cut_value"], r["classical_optimum"], r["approximation_ratio"])import numpy as npimport superquantx as sqx returns = np.random.default_rng(0).normal(0.01, 0.02, size=(250, 6))res = sqx.QuantumPortfolioAgent(backend="simulator", n_select=3, seed=0).solve(returns)print(res.result["selected_assets"], res.result["classical_selected_assets"], res.result["matches_classical_optimum"])BENCHMARK EVIDENCE
A quantum classifier, benchmarked against classical models
One example run, not a general result. The code and numbers come from the compare_to_classical example in the SDK README, reproduced with superquantx==0.3.0 on the built-in simulator: 100 samples, 2 features, 5-fold stratified cross validation with identical folds, seed 0. The tuned quantum kernel was not significantly different from a tuned RBF SVM (difference -0.020, p=0.54). Wall times are from that run and depend on the machine.
import superquantx as sqxfrom sklearn.datasets import make_moons X, y = make_moons(n_samples=100, noise=0.2, random_state=0)result = sqx.compare_to_classical(X, y, sqx.QuantumSVM(backend="simulator"), cv=5)print(result.table())| Model | Accuracy | Wall time | Grid | QPU circuits |
|---|---|---|---|---|
| QuantumSVM (simulator), tuned | 0.930 ± 0.051 | 1.16s | 25 | 119,000 |
| QuantumSVM (simulator), untuned | 0.670 ± 0.125 | 0.13s | 1 | - |
| SVC (RBF), tuned C and gamma | 0.950 ± 0.055 | 0.91s | 25 | - |
| Logistic regression, tuned C | 0.860 ± 0.097 | 0.22s | 4 | - |
Known limitations in 0.3.0: the tuned comparison fails when the quantum model uses feature_map="auto", and the inner-validation preprocessing in the tuned comparison is still under review. Treat these numbers as an illustration of the method. Benchmarking guide
COMMAND LINE
Check your install in one command
list-backends runs a real Bell-state check on every installed backend. benchmark compares a quantum kernel against tuned classical models.
$ uv add "superquantx==0.3.0"$ uv run superquantx -v version # versions of installed backend SDKs$ uv run superquantx list-backends # real Bell-state check per backend$ uv run superquantx run qsvm --data iris -b simulator# or run once without adding it to a project$ uvx --from "superquantx[pennylane]==0.3.0" superquantx benchmark -d moons -b pennylaneMCP SERVER
Six tools for AI assistants
An optional MCP server speaks stdio, runs on local simulators by default and annotates every tool read-only. Start it with superquantx-mcp from your MCP client.
LIMITATIONS
Research software: current limits
OPEN SOURCE · APACHE 2.0
SuperQuantX is open source. Build with us.
Install the release, read the code, open an issue or send a fix. Contributions are welcome.
Want help applying SuperQuantX to a research question? Talk to us about an assisted pilot.