Meet Brian

Brion's quantum agent — wired to hardware, not to a script.

Brion Statevector Simulator · 20q Brion GPU Statevector Simulator · 25q · 8 GPUs Origin Wukong · 166q Origin WK_C180_2 · 17q Origin PQPUMESH8 · 3q

Ask Brian

Running on Brion's own hardware. Ask him to run a circuit, compute a molecular ground state, or tell you whether something is feasible — the tools he calls appear under each answer.

What Brian can reach, right now

Probed live every time this page loads — not a configuration file. A resource that is down appears with the reason, because an agent that hides an outage is an agent whose answers you cannot trust.

Brion Statevector Simulator simulator

20q online

Brion GPU Statevector Simulator simulator · 8 GPUs across 2/2 nodes

25q online

Origin Wukong quantum hardware · Origin Quantum

166q online

Tianyan Quantum Cloud quantum hardware · China Telecom Quantum Group — cqlib is not installed (No module named 'cqlib'). Install with: pip install cqli

504q offline

Origin WK_C180_2 quantum hardware · Origin Quantum

17q online

Origin PQPUMESH8 quantum hardware · Origin Quantum

3q online

Brian caps his own circuits at 24 qubits. The hardware goes higher; a conversation should not be able to queue an hour of compute by accident.

Run one of his tools

These are the same tools Brian calls. Each one executes on a real backend and returns measurements — nothing here is a canned response.

Pick a tool above.

The feasibility tool is the interesting one. Ask whether we can fold
a peptide and the answer is yes, with the qubit count and the paper.
Ask about mRNA and the answer is no, by a factor of 22, with the
classical method that does solve it.

An agent that only says yes is not an expert. It is a salesman.

His tool surface

Five tools, each one a real capability. The schemas are published in the Anthropic tool-use format, so the reasoning layer is configuration rather than a rewrite.

list_quantum_resources

Live status of every quantum resource Brion can reach right now: the GPU simulator fleet, the in-process simulator, and any QPU hardware. Call this before claiming anything about what is or is not available - it is measured, not cached.

run_quantum_circuit

Execute a quantum circuit and return real measurement counts. Operations are {"gate": "h", "qubits": [0]} with an optional "theta" for rotations. Supported gates: h, x, y, z, s, sdg, t, tdg, rx, ry, rz, p, cx, cz, swap, ccx, measure. Add explicit measure operations, or every qubit is measured.

compute_ground_state_energy

Estimate the ground state energy of a Pauli-sum Hamiltonian using VQE on real quantum backends. This is the tool for molecular chemistry. Below 15 qubits it also returns the exact classical answer, so the estimate can be checked rather than trusted. Note the result is electronic energy only - nuclear repulsion is a classical constant the caller adds.

classify_sample

Run the deployed qnn-hybrid quantum classifier on one sample. This is a trained model with measured held-out accuracy, not a demonstration.

assess_quantum_feasibility

What a class of problem costs in qubits, from the published literature, against what Brion actually has. Use this before promising anything about proteins, RNA, molecules, portfolios or cryptography. It returns real citations and the classical alternative where the quantum route is not yet viable.

Where Brian is today

Brian's hands are built and connected: he reaches an 8-GPU simulator fleet across two nodes, an in-process simulator, and Origin Wukong — a 180-qubit superconducting quantum processor. A Bell pair submitted through him returns from real hardware in about thirteen seconds, with roughly one shot in two thousand landing outside the entangled state. That residue is device error, and it is the clearest evidence there is that the computation happened on physical qubits rather than in a simulator.

His reasoning layer runs on Brion's own GPUs — an open-weights model served on the same fleet that runs the simulators, with no external API in the path. It reads your question, decides which tool to reach for, and explains what came back. Every quantitative claim he makes about quantum hardware comes from a tool call you can see in the trace under his answer.

What he will not do

Brian carries a feasibility tool with published qubit counts and citations, and he is built to use it before promising anything. Ask him to fold a 300-residue protein and he will tell you it needs about 600 qubits against the 24 he is allowed, cite Robert et al., npj Quantum Information 7, 38 (2021), and point you at AlphaFold. Ask about mRNA secondary structure and he will note it scales as roughly n² in sequence length, so a 30-base sequence already wants 435 qubits — and that ViennaRNA solves it classically in milliseconds.

This is the useful behaviour, not a limitation of it. A scientist can work with a system that knows its own edges. Nobody can work with one that agrees with everything.

Call his tools yourself

curl -X POST https://brionquantum.com/api/v1/brian/tools/run_quantum_circuit \
  -H "X-API-Key: $BRION_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"circuit": [{"gate":"h","qubits":[0]},{"gate":"cx","qubits":[0,1]}],
       "shots": 2000, "backend": "originq"}'

API documentation · The trained model · Technology