A research agent with its own laboratory: it drafts the experiment, probes it on free simulators first, escalates to paid compute only with your approval — and freezes every run into a reproducible record.
Four lines of work feed the products on this site. Each one says what is measured, on what, and what is still in progress.
The SynapseX model family behind Chat and the API Platform is trained on European HPC GPU resources, under EuroHPC allocation EHPC-AIF-2025LS13-019. We study what makes a model useful for code, science and quantum work, and how to measure that honestly.
Circuit placement, routing and optimisation benchmarks, measured on simulators and labelled as simulation. A method that loses is reported as a loss, and hardware results are reported only once a partner QPU run exists.
Every run is frozen into a sha256 manifest of code, dependencies and backend, sealed under a Merkle root. That proves integrity and lets anyone re-run it. It does not prove who ran it.
Variant interpretation as a reproducible lab: the agent gathers the public evidence, writes and runs the analysis, and escalates a subproblem to quantum only where that is honest.
Describe the question in plain language. The agent drafts hypotheses, circuits and analysis code in a persistent Labs workspace — Jupyter notebooks included.
Prototype and debug quantum algorithms against exact simulation, sweeping parameters from Jupyter notebooks the agent reads and edits with you — and catch the bug before it costs a paid run.
Run error-correction Monte Carlo on repetition and surface codes and compare device topologies on the free tools. Noisy circuit simulation is in development.
Compare QEC code families and device topologies on the free tools — then let the agent argue the trade-offs for your device.
Small-molecule and small-ansatz studies on the free simulator. Larger electronic-structure runs are in development.
Run manifests and evidence bundles designed to be checked by someone who wasn't in the room.
Quantum marketing runs on multipliers. We publish a number only when a run record stands behind it, and every number is labeled simulation or hardware.
Everything here can be opened by someone outside the company. Technical reports join the list when they are public, with what is needed to check them.
Releases link to their public package pages. Compute allocations are listed for context; they are not results.
Trust in quantum results shouldn't require trusting us. The workflow assumes you'll verify — so every layer leaves something verifiable behind.
Code, dependencies and backend are hashed into a manifest at dispatch — the run is re-runnable, the record tamper-evident.
Free simulation answers most questions. Paid compute, when it opens, will ask for an estimate and your approval before a credit moves.
Releases link to their public PyPI pages, and the evidence seal can be recomputed offline by anyone.
Every number is labeled simulation or hardware, and nothing is called hardware until it ran on hardware.
Install from PyPI, keep your notebooks, script the same lab the agent uses.
Start free: simulators cost 0 credits within the Free limits, up to 100 simulation runs a month, one at a time, up to 60 seconds each, on CPU only. Partner QPU access is in development.
SynapseX Labs · pip install synapsex · chat.synapsex.ai