Cebule
Cebule is a computational chemistry platform for pharma, biotech and materials R&D. Molecular Quantum Solutions pairs over 200 million optimized molecules with quantum chemistry, thermodynamics and machine learning, through a dashboard, a Python SDK or a research agent.
What is Cebule?
Cebule is the platform built by Molecular Quantum Solutions (MQS), a deep tech company based in Copenhagen and hosted by Quantum Denmark at the Niels Bohr Institute. Its stated positioning is Acceleration through Quantum AI, and the home page promises compounds computed and developed in hours rather than months, R&D cycles ten times faster, and accuracy at DFT and CCSD(T) level.
Three named building blocks make up the product: Cebule Dashboard, a visual interface; Cebule SDK & HPC, a Python library backed by CPU and GPU nodes; and Cebule Agent, an agentic assistant. Beneath them sits a searchable database of more than 200 million molecules. It draws on PubChemQC PM6 — 221 million PM6-optimized structures, over 80 million of them neutral, over 35 million spin-flipped and over 50 million cationic, each entry holding 3D coordinates, vibrational spectra, orbital energies, HOMO-LUMO gaps, dipole moments and Mulliken charges — and on QMugs, 685,917 drug-like molecules from ChEMBL computed at two levels of theory, GFN2-xTB and DFT ωB97X-D/def2-SVP.
The compute stack is unusually wide. Electronic structure runs from DFT and plane-wave DFT (BP86 and ωB97X-D functionals; def2-TZVP, def2-TZVPD, def2-SVP and 6-31g** basis sets) through semi-empirical PM6 and PM7 and tight-binding GFN1-xTB, GFN2-xTB and g-xTB, to machine-learned interatomic potentials such as MACE, ORB, PET-MAD, GRACE-2L and UMA. Solvation and thermodynamics cover COSMO, PCM and IEFPCM plus the COSMO-RS/COSMO-SAC, NRTL and UNIFAC families, whose parameters export to process simulators. Molecular dynamics spans Born-Oppenheimer, Car-Parrinello and path-integral methods alongside classical force fields through OpenMM and LAMMPS, with native polymer support in pSMILES notation. Machine learning adds three graph neural network architectures, and the quantum computing layer offers variational algorithms, subspace expansion, phase estimation, qubit reduction, tensor networks, zero-noise extrapolation and a genetic circuit generator called Xenakis.
Lab Connect reaches into the laboratory itself with OPC-UA gateways, MQTT messaging, a PostgreSQL time-series schema, ROS2 robot automation and closed-loop Bayesian optimization. QPU connectors to IBM Quantum, IQM or Xanadu run on the customer’s own credentials. Delivery is deliberately infrastructure-as-code rather than pure SaaS: a CloudFormation Template provisions the whole environment inside the customer’s AWS account, and the platform is also offered through AWS Marketplace.
What it does
- Search more than 200 million molecules by SMILES, InChI, InChIKey, ChEMBL identifier, CAS number, formula, mass or charge, with wildcards and Boolean operators
- Run quantum chemistry calculations — DFT, PW-DFT, PM6, PM7, GFN1-xTB, GFN2-xTB — on the Cebule HPC
- Predict mixture properties such as solubility, viscosity and phase equilibria with COSMO-RS/SAC, NRTL and UNIFAC
- Train and fine-tune property prediction models using the built-in graph neural network library
- Chain tasks through connected_task_id to build a multi-step pipeline, for example GEOMETRY_OPT to COSMO to SIGMA to SOLUBILITY
- Ask the research agent for runnable Python scripts fitted to your own use case
- Deploy the full stack into your own AWS account with the CloudFormation Template and connect it to your laboratory equipment
When to use Cebule / When not to
A quick filter to help you decide if Cebule is the right fit.
When to use Cebule
- R&D teams in pharma, biotech and chemistry that run high-throughput molecular screening and want to stop stitching a dozen separate tools together
- Academic researchers and PhD students in quantum chemistry, who have a free tier to work from and an academic licensing model to ask about
- Process engineers who need binary interaction parameters exported into Aspen Plus, HYSYS, AVEVA PRO/II or DWSIM
- Data scientists and molecular ML engineers after a graph neural network library, generative GAN models, machine-learned potentials and descriptors, with a Kubeflow integration
- Organizations under data governance constraints that want the entire stack deployed inside their own AWS account
When not to use Cebule
- Teams that want a ready-to-use tool with no cloud account of their own: the Enterprise tier explicitly requires an AWS account
- Buyers who need written legal commitments such as a DPA, a privacy policy or terms of service, because the site publishes none of them
- Anyone expecting a mobile app: no iOS or Android application exists
- Users who need an interface in a language other than English, since the website, the documentation and the application are English only
- Newcomers looking for a consumer-grade or introductory chemistry tool, as Cebule assumes a background in quantum chemistry or in Python
How to use Cebule
A typical end-to-end flow, from setup to results.
- Sign up from the home page: the Sign Up and Start Free buttons lead to cebule.mqs.dk, accounts are authenticated through Auth0, and the site states that no credit card is required to get started
- Start without writing code in the Cebule Dashboard: search for molecules, visualize their properties, compare datasets and export the results
- For scripted work, install the Python SDK from the GitLab repository at gitlab.com/mqsdk/Python-SDK
- Open a session with import mqsdk, then mqsdk.Session(email, password), and query the database with session.data.search()
- Create calculations with create_task(): choose a task type and pass the input data, and the platform allocates the compute, runs the containers and stores the result on S3
- Chain steps into an automated pipeline by passing connected_task_id from one task to the next
- Work interactively in the hosted JupyterLab and Kubeflow environments, with adjustable compute power
- Ask the research agent in plain language for code examples and datasets when you are unsure where to start
- For an Enterprise deployment, subscribe through AWS Marketplace, launch the CloudFormation Template while choosing your AWS region and instance types, then open the Dashboard and JupyterLab from the URLs the stack returns
- Remember that only the Python SDK installs locally: the platform itself is not installed on a workstation
Pros & Cons
Pros
- Unusually broad functional coverage for a single product: database, quantum chemistry, thermodynamics, molecular dynamics, machine learning and quantum computing in one stack
- Two genuinely distinct levels of access, a no-code interface and a Python SDK, which opens the tool to bench scientists and developers alike
- The stack can be deployed inside the customer’s own infrastructure through a CloudFormation Template, with the AWS region chosen by the customer
- Public and legible pricing for the Free and Professional tiers, which is rare in this market
- Detailed public technical documentation on docs.mqs.dk, including the API specification and example notebooks on GitLab
- Verifiable scientific grounding: five research papers linked from the Community page (arXiv and DOI) and a named scientific board
- Claimed absence of lock-in: the SDK integrates with third-party packages and the workflows follow MLOps practice
Cons
- No legal document is published at all: no privacy policy, no terms of service, no legal notice, no cookie policy, no DPA and no mention of the GDPR
- The Enterprise tier requires an AWS account, which the site presents as a current requirement of that deployment
- There is no dedicated pricing page: the table lives in a section of the home page with no URL of its own, which makes it hard to cite or to track
- No contact page and no contact form, only a single email address published in obfuscated form
- The Request Demo, Schedule Demo and Request Meeting buttons carry no URL: they open a modal window, and no demo page exists
- Interface, documentation and website are in English only, with no language selector
- cebule.io, presented on the home page as the address of the Cebule Agent, disallows automated collection in its robots.txt, so that part of the product could not be documented
Pricing & Plans
A permanent free tier is available. Cebule Free provides 25 minutes of compute time and access to the platform for an initial trial, and the home page states that no credit card is required to get started. Cebule Professional is billed per usage, as blocks of compute minutes purchased up front: 700 minutes for $50 (€40), 2,500 minutes for $150 (€130) and 6,000 minutes for $300 (€260), the larger blocks being proportionally cheaper. Cebule Enterprise is priced on request and covers cloud or on-premise deployment with an optional European HPC provision, reached through a Request Meeting button. Subscription is also possible through AWS Marketplace, which brings the spend inside an existing AWS contract. The partnering page lists further commercial arrangements — technology licensing, an academic Cebule license with support, fixed-fee consulting and joint R&D programs — none of which carries a published amount.
- Cebule Free — “Perfect for exploring” — 25 minutes of compute time and platform access for an initial trial
- Cebule Professional — “Pay per usage” — includes Database
- HPC Engine and Model Library
- Research Agent
- JupyterLab & Kubeflow — blocks of 700 minutes for $50 (€40)
- 2
- 500 minutes for $150 (€130) and 6
- 000 minutes for $300 (€260)
- Cebule Enterprise — “Cloud/on-premise + optional European HPC provision” — deployment on AWS
- Azure or Google Cloud (other providers also supported)
- on-premise and/or supercomputer/HPC connection
- quantum computing integration
- automated laboratory connection — price on request
Data, GDPR & hosting
A consolidated view of how Cebule handles your data.
GDPR overview
The site carries no GDPR information whatsoever. Across the thirteen archived pages and the application bundle there is not one occurrence of GDPR, data protection, personal data or privacy: no privacy policy, no cookie notice, no data processing agreement, no Article 27 representative, no data protection officer and no dedicated address, only the generic contact@mqs.dk. What can be verified is factual rather than legal. The publisher, Molecular Quantum Solutions, is established in Denmark, inside the European Union, at the Niels Bohr Institute in Copenhagen, and the application authenticates against an Auth0 tenant in a European region. Being established in the EU is not a declaration of compliance: the site neither claims compliance nor denies it, and anyone with a regulatory requirement will have to obtain those commitments directly from the publisher.
Who owns the data?
Molecular Quantum Solutions publishes neither terms of service nor a privacy policy, so no document states who owns the data on the hosted Free and Professional tiers, or what MQS may do with it. The only ownership language on the site is marketing copy about the Enterprise deployment: “Your molecules, your simulation results, your proprietary structures never leave your storage”, and “You own the compute. You own the data.” Both describe the CloudFormation Template deployment running in the customer’s own account, where each team member connects a personal S3 bucket and task results are written to it automatically. They do not extend to the hosted tiers, the public database stays on MQS infrastructure, and no contract backs any of it.
Reuse rights
There are no terms and conditions to read, so the site never says whether you may reuse, republish or commercially exploit the results Cebule computes for you: no licence is granted and none is withheld. In practice, results from a customer-side deployment are written to your own S3 bucket, which puts them within your control, while the multi-terabyte public database remains hosted by MQS. Nothing published states what MQS does with inputs or results on the hosted tiers, and the site takes no position, either way, on training models with customer data. No sub-processor list exists, although the hosted application loads Auth0, Stripe and Hotjar. The self-deployable multi-agent framework does let customers choose which LLMs it calls.
Data retention & training
Hosting summary
No hosting country is named anywhere on the site. MQS states that it runs its own self-hosted HPC infrastructure on ARM processors, described as scalable to hundreds or even thousands of CPUs, but never says where those machines are located. The Enterprise tier mentions an optional European HPC provision: an option that can be requested, not a rule, and not a guarantee of European hosting. With a CloudFormation Template deployment the question moves to the customer, who chooses the AWS region when launching the stack; each team member can also connect their own S3-compatible object storage, and task results are written to it automatically. The multi-terabyte public database remains hosted by MQS. Because nothing explicit is declared, the hosting country and region fields of this listing are left empty rather than guessed. Note also that the domain resolves to a Cloudflare anycast address, which reflects a CDN edge and says nothing about where data is stored.
Things to keep in mind
Risks and trade-offs to weigh before adopting Cebule.
- No legal document is published on mqs.dk — no privacy policy, no terms, no legal notice, no cookie policy, no DPA and no GDPR mention — which leaves you without written recourse if anything goes wrong
- The incorporation date of the publishing company could not be established reliably, so that field is left unresolved in this listing
- The published address is a c/o domiciliation at Quantum Denmark, inside the Niels Bohr Institute, rather than a registered office of the company itself
- The data sovereignty claims apply to the Enterprise deployment in your own account, not to the hosted Free and Professional tiers, so do not read them as a promise covering the whole product
- The hosted application loads third-party services the site never documents: Auth0 for authentication, Stripe for payment and Hotjar for analytics
- The resolved IP address is a Cloudflare anycast node, so it tells you nothing about where data is actually hosted
- The Danish commercial register lists two employees for the publisher, which is worth weighing against the functional ambition of the platform
Setup & Integrations
Technical difficulty
Difficulty depends on the route. The no-code route is easy: sign up on cebule.mqs.dk and use the Dashboard straight away, with no technical skill required according to the site. The developer route is moderate: install the Python SDK, open an authenticated session, and call create_task, which assumes Python. The Enterprise route is the demanding one, requiring an AWS account, a CloudFormation Template to launch, instance types to choose and Kubernetes and SLURM integration. Hosted JupyterLab and Kubeflow environments come ready to use.
Deployment
Integrations
Supported languages
Behind Cebule
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
What is Cebule?
Do I need to know how to code to use it?
How many molecules does the database contain?
Is there a free plan?
How much does the first paid tier cost?
Is there an API?
Can I host Cebule myself?
Which quantum computing providers are supported?
Does the site publish a privacy policy or terms of service?
Is there a mobile app?
Should you pick Cebule?
Cebule is a dense, technically ambitious product, and that density is its main argument: one stack holding a 200-million-molecule database, quantum chemistry, statistical thermodynamics, molecular dynamics, graph neural networks and quantum computing methods, where most teams would otherwise assemble half a dozen separate tools. The scientific grounding is checkable, with published papers, a named scientific board and research partners including DLR, Fraunhofer SCAI, the BioInnovation Institute, planqc, Oxford Ionics and d-fine, and the technical documentation is public and detailed. Molecular Quantum Solutions says it serves pharmaceutical, biotechnology, chemical, materials science, food, cosmetics and biomaterials work, and the platform is reachable three ways: a no-code dashboard, a Python SDK with a documented REST API, and an agentic assistant.
The blind spot is documentary rather than functional. At the date of this review, mqs.dk publishes no privacy policy, no terms of service, no legal notice, no cookie policy, no data processing agreement and no mention of the GDPR. For a pharmaceutical or industrial buyer that is a genuine obstacle, and the sovereignty language MQS uses — your molecules never leave your storage — describes the Enterprise deployment inside the customer’s own account, not the hosted Free and Professional tiers, with nothing contractual behind it either way. The publisher describes itself as founded in 2019 and says it developed the Cebule™ platform over six years; we could not establish an incorporation date reliably, so that field is left unresolved here.
The publisher is established in Denmark, inside the European Union, at the Niels Bohr Institute in Copenhagen. Anyone weighing Cebule for regulated work should ask for the paperwork in writing first. Anyone curious about the science can start on the free tier and judge it directly.
- Choosing a selection results in a full page refresh.
- Opens in a new window.