MolAgent
Pre-seed · Building Early Access →
An expert AI agent · install on any machine · Tournai, BE

Materials discovery
at the speed of thought.

MolAgent is an AI agent you install on your workstation or HPC cluster — expert in quantum chemistry, molecular dynamics, coarse-grained modeling, and ML interatomic potentials. Pre-built skills, curated pipelines, and deep knowledge of every major simulation package. It proposes strategies, sets up calculations, monitors jobs, analyzes trajectories, and interprets results — so your team stops reinventing the wheel and starts producing science.

An expert across every level of theory
methods.proposed · curated_pipelines
Quantum
DFT · ab initio · TD-DFT
Atomistic
classical MD · force fields
Coarse-Grained
MARTINI · DPD · mesoscale
E = f(graph)
ML-Accelerated
MACE · NequIP · Allegro
$ molagent
The Agent
proposes · runs · analyzes
MolAgent is fluent across every level of theory. It doesn't pick for you — it proposes strategies, explains trade-offs, and once you decide, it sets up the calculation, launches it, monitors progress, and interprets the output.
install · run · ship results
4 levels
Theory levels covered
20+ tools
Simulation packages
100%
Local · your hardware
0 outbound
Your data stays yours
01 / problem

HPC clusters sit idle because expertise doesn't scale.

Every university, national lab, and industrial R&D center owns compute capacity they cannot fully use. The bottleneck isn't silicon. It's the rare specialists who know how to wield it.

01
Generic agents are blank canvases
Out of the box, they don't know LAMMPS conventions, VASP pseudopotentials, GROMACS topologies, or MLIP training pipelines. Every team rebuilds the same skills from scratch.
Industry-wide
02
Knowledge lives in one head
Input parameters, force-field choices, training recipes — scattered across individuals, lost when they leave, reinvented by the next person who joins.
Critical
03
Off-the-shelf models hallucinate physics
General LLMs invent force-field parameters, propose unconverged k-points, suggest unphysical thermostats. A plausible wrong answer burns weeks of compute and trust.
Dangerous
04
Time-to-first-result is too long
Hypothesis to converged simulation takes weeks of human attention. The bottleneck is the specialist's time — not the GPU's.
Systemic
02 / how it works

An AI expert agent built to live inside your HPC cluster.

A simple install on your workstation, lab server, or HPC login node. From the first command, MolAgent is fluent in your tools, your queues, and your simulation packages. No skill engineering required.

HOW IT FITS
molagent · v0.1.0
YOU
Researcher
  • Describe the problem
  • Review proposed strategies
  • Approve, edit, decide
  • Stay in control
INSTALLED ON YOUR MACHINE
MolAgent CLI
  • Pre-built skills + pipelines
  • Method proposal engine
  • Input / script generation
  • Job submission + monitoring
  • Trajectory + result analysis
  • Memory of your conventions
YOUR ENVIRONMENT
HPC · Workstation · Cloud
  • SLURM / PBS / LSF queues
  • CPU + GPU nodes
  • Installed simulation binaries
  • Storage + trajectories
01
Strategy Proposals
Decomposes your question into options. Suggests methods, level of theory, convergence approach. You decide.
02
Input Generation
Validated input files for LAMMPS, VASP, GROMACS, CP2K, ORCA — calibrated to your system and conventions.
03
Job Submission
Builds your SLURM/PBS scripts, submits, watches the queue, restarts on failure, scales when needed.
04
Analysis & Interpretation
RDFs, MSDs, free-energy landscapes, phonons, band structures. Generates figures and explains what they mean.
05
MLIP Pipelines
Active learning, data curation, training pipelines for MACE, NequIP, Allegro, and custom potentials.
06
Lab Memory
Learns your group's protocols, prior results, and conventions. Reuses what worked before.
07
Per-User Sessions
Each researcher has their own context. Shareable runs. Full audit trail of every reasoning step.
08
Local-First
Runs on your hardware. Open-source LLM under the hood. Your data, your weights, your control.
03 / methods

Literate in the full simulation stack.

From electronic structure to coarse-grained mesoscale. From static lattices to ensemble statistics. MolAgent speaks every language your team already uses — and proposes the right one for the question.

01 · Electronic structure
Quantum Chemistry & DFT
Ground and excited-state electronic structure, band structures, phonons. Automated k-point convergence, functional selection, and embedding schemes.
VASP Quantum ESPRESSO CP2K ORCA PySCF
02 · Atomistic dynamics
Molecular Dynamics
All-atom and classical MD from ps to µs. Enhanced sampling — metadynamics, umbrella, replica exchange. Transport coefficients and free-energy landscapes.
LAMMPS GROMACS NAMD OpenMM PLUMED
Q P N C C C C
03 · Mesoscale
Coarse-Grained Modeling
Mesoscale dynamics for soft matter, polymers, membranes, biomolecular assemblies. MARTINI, DPD, systematic coarse-graining from atomistic data.
MARTINI DPD ESPResSo HOOMD-blue VOTCA
E-equivariant message passing
04 · Equivariant GNNs
ML Interatomic Potentials
Training and deployment of equivariant graph neural network potentials. Active learning, uncertainty quantification, transferability across chemistries.
MACE NequIP Allegro SchNet GAP
candidate → score → select
05 · High-throughput
Screening Pipelines
Automated workflows for screening chemical and materials spaces. Crystal structure prediction, adsorbate binding, solubility, reaction barriers.
Materials Project OQMD AFLOW AiiDA Atomate
r (Å) g(r)
06 · Observables
Analysis & Visualization
RDFs, MSDs, free-energy surfaces, phase diagrams. Publication-ready figures with consistent visual language and reproducible notebooks.
MDAnalysis ASE pymatgen OVITO VMD
04 / roadmap

From agents to an operating system for discovery.

Aggressive but honest timeline. Funding-aware. Every phase de-risks the next.

Progress
Phase 1 of 6
Updated Q2 2026 · Next review Q3 2026
Q1 2026
Complete
Company + vision
Incorporation in Belgium. Technical architecture defined. Method library designed.
Q2 2026
In progress
Agents + grants
Working DFT & MD agents. Applications: NVIDIA Inception, Anthropic Startup, EuroHPC, Wallonia R&D.
3
Q3–Q4 2026
Planned
Alpha + pilots
Deployable alpha. 2–3 pilot partners. Materials discovery + MLIP training validation.
4
H1 2027
Planned
Beta appliance
First shippable server. SLURM/PBS. Methodology papers. Industry pilots.
5
H2 2027
Planned
GA + seed round
General availability. Subscription + hardware. Seed financing to scale team.
6
2028 →
Vision
Vertical packages
Drug discovery, batteries, catalysis. Open plug-in ecosystem. EU + US expansion.
05 / markets

Built for teams already sitting on unused compute.

Three customer archetypes. Same underlying pain. Different procurement paths.

01
Universities & research groups
Accelerate graduate research. Standardize methods across cohorts. Capture institutional knowledge before it walks out.
Computational chemistry Materials science Biomolecular labs
02
National labs & HPC centers
Lower the barrier for non-expert users accessing exascale resources. Maximize utilization of GPU partitions and specialized hardware.
EuroHPC facilities Government labs AI factories
03
Industry R&D
Simulation-driven discovery in-house with full data sovereignty. Drug discovery, catalysis, batteries, polymers, advanced materials.
Pharma & biotech Energy Semiconductors
06 / founder

Built by someone who has lived the problem.

Not an outsider's guess at what scientists need. A computational chemist turning a decade of practice into software.

// BACKGROUND founder.md

More than a decade at the intersection of molecular dynamics, quantum chemistry, and machine learning for materials. Peer-reviewed publications in the field. Active research professor appointment. Maintainer of open-source scientific software.

The product is not an outsider's guess. It's the codification of how an expert actually works — the heuristics, the debugging instincts, the method choices — into software that scales beyond a single person.

The founder maintains active research output, ongoing collaborations with leading European and North American academic groups, and a clear view of where current AI capabilities fall short of real scientific workflows.

// PROFILE scientist.yaml
[Your Name]
Founder & Chief Scientist
role Founder · MolAgent (Tournai)
academic Research Professor, computational chemistry
phd Computational Chemistry, UMons
expertise MD · DFT · Coarse-grained · MLIPs
output Publications · OSS maintainer
based Tournai, Wallonia · Belgium
Early access · Pilots · Collaborations

An expert co-pilot, one install away.

We're opening early access to a small cohort of research groups, HPC centers, and industry teams. If your scientists shouldn't be spending their days writing input files, let's talk.

Email
hello@molagent.example
HQ
Tournai, Belgium
Stage
Pre-seed