Aitomistic Logo
Protomia

Features

What Protomia Solution does — calculations, materials, writing, literature and your workspace.

Protomia runs real computational chemistry from a conversation. You describe the molecule and the task; it prepares the input, submits the job, watches the queue, and comes back with results, files and figures in the same thread.

It also does the work that surrounds a calculation — the writing, the literature, the scripts, the files. This page lists all of it. Where something is limited, that is stated.

Molecular calculations

TaskWhat you get back
Single pointenergy, dipole, HOMO/LUMO
Geometry optimisationoptimised structure, energy, an .xyz you can download
FrequenciesIR and Raman spectra, thermochemistry, zero-point energy
Molecular dynamicstrajectories, with fast ML potentials where full QM would be too slow
Transition-state searchthe TS, its imaginary mode, and the barrier
IRCthe reaction path down from a transition state, to confirm it connects the right minima
NEBa minimum-energy path between two structures
Reaction energeticsreaction energy and thermodynamics from reactants, products and transition states, with stoichiometry
UV–Vis spectravertical excitations, or a nuclear-ensemble spectrum for a realistic band shape
Emission / fluorescencethe full Kasha S₁ workflow — ground-state optimisation and frequencies, excited-state optimisation and frequencies, then the ground-state energy at the excited geometry
Excited-state geometries and frequenciescomputed on S₁, S₂, … rather than S₀
Orbitals and densitiescube files for HOMO, LUMO and density, ready for isosurfaces
Batch jobsthe same calculation across a set of molecules, submitted together
Molecular descriptorscomputed properties for a structure

Multi-stage work is chained for you: "optimise with AIQM3, then run OMNI-P2x for the UV–Vis spectrum" runs each stage on the previous stage's optimised geometry without you managing the handoff.

Periodic systems and surfaces

Crystals, slabs and adsorption, through CP2K:

  • cell and slab geometry optimisation, band gaps
  • supercells to any N×N×N
  • slabs by Miller index, with your choice of vacuum spacing and termination
  • adsorbate placement and adsorption workflows, with the CP2K output post-processed for you

Methods

  • Machine-learned quantum chemistry — AIQM1, AIQM2, AIQM3, and UAIQM, the universal and updatable AIQM model that spans nearly the whole periodic table.
  • Neural network potentials — including OMNI-P2x for excited states.
  • Semi-empirical — GFN2-xTB, ODM2*, OM2.
  • DFT and wavefunction theory — B3LYP/6-31G*, ωB97X-D/def2-TZVP, CAM-B3LYP and others, with your choice of basis set.

Element coverage, honestly: AIQM1 and AIQM2 cover H, C, N, O. AIQM3 adds F, S and Cl. For anything outside those, use UAIQM.

Engines, and what runs where

On the Hub, calculations run through MLatom, with PySCF as the default engine, xTB for semi-empirical work and CP2K for periodic systems. No licence of your own is needed.

Commercial programs such as Gaussian are not available on the Hub. What Protomia can do is prepare the input files for them, so you can run the job on a machine where you hold a licence and bring the output back.

Getting your molecule in

Give it whatever you already have:

  • a common name — "benzene", "triphenylamine"
  • a SMILES string
  • XYZ coordinates pasted into the conversation
  • an uploaded file.xyz, .pdb, .sdf, .mol, .json, or a Gaussian .gjf / .com
  • a picture of a structure — it will read the structure out of the image

An uploaded structure is authoritative: your coordinates are used as given, never regenerated from a name lookup.

Real jobs, not a chat about jobs

Protomia submits actual cluster jobs. It reports the queue state, and when a long calculation finishes you can pick the conversation back up where you left it, including in a later session. Ask it to check what is running, or to cancel something.

Results you can use

Structures render in the conversation, so you can look at a geometry without downloading it. Cube files come back ready for isosurfaces. Reaction work produces a spreadsheet of the extracted energies so you can check the numbers yourself. Everything lands in your own working directory.

Beyond the calculation

This is where a calculation usually turns into work you can publish.

Writing and publishing

  • Draft a manuscript or report in LaTeX from results you have just produced — a reaction study, a benchmark, a spectrum — rather than from a blank page.
  • Revise a Word manuscript against reviewer comments, with citation insertion handled deterministically through EndNote.
  • Prepare reviewer-response materials: parse the editorial email, scaffold the response letter, and keep every reviewer comment verbatim.
  • Create, edit, translate or repair Word documents directly.

Literature and knowledge

  • Search the public literature, then verify DOI and arXiv metadata rather than trusting a recalled citation.
  • Ask about MLatom itself — methods, input keywords, the Python API — answered from the documentation rather than from memory.
  • Learn from curated material, including the MLatom tutorials and the Computational Chemistry and AI course.
  • Get dataset recommendations for machine-learned potentials, filtered by the elements and properties you need.

Your own methods

  • Bring your group's own skills — reusable instruction files describing preferred methods, constraints, templates or workflow steps — and Protomia applies them consistently through the session, so project conventions survive from one calculation to the next.

Your workspace

  • Run shell commands and Python in your working directory, and read the logs.
  • Navigate and edit code — search a repository, inspect git history, make structured edits. Several users bring their own simulation code here rather than a molecule.
  • Read PDFs and fetch pages as part of a task.

What it will not do

  • It will not invent a geometry when you have given it one.
  • It will not quietly substitute a method you did not ask for. If the one you named is a poor fit, it says so and proposes an alternative.
  • It will not fabricate a citation — references are verified against DOI and arXiv metadata.
  • It is not a substitute for knowing what you are computing. It will set up, run and explain the calculation; the science stays yours.

Where to try it

Protomia runs on the Aitomistic Hub. Registration is free, and a free tier of compute is available for real calculations.

On this page