Governed
scientific discovery.
Build matter. See it. Rule on a plan and run it on real engines. Keep each result as Evidence with where it came from, on a thread that remembers how the science got there.
For materials scientists, chemists, physicists, engineers, inventors and R&D groups.
Scientific computation does the calculation.
AI helps reason around it.
Kymtx records which is which.
An investigation lives across papers, databases, Python environments, simulation packages, notebooks, AI chats, provider portals and structure viewers, carried between them by hand.
Copied from a run into a chat into a document, it no longer says which structure, method or version produced it.
A fluent paragraph and a converged calculation sit side by side with nothing to tell them apart.
Reproducing how a conclusion was reached means rebuilding the trail from memory.
Kymtx gives the investigation one governed workspace.
A thread is a persistent line of scientific inquiry, not a chat. It runs through one governed chain.
From the Matter Library, or by hand.
What the investigation is trying to establish.
Bounded accounts: structure, electronic, thermodynamics, reference, adversarial.
M frames what must be known and drafts the plan.
The plan is held on the Judgment Ledger until it is ruled.
The contract is checked and the work goes to a READY door.
A classical engine or a simulator runs the approved task.
Matched to its request; malformed output is held.
With its engine, input, run and class.
M reconciles it against the hypothesis and poses its reading.
Every step, as one graph.
Current state and the next scientific work.
M decides what must be known. SES decides how to obtain it from the READY doors. The Principal judges.
Atomic draws a structure in 3D, in 2D or as an outline: atoms and species, bonds, the unit cell and lattice, with inspection and measurement on the structure. It opens on diamond and says on screen what that is: a reference structure, not project Evidence.
A species is coloured from its observed emission lines in the NIST Atomic Spectra Database. The source and its retrieval date are on screen. Of the 118 elements, 98 take a visible or composite colour, 1 is shown in labelled false colour, and 19 have no lines and are drawn dashed. The conventional CPK colours are one click away.
Inspect the cell and its volume, repeat it, project along a lattice axis. Diamond opens as a 2×2×2 repetition of its conventional cell.
Two structures side by side, each with its own provenance, in overlay or difference view. When the compositions differ, Compare says so and maps nothing.
A library structure reads as a database reference. A structure built by hand reads by its own state. Neither is Evidence until a run makes it so.
Choose matter at any level, from isotope to molecule to crystal, and carry it into Atomic as a reference structure with its source named. Nothing in the library is project Evidence until a run makes it so.
Carbon, 12C, C2+, diamond, H2O, NaCl, fcc. Molecules, crystals, compounds, ions and isotopes each have a tab.
Built structures are kept in a library of the user's own, beside the references.
Molecules and bulk prototypes from ASE; isotopes from the NIST Atomic Weights and Isotopic Compositions; colours from the NIST spectral signature. The page's own footer says so.
ASSESS STRUCTURE validates the structure, shows the route decision for every engine with its reason, and poses the plan to Judgment. SES executes nothing until the Principal approves it.
Every structure wears a provenance label: DATABASE, USER BUILT or GENERATED today. More labels, and file import and export, are to build.
The Workbench is the present-tense scientific desk: the thread and its objective, the current scientific state, the structure in Atomic, the Evidence, the Quantum execution, the Corpus, the Manager's account, the next scientific work, and any authority the Principal still has to give.
Which structure, what would execute, what was excluded and why, and the external cost.
Validation, seats, plan, principal, SES, reconciliation, judgment: how far along the chain each assessment is, and which engines were selected, unavailable, not applicable or not justified.
Each node of the current state opens its surface on this thread, and each assessment links to Matter, Assembly and Simulation as well.
The Scientific Execution Steward stands between reasoning and execution. It validates contracts, routes the work, matches each return to its request and holds malformed output. It is not a reasoning or judgment seat: it cannot change scientific intent or rule on anything.
The approved plan goes out through SES; each result comes back through it and is written as Evidence by one writer.
A structure, a composite and a scene are all assessed through the same create, plan, pose, ruling, execute and reconcile chain.
A probe and a smoke test, never that a module merely imports. SES selects only from READY doors, and a capability that is missing is refused by name, not guessed.
The pose carries the execution plan: what is selected, what is not and why, and the external cost. Local runs read $0.
An Evidence row is a result with its origin attached: its class, method, source, uncertainty, the run that produced it and the conditions it ran under. It reads as a table, a plot or a field. Kymtx does not present every scientific-looking object as scientific fact.
A value read from a source, a value an engine computed, a simulator's result and a measured value never share a label.
A row can read SUPPORTED. That is the machine's reading; the verdict stays on the Ledger. Model proof is never physical proof.
A row a newer version replaced reads SUPERSEDED wherever it is cited, including from the ruling that rested on it.
A public datasheet or standard enters by its hash and page, posed to the Ledger first, and the same bytes are recognised if they arrive again.
The full vocabulary of classes, support states and verification levels is in the Docs.
Corpus draws one thread as a graph from its origin to now: the objective, the structure, the calculations, the runs and the Evidence each produced, joined by what was derived from what and what produced what. It is the same thread the Workbench and Evidence show, seen as history.
The counts sit under the graph by kind. Step between threads, fit the graph to the window, and read the zoom on the status line.
A failed run, a refused capability and a result ruled no stay in the history, not deleted as noise.
Follow one result back to its source across branches of a line of inquiry.
Kymtx is not a quantum computer app. Classical computation is used where it fits, simulation where it fits, and physical hardware only where it has a purpose. The Quantum surface shows a preparation as a chain: structure, active space, Hamiltonian, mapping, program, SES, backend, measurement, Evidence.
H2 in the STO-3G basis, prepared through PySCF: −1.845625 Ha on the exact simulator and −1.844728 Ha on the shot simulator at 4096 shots, each its own Evidence row.
The screen says which one ran, and a simulator's row carries the limitation of its model. One is never quietly answered by another.
Through the user's own provider account, with the scientific problem kept Kymtx-native. No provider is offered until its run has been walked.
A proposed scientific action is posed, held, ruled by the Principal, returned with proof, and acknowledged. The Ledger governs actions and judgment. It does not decide physics.
Only approved routes run, each recorded with its exact input, runtime and raw output.
Every pose carries the identity and digest of the records behind it, and still resolves to them after newer versions exist.
Twenty materials-science personas ship built in. A persona analyses, challenges and recommends; it acquires no authority.
Made on version 0 of a record, it still resolves to version 0 after version 1 exists, and says a newer one replaced it.
When the runs return, the Manager's reading of them against the hypothesis goes back to the Principal as its own decision.
The mushroom is on the Stage and in the Theater, as in Chorix.
The Principal rules on: scientific direction, paid and external execution, what leaves the machine, which candidate advances, and whether an AI conclusion is accepted.
Provenance is the trail behind a result. Reproducibility is being able to walk it again. A result without its computational environment is incomplete provenance.
A kind, an id, a version and a digest, for structures, assemblies and runs alike.
Backend and version, action, input, runtime and raw output on every run. Data files it depends on are checksummed again at run time.
A datasheet or standard enters by its hash and page, Ledger first.
The whole computational environment as one object, an evidence package a reviewer can open, and reference calculations shown beside Kymtx's own.
Kymtx runs on the machine where it is installed and keeps its record in a database in its own data folder there. The AI seats run on the subscriptions and keys the user already holds. Setup stops for the Principal at identity, legal, financial and IP-egress boundaries.
The engine does the calculation, and the Evidence row names it and its version. Setup probes the machine, finds what is installed, builds a plan and runs local setup under one approval.
Molecular electronic structure, structure operations, quantum program preparation. You do not have to assemble a scientific Python and quantum environment by hand.
Each waits on something the user supplies: a pseudopotential library, a thermodynamic database, an API key. The door says which.
Kymtx runs natively on Windows; the scientific engines run in a Linux runtime Setup points at.
A door reads one of twelve states, each with the sentence that says what is missing. They are listed in the Docs.
Implemented means merged in the app with a proof on file; each such row still owes the Principal's own walk. Partial names the part that runs. Planned is designed, not built. Research target is a horizon, not a feature.
Build, assess, rule, run, read the Evidence, reconcile.
Corpus draws the path.
Evidence per candidate, a comparison, a ruling on which advances.
The simulator half runs today.
Recorded demonstrations of real threads, from diamond in Atomic to a result traced back through Corpus, are to make.
A course as a workspace, with a stack per subject, and a one-seat trial for departments.
Fifteen Patent & IP personas ship today. Claim and evidence tracing is to build.
Ten Scientific Publication personas ship today. Flagging a statement when its Evidence is superseded is to build.
Kymtx is planned as a subscription; Chorix Studio and Pro stay a single purchase. The Kymtx fee will never include what outside science costs unless it says so in plain words.
Chorix is human-governed multi-AI orchestration. Ensemble runs it for a team. Kymtx is a standalone scientific workstation, not an edition of Chorix: it inherits the family's governance and adds scientific objects, execution, Evidence and provenance.
None of them. It is the workstation that joins them: the user's own AI models propose, scientific engines calculate, and the Principal rules. No quantum computer is needed.
No. The engine performs the calculation. Kymtx validates the request, routes it, records the run and keeps the result.
No. Evidence is written from the engine's return by one writer. An AI seat can read it and give an account; the account is posed for a ruling and stays labelled an account.
The record lives in the app's data folder on the machine. What a seat is asked goes to the AI provider the user chose for it. A full privacy statement is owed before this page publishes.
No, and the screen says which is which: a sphere is a representation, a spectral colour comes from observed lines, false colour is marked, CPK is a convention.
Not yet. Kymtx runs on Windows, with the engines in WSL2.
Kymtx uses reference data from the NIST Atomic Spectra Database and Atomic Weights and Isotopic Compositions, and runs ASE, PySCF, Quantum ESPRESSO, pycalphad and Qiskit Nature, with Materials Project as an optional data source. Each is the work of its own authors. What each is used for.
Results depend on their models, assumptions and inputs, and need interpretation. AI output may be wrong. The user remains responsible for scientific conclusions.
Drafting assistance creates no attorney-client relationship, and inventorship is a legal determination for qualified counsel.
