AI Assistant

Ask for the Calculation, Approve the Plan

The assistant turns a question into a planned computation, explains why it chose that method, and waits for you.

Get Started Request a Demo

A chemist at a laptop in a bright lab asking the AI assistant a question, with a molecule and suggestion chips on screen

AI-Native, with a Hand on the Brake

The chat knows the page you are on, the Study you are in and the molecule you selected. Live

Plain Language

Type or speak the question. The planner selects the tool, the level of theory and the server, and says why.

Offer With an Estimate

An offer card shows time, memory and server before you queue anything. You press Queue.

Quotes Stored Values

Answers cite stored results with numbered markers, so a number can be traced to its computation.

Refine With

A ladder suggests the next rung of theory for the same molecule. Higher rungs are shown as planned until they exist.

Intent Chips

Context-aware shortcuts limited to an allowlist: open a tab, plan a tool, run again, open a source or export.

Your Own Models

Uses the AI models your organisation configured, with no model fixed in the code.

Built in, Not Bolted On

The AI-native Chemistry Lab, live since 2026-10-19. Live

Accuracy Tiers

Quick, Standard or Best available, each expanding to the exact method and basis set; experts can always pick the level by hand.

Plans You Review

A side sheet shows the planned steps, the checks run before queueing and the estimated cost, before you approve.

Failed-Run Diagnosis

When a computation fails or does not converge, a banner explains why from the engine's own diagnostics and proposes a fix.

Provenance Gate

Reports are blocked if a number cannot be traced to a stored result. Unverified engine data adds a Limitations sentence and a badge automatically.

Rights-Tracked Papers

The assistant reads only papers the organisation has the right to use, and always cites the original publication.

Fallback, Never Leakage

If the main model is unavailable, it falls back to the organisation's own other models, or queues the request and notifies you.

Evaluated before every release: 77 of 77 behaviour cases and 38 of 38 safety cases in the offline test gate.

A researcher in a dim analysis room looking at molecular orbital lobes on a wide monitor beside an orbital energy diagram

Describe a Molecule

From a Sentence to a Checked Structure

Describe by typing or voice; get up to five candidates, or a set of up to twenty
Each candidate is validated by the engine and compared with PubChem
A badge shows a PubChem identifier, or says "AI-proposed, not found in PubChem"
Nothing is used until you press Use

Safe by Construction

Content from outside is marked as data and never followed as an instruction
AI output is validated before it is stored or shown as a structure
Chip actions are limited to an allowlist, with rate limits
No personal data is sent to Crossref or PubChem

A security officer opening a locked data-centre server cage with a badge reader

AI Assistant Questions

Can the assistant start a computation by itself?

No. It answers with an offer card such as "Queue this for that?" with an estimate of time, memory and server. Nothing runs until you press Queue.

Does the assistant invent values?

No. It is built to quote only stored results and to mark them with citation numbers. Suggested sources must match author, year and title in Crossref, or they are dropped.

What are the suggestion chips?

Short prompts that appear in the chat column from the page you are on and your recent actions. They can only open a tab, plan a tool, run again, open a source or export. They contain no numbers and make no claims, and read-only users get navigation chips only.

Which AI models does it use?

The models your organisation has configured: local models on your servers, Bridge models (shown to clients as VYROX AI Combo) or cloud providers. No model is hard-coded.

Can I speak to it?

Yes. The molecule editor and the start box accept voice through the browser speech feature as well as typing.

Run Your Next Molecule Through AiMolSim

Connect your own server once, then let your team queue computations from the browser.