The questions
worth answering
properly.

No. The reasoning core is a hand-encoded knowledge corpus — senior process safety engineers have built it entry by entry from major incident investigations, HAZOP guideword logic, LOPA methodology, IEC 61511, NFPA, OISD, API standards, and the documented failure modes of real equipment in real energy transition assets. That corpus is what Aby reasons from.

The large language model writes Aby's answers in plain English. It does not do the reasoning. She does. The practical difference: a model trained on the internet will produce a plausible-sounding HAZOP. Aby retrieves from a corpus built by engineers who have run real HAZOP studies on real plants. Every finding is cited to its source. Every answer carries a confidence score.

A search engine finds documents. A document management system stores them. Aby reasons across them simultaneously — holding the live plant state, the incident record, the regulatory requirements, and your operating procedures in a single coherent view and telling you what the combination means.

The Texas City explosion in 2005 killed fifteen people. The information needed to prevent it existed in the plant's data — in three separate control rooms, in maintenance records, in near-miss logs. What did not exist was a single view of what all of it meant together. That is exactly what Aby provides.

Because a DCS was never built to do what StarFarer does, and StarFarer isn't built to do what a DCS does. Your DCS watches temperature, pressure, and flow in real time and pulls the trigger on an interlock the instant a value crosses a line. That is its entire job, and it is essential, fast, and automatic — StarFarer does not compete with it, monitor in its place, or touch your control systems at all.

What a DCS has no way to do: read a P&ID and reason about the hazard behind a deviation, write or check a HAZOP, run LOPA arithmetic, or notice that a five-year-old safety case has gone stale as the plant physically changed around it. That work has always sat with a human process safety engineer — on paper, in their judgement, and walking out the door with them when they retire. StarFarer is built for that gap specifically, not the one your DCS already covers.

VyroNyx is a process safety and energy engineering practice. The corpus was built by senior engineers working on real studies — HAZOPs, LOPAs, safety cases, regulatory submissions — encoding reasoning chains, not documents. It currently holds 966 entries, 18 major incident case studies, and 24 regulatory frameworks.

This is slow to build and impossible to shortcut. It is also exactly why it produces useful output rather than plausible-sounding noise. Our paper on propagation containment in solar-BESS-electrolyser architectures was accepted at IChemE Hazards 36 — that research is encoded in the corpus.

Yes. Upload a P&ID as an image or PDF and Aby will extract instruments, equipment, and flow paths, map the topology, identify the node boundaries, and begin HAZOP analysis on the structure she reads. She handles scanned drawings, aged PDFs, and layered CAD exports.

P&ID reading is the capability our engineering users test first and trust most. It is the reason the platform exists — the ability to connect what the drawing shows with what the incident record and regulatory framework say about that specific configuration.

Every answer Aby produces carries a confidence score from 0 to 100, with visible components: corpus match strength, reasoning chain length, regulatory citation coverage, and assumption count. A score above 80 means the answer is strongly grounded in the corpus. A score below 50 means Aby is extrapolating — and she will tell you so.

SPECULATIVE means speculative. In process safety, a confident wrong answer is the most dangerous possible output. The Aby Number is built specifically against it. We would rather the score say 40 and be honest than say 90 and be wrong.

Yes — and we would rather tell you that directly than have you find it halfway through evaluating us. There is a real and growing category of AI-assisted process safety software, including tools built specifically for HAZOP acceleration. If you are doing due diligence, you should find them, and you should ask us how we compare.

What we believe is different, and what is worth actually testing rather than taking our word for: a visible, componentised confidence score on every single answer — not just fluent-sounding text — that tells you when the corpus is thin instead of hiding it. A Living Safety Case that keeps tracking whether a study has gone stale as your plant changes, rather than a one-time report that's accurate the day it's issued and silently wrong a year later. And an AI HAZOP Committee that argues a finding from four independent perspectives before a chairman synthesises them, surfacing disagreement as a minority report instead of smoothing it away.

Test any AI process safety tool on whether it tells you when it doesn't know. That is the actual differentiator in this category right now — not whether it can read a P&ID at all, which several tools can now do.

Never. Your data serves only your queries. It is not shared with other clients, not used to improve Aby's answers for anyone else, and not folded into the general corpus without your explicit written consent. This is written into the DPA we provide.

The architecture enforces this at the data layer — your plant data lives in isolated storage, accessed only by your session, with no pathway to cross-client learning.

On Supabase infrastructure with servers in Mumbai (ap-south-1 region) for Indian clients, and regional equivalents for US and EU deployments. Data at rest is encrypted. Data in transit uses TLS 1.3. We do not use your data for any purpose outside of answering your queries.

Request deletion at any time. We comply within 48 hours and confirm in writing. All plant data, session history, and uploaded documents are permanently deleted. A platform that traps your data doesn't deserve it.

Yes, for any client with EU operations (GDPR requirement) or any client whose internal policy requires one. We provide a standard DPA at contract stage. If your legal team has specific requirements, we will work through them.

What we do have: encrypted storage, TLS in transit, isolated client data architecture, role-based access, and written data handling commitments in the DPA. For most engineering teams evaluating a technical platform, these are the substantive controls — the certification is the audit that confirms them.

StarFarer uses multiple frontier models routed by task type — the best available model for each reasoning requirement. We do not disclose the full routing architecture for commercial reasons, but the design principle is explicit: no single vendor dependency. Your plant intelligence is not hostage to any one provider's API terms or pricing changes.

What matters to you: Aby's reasoning comes from the corpus, not from the model. The model writes the answer. The corpus determines whether the answer is right.

Confident where the corpus is strong. Cautious where it isn't — and Aby will tell you which is which via the Aby Number. For HAZOP analysis on hydrogen, solar, BESS, and process safety domains, the corpus depth is significant. For highly plant-specific questions where your facility data hasn't been loaded, the answers are general and the score will reflect that.

The rule: use StarFarer to accelerate the work of an experienced engineer, not to replace one. The output is as good as the engineer reviewing it.

This is not a limitation unique to AI — the same applies to any analysis tool. What StarFarer does is produce the analysis that the engineer then reviews, rather than the engineer producing it from scratch. The competent review step is unchanged. The time to reach it is significantly shorter.

The Aby Number exists precisely because we expect this to happen and want you to know when. When Aby is uncertain, the score is low and the answer is flagged SPECULATIVE. When the corpus doesn't cover a specific scenario well, she will tell you that rather than invent something plausible.

When a factual error does occur — wrong regulatory clause, incorrect equipment parameter, a connection the corpus should have made but didn't — we want to know. Flagged errors are reviewed by our engineering team and, where confirmed, the corpus is updated. The platform gets more accurate with every correction.

StarFarer accelerates what a competent engineer can do and extends their reach — queries answered in seconds instead of days, a propagation map that would take weeks to build manually, a regulatory gap register that updates as standards change. The combination of a strong engineering team and StarFarer is the target state. StarFarer alone is not.

If you are looking to replace your process safety function with a software platform, we are not the right product and this is not the right conversation.

StarFarer is a decision-support tool. Professional liability for engineering decisions made using its output rests with the competent engineer who reviewed and acted on them — the same position as any engineering software tool, from simulation packages to HAZOP facilitation tools.

Our Terms of Service reflect this clearly. We do not accept liability for engineering decisions made on the basis of unreviewed AI output. No responsible engineering software vendor does, and any that claims otherwise is misrepresenting the nature of professional engineering responsibility.

For the Tier 1 Baseline: a node list from your P&IDs, a fluid inventory, and a marked-up plot plan. Documents every operating facility already holds, in whatever format they exist — scanned drawings, PDFs, Excel registers. Translating your data into the format StarFarer needs is our job during onboarding, not yours.

Tier 2 adds: equipment register, safeguard inventory, maintenance history. Tier 3 adds: incident and near-miss logs, operating procedures, alarm historian export. Each tier is optional and delivers standalone value.

Five working days from receipt of your Tier 1 data. You receive: HAZOP drafts for every node in your P&ID, your propagation map showing how failures travel through the topology, dispersion and jet-fire physics for your specific fluid inventory, and a clause-level regulatory gap register against the standards applicable to your facility.

This is delivered on your actual plant, not a demo environment. You judge the quality of the output before committing to anything.

Freelancer — $99/month: Full platform access, general safety intelligence, up to one plant loaded, standard support. Suited to independent process safety consultants.

Consultancy — $349/month: Multi-user access, up to five plants, priority support, team query history. Suited to small consultancies running concurrent client studies.

Enterprise — $1,199/month: Unlimited users and plants, dedicated onboarding, DPA included, SLA, custom integrations on request. Suited to operating companies and large consultancies.

All tiers begin with the free Baseline. You do not commit until you have seen what the platform produces on your own plant.

Yes — that is the Baseline. Submit your Tier 1 data. We run the full analysis. You receive the output in five working days at no cost. If it isn't useful on your actual plant, you owe us nothing and walk away with the analysis.

We structured it this way deliberately. A demo on sample data tells you nothing about whether the platform works on your plant. Your own plant tells you everything.

If live system integration is a requirement for your use case, raise it in the initial conversation. We will tell you honestly whether it is on a timeline that works for you or not.

Nothing happens automatically. There is no credit card required for the Baseline. At the end of five working days you receive your output. If you want to continue, we have a conversation about the right subscription tier. If you don't, the plant data is deleted on request and that is the end of it.

No automatic billing. No dark patterns. No renewal reminders designed to catch you out.

An engineering company that builds AI. The distinction matters. VyroNyx delivers process safety and energy engineering as a professional practice — HAZOP studies, LOPA analyses, safety cases, regulatory submissions. StarFarer was built inside that practice to solve problems we encountered while doing real engineering work, not alongside it.

The people building the platform are the same people who ran the studies the corpus is built from. That is not a common architecture for AI product companies, and it is the reason the output is useful to engineers rather than impressive to investors.

What we can tell you: your plant data is exportable at any time in standard formats. The analysis StarFarer produces is yours — not locked in a proprietary schema. The regulatory frameworks and incident corpus that underpin the reasoning are encoded knowledge, not a black-box model you lose access to when a company closes.

The continuity risk is real. The mitigation is that what you get from StarFarer is useful engineering output you own, not a dependency on a platform that could disappear.

Yes. Green hydrogen is the primary vertical the platform was built toward. The corpus covers electrolyser safety, BESS thermal runaway propagation, hydrogen embrittlement, hydrogen dispersion modelling, NFPA 2, ISO 22734, MNRE and PNGRB regulatory frameworks for India, and IEC 62443 for hydrogen facility control systems.

The IChemE Hazards 36 paper accepted for oral presentation covers propagation containment in integrated solar-BESS-electrolyser architectures — the topology that defines green hydrogen FEED studies. This is the specific frontier the corpus is deepest in.

Paper 162: propagation containment in integrated solar-BESS-electrolyser architectures. The core question: when a BESS thermal runaway event propagates to adjacent electrolyser equipment, what are the credible consequence scenarios and what containment measures are effective across that specific multi-hazard topology?

This is a domain where very little published guidance exists and where the energy transition is building large facilities right now. The paper is in peer review for full publication in 2026. The reasoning from that work is encoded in StarFarer's corpus.

Email contact@vyronyxgroup.com or reach us on LinkedIn at linkedin.com/company/vyronyx-energy. We read everything and respond to genuine enquiries. If your question reveals a gap in this FAQ, we will add it.