Start free — three reports, no card.
What a run produces
A report with six parts, in this order, because that is the order a reviewer reads them in:
- The question, restated as constraints. Not “corrosion resistant” but the property, the operator, the value, the unit and the condition. If a requirement cannot be made testable, it is recorded separately as a qualitative requirement rather than quietly turned into a number.
- Candidates considered — including the ones that did not survive, with the reason each was dropped. A shortlist without its rejections is a conclusion, not an analysis.
- The values, each with its source, its entry identifier, its condition and its evidence level.
- The comparison — screening against constraints, then ranking, with derived quantities carrying the level of their weakest input.
- What the report does not cover. Mandatory, never empty.
- Sources, each one a link that lands on something containing the number.
What it will not do
It will not give you a number it cannot attribute. If the open databases do not hold a value, the report says the value is not established and why, rather than producing a plausible figure. This is the single most common complaint about general-purpose models applied to materials work, and the reason we exist.
It will not decide. It does not tell you which material to use. It assembles the material for your decision, in a form you can defend, and the responsibility stays where it already was.
It will not pretend to have data it does not have. Commercial grade designations bridge incompletely to composition-indexed databases. Formulated materials — polymers, composites, coatings, concretes — are outside our data layer entirely. The full coverage picture.
How it differs from asking a general model
A general-purpose model will answer a property question fluently and will usually be close. The problem is not accuracy in the average case; it is that the answer arrives with no way to tell the good cases from the bad ones, and no way for you to check.
Under the E-ladder, a value stated without a resolvable citation is E0 — inadmissible in design documentation regardless of whether it happens to be right. The agent’s job is to move values up that ladder by finding what they rest on, and to say so when it cannot.
You can test this in about two minutes, on any tool including ours: take a number, try to reach its exact source. That test, formalised, is the tool audit.
Frequently asked
Is this a wrapper around a language model? A language model does the reading and the writing. The databases do the facts, and every value in the output is traced to one. The difference you can check is whether the citation resolves — run the same question here and in a chat window and compare what you can click.
Can I use the output in design documentation? That is what it is for. Read the limitations section first, and note that a value below E5 used as a design allowable requires your explicit acceptance — the report records that as your decision, not ours.
What if it gets something wrong? Report it. Every value has a source, so a wrong value is a traceable wrong value, which is the whole point of building it this way. We do not claim accuracy we have not measured.
Does my data leave my network? Queries are processed in the cloud. If your project data cannot leave your network, this product is not currently for you and we would rather say so now.
What this page does not cover
- Measured accuracy. We have not published an accuracy benchmark. When we do, it will include the cases we lose.
- On-premise deployment. Not available. It is the condition for working with organisations that cannot use cloud tools, and it is not built.
- Speed under load. No published latency figures.