The questions people actually ask.

You describe a system in plain language — or upload a paper or PDF — and AutoSim writes the simulator, proposes parameters with cited priors, fits them to data by Bayesian calibration, tests the fitted model on data it never saw, and returns a scored model with a readable report. All six stages run as one pipeline.

No. You need to be able to describe the system and what you are trying to decide. The generated Python is there if you want to read or edit it, but nothing in the workflow requires you to.

A description of the system, and data if you have it. With data, AutoSim calibrates and validates against it. Without data, it still builds and runs the model but will restrict the claims it is willing to make and tell you so.

Minutes for a first calibrated version of a typical model. Heavier calibrations take longer because they are doing real inference, and progress is shown per stage rather than behind a spinner.

Yes — 500+ scenario templates across 20 domains, plus every model published on the marketplace, can be opened, forked and recalibrated on your own data.

You do not take our word for it. Parameters are fitted by ABC-SMC, which reports posterior uncertainty rather than a single guess; the fitted model is then scored on held-out data. Fit and out-of-sample performance are reported separately, so a model that fits beautifully and generalises badly is visible as exactly that.

No. The AI proposes structure, code and prior ranges. Every reported number comes out of executed code. AI narrative is never blended into a numeric score.

AutoSim screens for that before calibrating. If the data cannot distinguish between parameter settings, it says so, narrows the claim class, and tells you what additional data would resolve it — rather than returning a precise-looking answer that is not supported.

Every stage leaves a dated artifact: the specification, the priors and their citations, the calibration posteriors and convergence diagnostics, the out-of-sample results, and the stated limitations. The model card exports with all of it.

Yes — that is the primary benchmark. The engine is run against problems whose true parameters are known in advance across multiple domains, and the record of whether the credible interval covers the truth is published on the proof page.

Simulations execute in your browser through WebAssembly, so the numerical work happens on your machine. Your text description is sent to the AI to generate the model; your dataset is not required to leave the browser to be calibrated against.

Yes. Private and on-premises deployment is part of Enterprise, along with SSO and audit logs.

You do. You can keep them private, export them, or publish them to the marketplace on your own terms.

No, it is a real plan. The free tier includes the full six-stage pipeline and unlimited in-browser simulation, with a daily AI token allowance.

AI usage only — generating and revising models, reading documents, proposing priors. Running simulations costs you nothing on our side because your browser does that work.

Generation pauses until the allowance resets; already-built models keep running locally. Usage and remaining allowance are shown live, and quotas are hard by default so a plan cannot overspend.

Yes. Publish a calibrated model to the marketplace and earn on every run other people make with it.

Yes — that is AlphaIQ's bespoke engagement path, for teams who want the model built alongside them rather than self-serve.