Run the science — experiments, 140+ data sources, on your own cluster

01 — 140+ scientific data sources

One surface over the data you already use.

A unified registry reaches space agencies, particle physics, gravitational-wave, genomics, neuroscience, medical, climate, materials and more — one API surface, with an honest "not wired" response when a source isn't configured rather than a fabricated answer. For research →

02 — Experiment engine

Design and track experiments.

An early experiment engine helps frame a question into phases and sub-tasks and tracks the run. (Phase one — the planning and tracking surface; we describe what's here today, not a finished autonomous researcher.)

03 — Forecasting on your telemetry

Extend a series, natively.

The structured-ML forecasting workload runs natively inside the main service, no separate daemon — give it a recent window of a series (a sensor stream, a load curve) and it extends it. It forecasts a series from its recent pattern; it does not predict a real-world event.

04 — Forward simulation of named domains

Roll a known model forward, on CPU.

Compact files can roll the dynamics of a named physical domain forward through a known model, on an ordinary CPU — e.g. bodies moving under gravity. You give an initial condition and a number of steps and get the step-by-step evolution to animate or analyse.

Forward model, not inversion or prediction. This is a forward simulation of a known model — not an inversion of field data, and not a prediction of a real-world event. It is faithful over a bounded window (chaotic systems are precise at short horizons and diverge at long ones — that's physics, not a flaw).

05 — Watch it render

See it render, not just describe it.

Ask Eldric about a real, cataloged earthquake in plain language and it doesn’t hand you a paragraph — it computes the seismic wavefront and draws it spreading through a cross-section of the crust, frame by frame, right in the chat. For a large event you watch the wave radiate from the length of the fault rupture and bend as it crosses the crust’s velocity layers. It runs natively on CPU — no GPU, no cloud, no external service — on infrastructure you control. Try it in the chat →

Real events only, and honest about it. Eldric renders wavefronts for events that exist in a real catalog; ask about something uncatalogued and it tells you it can’t find it rather than inventing one. This is a visualization of a known physics model — how the wave spreads from the fault through the crust’s layers — not earthquake prediction and not an inversion product.

Get started

Bring the science in-house.