Zero-shot time-series forecasting with Google's TimesFM 3.0 foundation model. Give it a history, get a quantile forecast — no training, optional covariates. Weights are under the TimesFM Non-Commercial License v1.0.

Solves crystal structures from nanocrystalline powder X-ray diffraction: pass a two-column pattern (.xy/.xye, Q or 2theta) and get ranked candidate CIFs with pattern-fit statistics. CDVAE-based graph diffusion (Guo et al. 2024).

Dated numeric series ready to forecast: search and retrieve any of FRED's ~800k macroeconomic series with unit transformations and frequency aggregation, plus relative Google Trends search interest for up to five terms.

is a lightweight EDM diffusion transformer for crystalline materials discovery. Generate novel crystals de novo (atom types + coordinates + lattice) or predict structures for fixed compositions (CSP). Trained on MP20; optional MPTS-52 CSP checkpoint for larger cells.

Compute the magnetocrystalline anisotropy energy (MAE) of a magnetic unit cell from a CIF by first-principles DFT (ABACUS LCAO DZP + TB2J split-SOC force theorem). Built for large cells (~20+ atoms) that fast ML predictors don't cover. Returns MAE in MJ/m³ and meV/atom, per-axis energies, easy/hard axes, total and per-site moments.

Predict ZT_max and thermoelectric properties for inorganic crystal structures using first-principles methods: phono3py BTE for lattice thermal conductivity, BoltzTraP2 for electronic transport from ABACUS PBE±SOC bands, and MEGNet-HSE scissor correction when DFT is gapped without SOC. Falls back to Slack + SPB models when the first-principles path is unavailable.

Native agent-authored routes for apollo.

Generate novel metal-organic framework (MOF) structures as CIF files with MOFFlow-2, a torsion-aware flow-matching generative model. Pass a sample count (1-20) and optionally condition on a target CO2 uptake value; get back a zip bundle of CIFs plus per-structure stats (formula, atom count, density).