OmniScientist Data

Raw scientific data behind the runs in OmniScientist: An Omni-Modal Omni-Discipline AI Scientist, one folder per case. The agent reads these files directly, so what is here is what it saw.

• Paper: https://arxiv.org/abs/2608.13558
• Project page: https://omni-scientist.github.io
• Code: https://github.com/Omni-Scientist/OmniScientist

13 of the 36 cases in the paper are here. The rest are left out because their upstream terms grant no redistribution right, or because we have not verified that they do. The code repository carries fetch recipes for those, in docs/DATASETS.md.

Cases

Case	Modality	Items	Size	Upstream	License
animal_track	GPS trajectories	49	1.5 MB	Movebank MPIAB white stork study	CC0
birdaudio	field audio	2,000	1.76 GB	QMUL Bird Audio Detection	CC BY 4.0
calms21	video frames	2,500	274 MB	CalMS21	CC BY 4.0
chestxray	radiographs	3,000	2.73 GB	Kermany et al.	CC BY 4.0
comma2k19	driving frames	3,000	174 MB	comma2k19	MIT
dna	sequence table	1 file	5.1 MB	Nucleotide Transformer H3	see upstream card
feynman	equation tables	12	7.8 MB	AI Feynman / FSReD	AI Feynman terms
galaxy_xsurvey	survey images	105 galaxies x 2 surveys	2.75 GB	Galaxy10 DECaLS and SDSS	Galaxy Zoo / DESI Legacy terms
histopath	H&E tiles	16	0.8 MB	Kather Texture 2016	CC BY 4.0
kg_biokg	knowledge graph	8 files	1.4 MB	OGB ogbl-biokg	CC0
med_ct3d	3-D CT volumes	1,496	393 MB	MedMNIST v2	CC BY 4.0
stead_seismic	3-component waveforms	1,500	110 MB	STEAD	CC BY 4.0
supercon	material property tables	3 files	36.5 MB	UCI Superconductivty	CC BY 4.0

Download

One case:

hf download BradNLP/OmniScientist --repo-type dataset \
  --include "data/stead_seismic/*" --local-dir .

All of it, 8.24 GB:

hf download BradNLP/OmniScientist --repo-type dataset --local-dir .

Layout

data/<case>/SOURCE.md     upstream dataset, license, attribution
data/<case>/series.json   case specification, item list, labels
data/<case>/...           the raw files

series.json holds the fields the engine reads (role, subject, property, request) and a members list with one entry per item: file path, label, and whatever per-item metadata the case carries. stead_seismic members carry magnitude, epicentral distance and station code; galaxy_xsurvey members carry RA/Dec and which survey the image came from.

Paths in members[].file are relative to the case root the engine uses, which holds the raw files one level down in a subfolder. To run a case with the engine, restore that shape:

hf download BradNLP/OmniScientist --repo-type dataset \
  --include "data/stead_seismic/*" --local-dir .
mkdir -p cases/stead_seismic
mv data/stead_seismic cases/stead_seismic/data
mv cases/stead_seismic/data/series.json cases/stead_seismic/

There is no train/test split in this repository. The Hub viewer covers the image cases; the audio, waveform, volume, table and graph cases have no Hub loader and need a download. The evaluation protocols used in the paper are listed in docs/DATASETS.md.

Licensing

Each case is redistributed under its upstream license, named in that case's SOURCE.md, with attribution to the original publisher. Check that file before redistributing anything from here. Cases whose upstream terms grant no redistribution right are not in this repository. Nothing here is intended for clinical use.

Citation

@article{omniscientist2026,
  title   = {OmniScientist: An Omni-Modal Omni-Discipline AI Scientist},
  author  = {Li, Bobo and Fei, Hao and Ju, Tianjie and Lee, Mong-Li and Hsu, Wynne},
  journal = {arXiv preprint arXiv:2608.13558},
  year    = {2026}
}

If you use a single case, cite its upstream dataset as well.