Pre-trained vision-language-omics transformer for glioma.
Pre-trained vision-language-omics transformer for glioma.
GLIO-AI is one of 11 subsystems that compose AURORA-GLIO. It is independently usable but shares the same patient representation, provenance log and governance layer as every other AURORA module.
GLIO-AI is the per-module foundation stack: a vision-language-omics transformer pre-trained on permissively-licensed neuro corpora, then conditioned on glioma-specific tasks. It is distilled to clinic-edge sizes so it runs on hospital hardware. It is patient-conditioned but never patient-identifying.
The foundation pass is the cheapest layer for downstream subsystems to call. GLIO-DIAG, GLIO-INFIL and GLIO-RAD all start from its embeddings. The weights are versioned, the training data hashes are pinned, and the model card is published alongside the release.
GLIO-AI can be installed standalone, or pulled in as part of the full AURORA-GLIO stack. Both ship the same code and weights, signed end-to-end. Pilot partners have access today; public alpha opens Q4 2026.
pip install aurora-glio[ai] aurora glio.ai download # signed, hash-verified aurora glio.ai embed case.aurora --layer last
GLIO-AI moves faster when the right people are in the room. Whether you carry a service line, a registry, or a few quiet weekends — there is a way in.
Sponsor GLIO-AI for your unit. Co-design the surface, the audit log and the override semantics with the team that built it. Pilot sponsors keep a council seat for the duration.
Run a research question on GLIO-AI. Federation means you can train across populations you cannot see alone. Whitepapers carry co-authorship, not acknowledgements.
GLIO-AI ships starter tasks tagged 'good-first-issue' on the pilot repo. SDKs in Python, TypeScript and Rust at public alpha. MIT, no CLA.
AURORA is in private pilot today and opens to the world at public alpha in Q4 2026 under MIT. If you carry the weight of these diseases — as a clinician, scientist, builder, patient or advocate — there is a seat at the table.