AI Models
Liatir exposes only AI Models distributed as signed Runtime Boxes. Each box is an immutable, target-specific bundle containing the model, interpreter, dependencies, scientific runner, and legal notices tested together.
Available models
| AI Model | Input | Published support |
|---|---|---|
| Geneformer V1 10M | human AnnData | macOS arm64 Metal; Linux CUDA 12.9; Windows through WSL2 |
| scGPT Whole-human | human AnnData | macOS arm64 Metal; Linux CPU; Linux CUDA 12.9; Windows through WSL2 |
| UCE 4-layer | multi-species AnnData | macOS arm64 Metal |
| Boltz-2 | protein, DNA and RNA sequences; small molecules | Linux CUDA 12.9; Windows through WSL2 |
| Protenix base v1.0.0 | protein, DNA and RNA sequences; small molecules | Linux CUDA 12.6; Windows through WSL2 |
These are the complete product catalog. Models without a published Runtime Box are not shown as previews and cannot be installed through a legacy path.
Installation and removal
AI Models are installed globally for the app, not per workspace. Liatir checks host compatibility before downloading, verifies the selected box before activation, and records the exact activated release. Removing a model removes its Runtime Box from the device.
Running a model
The single-cell models run through Single-cell Embedding, either directly from the model page or as a pipeline step. Boltz-2 and Protenix run from the Structure Prediction page under Tools, and Boltz-2 also from Protein–Ligand Affinity. Runs produce Jobs, Results, durable output files, and Runtime Box provenance.