Appearance
Chai-1 Local Structure
Chai-1 is a local molecular structure prediction runtime. In Liatir it is available as a managed AI Model for Linux CUDA machines.
What it does
Chai-1 predicts molecular structures from biological sequence and molecular inputs. It is part of the protein structure prediction family in Liatir.
When to use it
Use Chai-1 when you have access to a compatible Linux workstation or server with CUDA and enough GPU memory. It is not intended for normal macOS CPU or Apple Metal testing.
Inputs in Liatir
- Protein FASTA/FAA file, or an inline protein sequence.
- Optional ligand-related inputs when supported by the tool.
- Prediction options exposed by the Liatir protein structure runner.
Outputs
Liatir can produce:
- structure files when prediction succeeds;
- logs and runtime metadata;
- viewer-compatible artifacts;
- provenance with model, runtime, input, parameters, and output files.
Hardware and installation
The local Chai-1 runtime is restricted to Linux CUDA hosts in Liatir. It requires Python 3.10 or newer and a GPU with bfloat16 support. A100, H100, or L40S-class GPUs are recommended by the upstream project.
Liatir blocks installation on incompatible hosts and explains why.
Limits and cautions
Chai-1 is not a fallback for macOS or CPU-only testing. Use Boltz-2 CPU mode for local compatibility testing on machines without CUDA, understanding that CPU runs can be slow.
This model is currently hidden from normal Liatir catalog surfaces until a Linux CUDA validation environment is available.