Supported libraries
Calcify maintains a mapping from common Python libraries to their Chelis equivalents. When a translation encounters calls to a supported library, it emits the corresponding Chelis target rather than falling back to a bridge call.
Library mapping
Section titled “Library mapping”| Python library | Chelis target |
|---|---|
| NumPy | chelis-std tensor primitives + Nautilus |
| SciPy | Nautilus |
| pandas | Coral |
| PyTorch (tensor ops) | chelis-std |
| PyTorch (model loading) | Hydronnx (where available) |
| sklearn | School |
How mappings work
Section titled “How mappings work”Each mapping covers the subset of the Python library's API that has a direct Chelis equivalent. Functions outside that subset fall to Tier 2 (bridge) or Tier 3 (agentic) translation.
For example, NumPy array creation and element-wise arithmetic translate to chelis-std tensor primitives at Tier 1. NumPy functions that rely on C extension internals or untyped dispatch remain in Python at Tier 2.
Adding mappings
Section titled “Adding mappings”Library mappings are Reef packages. To add support for a new library, publish a mapping package that implements the calcify-mapping interface. See the Reef documentation for the package specification.