Installation¶

Prerequisites¶

• Python >= 3.14
• PyTorch >= 2.0
• didactic >= 0.6.0
• panproto >= 0.45.0 (provides the schema/lens machinery)
• panproto-grammars-all >= 0.45.0 (ships the QVR tree-sitter parser)

The didactic, panproto, and panproto-grammars-all packages are pulled in automatically by pip install quivers.

From PyPI¶

pip install quivers

From Source¶

To install directly from source:

git clone https://github.com/FACTSlab/quivers
cd quivers
pip install -e .

Development Installation¶

If you intend to run tests and contribute:

pip install -e ".[dev]"

This adds:

• pytest >= 7.0: test runner
• pytest-cov: coverage reporting
• ruff: linter and formatter
• pyright: static type checker

The [docs] extra (mkdocs, mkdocstrings, mkdocs-cinder, pymdown-extensions, pygments) is needed to build the documentation site locally.

Verify Installation¶

Check that the import works:

import quivers
print(quivers.__version__)

Run the test suite:

pytest tests/

Dependencies¶

Quivers depends on:

• torch (>= 2.0): differentiable tensors and automatic differentiation
• didactic (>= 0.6.0): typed-data layer that backs every value-type in quivers (dx.Model, dx.TaggedUnion, dx.Lens)
• panproto (>= 0.45.0): schema/theory machinery used to extract a Schema from each .qvr program for diff/migrate workflows
• panproto-grammars-all (>= 0.45.0): ships the QVR tree-sitter parser registered with panproto; quivers does not run a hand-written lexer or recursive-descent parser

All functionality is built as pure Python atop PyTorch; no other system dependencies are required at runtime.