Installation

Python 3.12 or newer is required.

Core library

pip install schematize

The core package includes the pipeline, schema models, and evaluator. It has no retrieval dependency — you bring your own document store or install one of the optional adapters below.

Optional extras

Weaviate adapter

pip install "schematize[weaviate]"

Adds WeaviateRetriever for hybrid-search retrieval against a Weaviate instance.

HuggingFace adapter

pip install "schematize[huggingface]"

Adds HuggingFaceRetriever and MMLWRobertaV2Retriever for FAISS-indexed retrieval over HuggingFace datasets.

CLI scripts

pip install "schematize[scripts]"

Installs Hydra-based CLI runners (schematize-run, schematize-run-mocked, schematize-evaluate). See the CLI reference.

Notebooks

pip install "schematize[notebooks]"

Adds pandas, matplotlib, Jupyter, and ipykernel for exploratory work in notebooks.

Development

pip install "schematize[dev]"

Adds pytest, coverage, ruff, mypy, and pre-commit.

Multiple extras at once

pip install "schematize[weaviate,dev]"

Installing with uv

uv add schematize
uv add "schematize[huggingface]"

Environment variables

Set these before running the pipeline. See Configuration for details.

# LLM access (required)
API_KEY=sk-...
API_URL=https://api.openai.com/v1   # optional, for self-hosted endpoints

# Weaviate (only needed with [weaviate] extra)
WV_URL=localhost
WV_PORT=8080
WV_GRPC_PORT=50051
WV_API_KEY=...

Scripts auto-load a .env file in the working directory via python-dotenv.

CLI scripts

After installing with [scripts], three console scripts are available:

Command	Description
`schematize-run`	Interactive pipeline (prompts for user input at each step)
`schematize-run-mocked`	Replay a pre-written test case (no live LLM calls)
`schematize-evaluate`	Run the schema evaluator against expert questions

schematize-run
schematize-run-mocked +case=en_age
schematize-evaluate +case_name=age