strata-client

A lightweight Python client for a Strata server. Depends only on httpx + pyarrow, none of the server's stack (no pyiceberg / fastapi / duckdb / pydantic, no Rust extension), so it drops into any analysis venv, training image, CI job, or notebook without dragging the deployable service along.

pip install strata-client
pip install "strata-client[duckdb]"   # or pandas, polars, datafusion, all

from strata_client import StrataClient

with StrataClient() as client:  # resolves the server URL from STRATA_SERVER_URL,
    # STRATA_HOST / STRATA_PORT, or [tool.strata]
    art = client.materialize(
        inputs=["file:///warehouse#db.events"],
        transform={"executor": "scan@v1", "params": {}},
    )
    table = client.fetch(art.uri)

    # Persist a result computed locally, with its lineage
    clean = client.put(
        inputs=[art.uri],
        transform={"executor": "clean@v1", "params": {}},
        data=table,
        name="team/dataset/clean",
    )

    # Registry: names, aliases, tags, audit
    client.set_alias("team/dataset/clean", "champion", clean.artifact_id, clean.version)

An unnamed result such as art is a cache entry: the server keeps it while it is used and its retention sweep may collect it later, after which the same call computes it again. A named one (clean) is kept.

The client and the server distribution (strata-notebook) are independent: they share only the JSON wire protocol, and neither depends on the other. Code written against the old strata.client module imports strata_client instead.