⛟ Export → Parquet

CLI tool to export a Qdrant collection to a Parquet file.

• All vector types: unnamed, named, multi-vectors and sparse.
• Parallel: ranges of points are exported by worker processes.
• Resumable: an interrupted export continues where it stopped.

Usage

Requires uv.

uvx qdrant-to-parquet my_collection out.parquet            # localhost

export QDRANT_URL=https://xyz.cloud.qdrant.io QDRANT_API_KEY=...
uvx qdrant-to-parquet my_collection out.parquet            # Qdrant Cloud

Exit codes: 0 done, 1 failed, 2 invalid arguments, 130 interrupted, 143 terminated.

Output

Rows are in point id order. A missing vector is null. The file metadata holds the collection's configuration.

-- DuckDB
SELECT id, payload->>'title' AS title, len(vector) AS dim FROM 'out.parquet';

Library

from qdrant_to_parquet.exporter import ClientConfig, export_collection

if __name__ == "__main__":
    config = ClientConfig(url="http://localhost:6333")
    export_collection(
        config.make(), "my_collection", "out.parquet", workers=4, client_config=config
    )

Development

uv run pytest                                      # without a server
QDRANT_URL=http://localhost:6333 uv run pytest     # also against a server, over gRPC and REST