Quiver is a feature serving system for machine learning inference. It resolves feature requests across different backends in a single gRPC call and returns the results as columnar data via Apache Arrow Flight for direct consumption by model inference pipelines.

from quiver import Client

client = Client("localhost:8815")

table = client.get_features(
    feature_view="user_features",
    entities=["user:1000", "user:1001", "user:1002"],
    features=["score", "country", "last_active"],
)

df = table.to_pandas()

# Python client
pip install quiver-python

# Server (Docker)
docker run -p 8815:8815 -p 8816:8816 \
  -v $(pwd)/config.yaml:/etc/quiver/config.yaml \
  ghcr.io/olamyy/quiver-server:latest

See the docs for more details on how to get started, architecture, configuration, and the Python client API.