Sub-millisecond feature serving for ML models. Batch pipelines, streaming updates, and production monitoring - all built on Valkey Hashes.
6 guides from basic feature writes to production monitoring and versioning
Define entities, write features, and read them back in 5 minutes
Single lookups, batch pipelines, selective fields, and multi-view vectors
Sliding windows, rolling averages, and HyperLogLog cardinality
Real-time pipelines with Valkey Streams and consumer groups
Serve features to scikit-learn, FastAPI, and LLM chains
Freshness monitoring, versioning, health checks, and observability
Create entities, write features, fetch feature vectors, and measure latency in real-time
HSET fs:v1:user:user_123
age "28"
ltv "1250.5"
segment "premium"
EXPIRE key 3600
One Hash per entity. Atomic write + TTL. ~0.2ms.
HGETALL fs:v1:user:user_123
-- or select fields --
HMGET key age ltv segment
Full vector or selective fields. ~0.1ms.
PIPELINE:
HGETALL user_001
HGETALL user_002
... x100
EXECUTE → 1 round-trip
100 entities in ~0.3ms. One TCP round-trip.
XADD fs:stream:user *
entity_id user_123
features {json}
XREAD ... → HSET
Publish updates via Streams, materialize to Hashes.
Full Python library with Entity, FeatureView, OnlineStore, StreamingWriter, and FeatureMonitor. Docker compose included.