Real-Time Serving
Live state for apps and agents, a millisecond away.
Eon Yukon enables real-time serving: it keeps aggregates current as data lands and serves the latest value for any key, such as windows, distinct counts and top-k, in under 1 ms through a Redis-compatible API. Application and AI teams get live profiles and features without a cache to fill or sync jobs to run.
Useful summaries, current and a key away
Eon Yukon updates configured aggregates alongside detailed records in the same ingestion pass, then serves them by key, so a read never waits on a scan or a cache refill.
Under 1 ms by key
Hot keys are served from memory, warm keys from local disk and everything else from the lake. A read is a key lookup, with no query to plan and no table to scan.
Any Redis client
Services and agents read with GET, MGET and HGETALL over the Redis protocol, using the client libraries they already have.
No cache to keep in sync
State commits to object storage in a dedicated account in your region, shared with you, or in your own account with BYOC as the source of truth and moves into memory on its own. There are no jobs copying the lake into Redis or Aerospike.
Scales with traffic
Elastic read replicas and sharding by key grow with read traffic, so a launch or a sale doesn't need a capacity plan for the cache.
From event to live row
Eon Yukon serves the state it commits to object storage, in tiers that follow how often each key is read.
Eon Yukon folds each event into the aggregate for its key, in the same pass that writes the Apache Iceberg table.
The aggregate is committed to object storage. Hot keys move into memory and warm keys onto local disk without a sync job.
A checkout service or support agent calls HGETALL for a customer and gets the profile back from memory, current to within seconds of the last event: 0.4 ms in the example above.
Put live state to work
01
Personalize every session
Read a customer's recent orders, 30-day spend and top category by key at the moment of the request, updated within seconds of each event.
02
Give agents live customer context
A support or shopping agent reads a customer's live profile in under 1 ms, instead of querying the warehouse mid-conversation.
03
Serve features to models
Each feature is computed once as data lands: one definition writes the lake tables models train on and the value applications read by key in under 1 ms.
04
Retire lake-to-cache sync jobs
Jobs that copy the lake into Redis or Aerospike go away. Applications read the lake by key directly.
See how Eon works in your environment
Bring a lookup your app makes today and we will show you that value served by key from Eon Yukon, with read latency and freshness side by side.
