Flink’s model,
Rust’s runtime.
A stream processing engine written in Rust for the heavy lifting: ingest, clean, enrich and aggregate events in motion — feeding feature-ready data straight into the MLOps platform or anything downstream.
Built for events that don’t batch
The familiar stream model — sources, windows, sinks — on a runtime without a JVM.
Throughput
Millions of events per minute per cluster, with backpressure handled for you — no tuning forum threads.
Low, predictable latency
Sub-millisecond per record, no GC spikes. Rust the whole way down, so p99 stays honest.
Memory safety
State handling is checked at compile time. No segfaults at 3 a.m., no heap explosions mid-window.
A tiny DSL
Declare sources, windows and sinks in a handful of Python lines. No 400-line job classes.
Sources → windows → sinks
The whole job is one declarative file.
Sources
kafka("ticks")
http("news-rss", 30s)
files("uploads/")
Window & aggregate
enrich(news, on=ticker)
tumbling(5m)
agg(mean, sum)
Sinks
mlops_dataset(…)
parquet("lake/")
webhook(…)
# declare the pipeline in Python from absolutno.stream import pipeline, source, sink @pipeline(name="market-etl") def market_etl(): ticks = source.kafka("ticks", key="ticker") news = source.http("news-rss", every="30s") joined = ticks.enrich(news, on="ticker", within="5m") agg = joined.window(tumbling="5m").agg( avg_sentiment=mean("sentiment"), volume=sum("volume"), ) sink.mlops_dataset("features_live", agg)
What people run on it
Anything that moves faster than a cron job.
Real-time ETL for ML
Clean, join and window events so the MLOps platform always trains on fresh features.
Event enrichment
Join live streams against reference streams and lookups within a sliding window.
Continuous aggregation
Rolling metrics and materialised views that update as events arrive — never batched.
Direct integration
Pipelines output straight to the MLOps platform — sink.mlops_dataset(...) lands a versioned, ready-to-train dataset. No glue code, no export step.
Pricing
Free to start. Solid in production.
Start on the sandbox. Upgrade when the stream carries real data.
Developer
One pipeline, on us.
- ✓1 pipeline
- ✓2 sources / 2 sinks
- ✓50k events per minute
- ✓Local + cloud sandbox
- ✓Community support
Production
Pipelines that never drop a record.
- ✓10 pipelines
- ✓Exactly-once sinks
- ✓1M events per minute
- ✓State snapshots & replay (30 days)
- ✓Schema drift alerts
- ✓Email support
Enterprise
Managed clusters in your regions.
- ✓Dedicated stream clusters
- ✓Unlimited pipelines & sources
- ✓Cross-region failover
- ✓On-prem / self-hosted option
- ✓99.95% SLA, named engineer
Compare plans
| Compare | Developer | Production | Enterprise |
|---|---|---|---|
| Pipelines | 1 | 10 | Unlimited |
| Sources & sinks | 2 each | 10 each | Unlimited |
| Throughput | 50k events/min | 1M events/min | Custom clusters |
| Exactly-once sinks | — | ✓ | ✓ |
| State snapshots & replay | 7 days | 30 days | Custom retention |
| Direct MLOps sink | ✓ | ✓ | ✓ |
| Cross-region failover | — | — | ✓ |
| Support | Community | 24/7 + named engineer | |
| SLA | — | — | 99.95% |