Stream processing.
Exactly-once streaming compute for real-time pipelines, enrichment and detection.
Process data in motion with exactly-once semantics — windowed aggregations, joins and enrichment over streams, feeding dashboards, alerts and the model in real time.
- Category
- Analytics
- Deployment
- Managed → air-gapped
- Governance
- IAM · encryption · audit
Real-time used to mean a second pipeline and a second set of bugs.
Batch handles most analytics, but some questions can't wait for the nightly job — fraud as it happens, an alert before the outage, a feature computed the moment an event lands. The classic cost of real-time is a parallel system with its own framework, its own state model and its own failure modes, computing numbers that don't quite reconcile with the batch ones. Aether's stream processing shares the lake, the catalog and the governance with everything else, so streaming and batch agree on the data instead of arguing about it.
Exactly-once and stateful, so the numbers are right under failure.
The hard part of streaming is correctness when something fails mid-flight. Aether processes streams with exactly-once semantics and checkpointed state, so a restart doesn't double-count and a window doesn't lose events. Tumbling, sliding and session windows over event time, plus joins across streams and reference data, let you express real aggregations and enrichment — not just a firehose you have to make sense of downstream.
Sub-second from event to action — including the model.
Because processing runs on the same platform as the model, a stream can score, classify or embed in flight — a transaction is checked for fraud, a log line is triaged, a sensor reading is anomaly-flagged, the moment it arrives. The output feeds dashboards, alerts and the model live, with the same lineage and governance as the rest of the data. Real-time stops being a separate stack and becomes another way to read the same governed tables.
Score a payment stream for fraud in flight, sub-second, with the model as a step in the stream — not a callout to a separate service.
stream("payments")
.window({ type: "sliding", size: "5m", every: "30s" })
.map((p) => ({
...p,
risk: aether.score("fraud", p), // model, in the stream
}))
.filter((p) => p.risk > 0.9)
.to("alerts.high_risk_payments"); // sub-second to actionThe fraud score is computed inside the stream, not by shipping each event to an external endpoint and waiting. State and windows are exactly-once, so a restart doesn't double-alert. And the stream reads and writes the same governed tables as the warehouse and the model.
Three steps to running.
Read from queues, the event bus or change feeds with exactly-once semantics.
Windowed aggregations, joins and enrichment over event time.
Feed dashboards, alerts and the model sub-second from event.
Stream processing, in full.
Stateful processing with checkpointing and no double-counting.
Tumbling, sliding and session windows over event time.
Enrich and correlate multiple streams and reference data.
Sub-second from event to action for detection and alerting.
Provision it in a few lines.
Every service is reachable from the same SDK, CLI and infrastructure-as-code — one identity, one bill, one audit trail across the whole catalog.
# Provision stream processing on Aether Cloud
aether analytics create \
--service stream-processing \
--name app \
--region us-1 \
--deploy managed # or vpc | on-prem | air-gappedAt a glance.
- Semantics
- Exactly-once, stateful
- Windows
- Tumbling · sliding · session
- Joins
- Stream + reference data
- Latency
- Sub-second
- Checkpointing
- Built-in
Built for real work.
Real-time dashboards
Anomaly and fraud detection
Live feature computation
On one model, not stitched together.
The usual stack runs stream processing in one product, the model in another and the data in a third — and the seams between them are the cost. Aether Cloud runs it on the same platform that serves the model, governs your identity and deploys into your boundary, with the rest of the catalog one hop away.
No stitching a vector DB to one place, a warehouse to another and a model to a third — stream processing sits next to the rest of the catalog, one identity, one bill.
The provider that runs Aether runs your stream processing — so the data and the model never leave the same governed boundary to talk to each other.
Need a capability that isn’t here yet? The model writes and deploys it into the same boundary — the catalog is a starting point, not a ceiling.
Good to know.
Exactly-once streaming compute — feeding the model and digital twins live, on one governed platform.
No — stateful processing with checkpointing and exactly-once guarantees.
Real-time dashboards, anomaly alerts and live feature computation for the model.
Your boundary, your choice.
Pairs well with.
Managed Spark and distributed processing for large-scale transforms and ML feature pipelines.
Visual and code-first pipelines with scheduling, lineage and schema management.
Governed dashboards and self-serve exploration over the warehouse and lake.
Full-text and hybrid search as a service, with relevance tuning and analyzers.
Run Stream processing on Aether Cloud.
Exactly-once streaming compute for real-time pipelines, enrichment and detection. Deployable managed, in your VPC, on-prem or fully air-gapped — talk to us about the configuration your workloads and your boundary require.