winkComposer · open-sourcing soon
Composable Streaming
Intelligence.
Turn streaming data into actionable insights — from edge to cloud. A high-performance JavaScript framework for IIoT. Signal conditioning, anomaly detection, and health assessment, underpinned by neural-network intelligence.
The same pipeline runs on a Raspberry Pi, a gateway, or a server.
What it surfaces
Every stream has a story.
Composer reveals it.
Two recurring patterns — drift and classification. The hero already shows a third: bearing degradation on real NASA data. More use cases span telematics, HVAC, and IT infrastructure.
A control loop hides what variance reveals.
A catalyst slowly degrades in the Tennessee Eastman benchmark. Temperature stays flat — the control loop compensates. Pressure oscillations grow from ±1 to ±50 kPa. Three detectors catch the instability before the reactor trips.
Average speed says nothing. Its behaviour says the road.
Real GPS from a truck on an Indian highway averages 40 km/h whether it’s cruising or crawling through jams. A flow reads the same speed stream three ways and fuses them into Highway, City, or Jam.
WiFi AP health, wash-cycle states, server latency, and more — the library grows with the community.
Explore all use cases →How it fits
One moving part.
Composer is the processing layer in a stack of proven, boring, open-source parts. No re-platforming — it slots into the brokers, historians, and dashboards you already run.
Standards
Aligned with NIST AI RMF 1.0 (AI risk management) and ISO 13374 (condition monitoring).
The output layer
Your operations,
at a glance.
Composer computes. QuestDB stores. Mosquitto alerts in real time. Grafana shows. Health scores, evidence, cycle signatures — the same intelligence your pipeline produces, in dashboards your team already knows.
winkComposer · QuestDB · Mosquitto · Grafana — open source, end to end
Express what, not how
Building blocks.
Endless combinations.
Each block, or "node", does one thing well: connect, filter, smooth, detect, classify, act, and more. Compose them through one declarative flow language into any pipeline you need.
Read the flow languageAI-native · via MCP
Your fleet,
in plain English.
Every insight is pre-computed, typed, and stored. Any LLM retrieves the facts via the Model Context Protocol — then reasons over them. No prompt engineering. No sampling. No hallucinated metrics.
How MCP grounding works
Proven at scale
Same engine.
Any scale.
An 8-node pipeline, benchmarked end-to-end. Your flow runs the same whether you ship it to a Pi in a gatehouse or a server tracking a full fleet.
Same engine powers every use case on this site — in your browser.
Transitioning to open source
Follow the journey.
winkComposer is in active development. Interfaces will break, details will go stale, whole sections will reshape. If something looks wrong — please tell us.