Fabric / How it works
From intent
to a system
you control.
Connect the data you have. Describe what you want to build. Fabric turns the workflow into a running system, with every step visible and editable.
See how it worksWorking with Fabric
Your path through Fabric.
Start with the outcome.
Tell Fabric Agent what you want to build: join data feeds, monitor a geofence, publish a dataset, or create an operational view. It drafts the workflow from known nodes and recipes.
Review the draft, supply credentials and source details, and adjust the configuration. The result is a graph you can work with directly.
Drafts are checked for unknown nodes, invalid connections, type mismatches, and missing inputs before use.
Under the surface
A visible graph.
A managed runtime.
Fabric compiles your workflow into a streaming runtime. You work with sources, rules, and outputs; the platform handles execution between them.
The graph defines execution.
Your configured nodes become a running workflow. Typed connections and validation make inputs, transformations, and outputs explicit. AI helps you author it; Fabric executes the final graph.
Data can arrive unevenly.
Processing stages run concurrently. Streaming transport buffers records, applies backpressure when a stage needs time, and can replay delayed or bursty feeds.
Outputs have a contract.
Each API output publishes the fields defined upstream. External consumers use a scoped endpoint and credentials, while the runtime’s internal data transport stays private.
A closer look at the architecture
- Go + Kafka
- Concurrent node execution with streaming transport, buffering, backpressure, and replay.
- PostgreSQL + spatial storage
- Workflow state, schemas, output metadata, and geospatial data, including vector tiles.
- REST + WebSockets
- APIs for external delivery; live updates for views and debugging.
Measured performance
3.25 million records.
One replay.
In a 10-minute AIS replay, Fabric ingested 3,245,193 source records with four workflows running on an AWS m7i.large instance.
- Average source ingestion
- 5,378records / second
- Peak source ingestion
- 10,875records / second
- Running together
- 4workflows
Test setup and output measurements
Paneo internal measurement, 29 June 2026. Performance depends on workflow complexity, source data, and deployment.
The replay used public AIS data and a test geofence at the Strait of Hormuz. It exercised a multi-join workflow with geofence event detection and dashboard delivery.
- Event processing
- 2,627,919 decoded AIS position records reached the Area Event Detector node, producing 5 new geofence entry/exit events.
- Output latency
- At the final profiler sample, 6 retained event rows had source-receipt-to-dashboard-output latency of 183 ms median, 246 ms average, and 395 ms P95 and maximum.
- How to read this
- The latency figures describe those retained output rows in this replay. The event count reflects the selected geofence and rules; it is not a measurement of detection accuracy or false-alarm reduction.
Deployment
Close to your data.
Under your control.
Fabric runs as a private instance. Existing deployments include AWS and on-premises Docker installations, so the runtime can sit close to your sources and systems.
You focus on the workflow. Paneo manages the infrastructure and ongoing operation. Together, we define the access, storage, and service needs of your deployment.
How we work with you Follow a complete product exampleYour next
milestone.
Let’s get there.
You have a deployment ahead.
A customer waiting.
A data path to make work.