Visalytix - Turn Any Sensor Stream into Intelligence

How it works

From camera to query in four calls.

01 / 04

Connect a sensor

Register any RTSP source as a sensor. Nothing is pulled until a stream starts, so an idle camera costs nothing.

Register a sensorHTTP

POST /sensor/
{
  "name": "loading-dock-02",
  "url": "rtsp://10.0.4.21:8554/live",
  "streamtype": "normal"
}

02 / 04

Start the stream

The sensor driver publishes frames onto NATS JetStream. Capture and analysis are decoupled, so modules scale independently of the cameras feeding them.

Start streamingHTTP

POST /streaming/start/{sensor_id}
{ "fps": 15 }

# frames publish to NATS JetStream, one subject per stream

03 / 04

Run a module

Attach a module to one or more streams. Face detection runs on Intel OpenVINO, tracks each face across frames and scores every detection.

Launch an analytics jobHTTP

POST /analytics/
{
  "stream_ids": ["6aa1c16959606bb372802b63"],
  "analysis": ["face_detection"]
}

sensor-01 · face_detection672×384 · recorded

tracked in one frame: 3faces
confidence on clear faces: 0.99+
annotated frames served live: ~17fps

Recorded from the running system. Source footage: Intel sample-videos (opens in a new tab).

04 / 04

Query the events

Every detection is written back as a timestamped event with its confidence and bounding boxes, filterable by job, stream, type or date range.

List eventsHTTP · JSON

GET /events/?analytics_id=6aa1c16a59606bb372802b64&limit=20

{
  "total": 11,
  "limit": 20,
  "skip": 0,
  "events": [
    {
      "event_type": "face_detection",
      "stream_id": "6aa1c16959606bb372802b63",
      "frame_number": 274,
      "timestamp": "2026-09-09T20:29:05.650Z",
      "confidence": 0.999,
      "detection_count": 3,
      "metadata": {
        "model": "face-detection-adas-0001",
        "detections": [
          { "bbox": [120, 80, 250, 220], "confidence": 0.999 }
        ]
      }
    }
  ]
}

Four calls to go from a camera to a queryable event.

Try it on your footage

Modules

Modular by design.

Each module runs as its own container on the stream bus, so adding one never touches ingestion, transport or the event store. One ships today; the sheet says which.

Intelligence modules and their availability
Module	Runs as	Status
Face detection & trackingLocates and tracks faces across frames, scoring every detection.	OpenVINO container on the stream bus	Shipping
Your own modelPackage a model as a container that reads frames from the bus.	Any container that consumes NATS subjects	Roadmap
Module chainingFeed one module's output topic into the next module's input.	Output topic to input subject	Roadmap
Object detectionGeneral object classes beyond faces.	OpenVINO container on the stream bus	Roadmap

Deployment

Your cloud, your racks, or both.

The same containers run everywhere, on one host with Docker Compose or across a cluster with Kubernetes. Moving between clouds, or off them, is a deployment change rather than a rewrite.

Illustration: a server rack, a network switch and a security camera inside one room.

On-premises

On your own hardware

Run the whole stack with Docker Compose on a single Linux host, or across your own Kubernetes cluster. Footage never has to leave your network.

$docker compose up -d

Illustration: clusters of server racks in a cloud region, joined by data lines.

Public cloud

AWS or Google Cloud

Amazon EKS with Terraform for the cluster, node groups and container registry, or a single Compute Engine VM that starts for a demo and stops when you are done.

$terraform apply

Illustration: cameras in a building streaming across to server racks in a cloud region.

Hybrid

Cameras on site, compute in the cloud

Keep the cameras where they are and stream RTSP to processing in your own cloud account, over the network you already have.

$rtsp://site-a/cam-01 → your cloud

Same stack everywhere

Docker Compose
Kubernetes
Terraform
Compute Engine
NATS JetStream
MongoDB

Running somewhere else? It is plain containers, so bring your environment to the demo call and we will walk through it.

Turn any sensor stream into intelligence.

Plug and play intelligence modules, and chain them into real-time pipelines.

Every stream, every module, one event log.

From camera to query in four calls.

Connect a sensor

Start the stream

Run a module

Query the events

Chain engines on real footage.

Modular by design.

Your cloud, your racks, or both.

On your own hardware

AWS or Google Cloud

Cameras on site, compute in the cloud

Point it at
your cameras.

Turn any sensor stream into intelligence.

Plug and play intelligence modules, and chain them into real-time pipelines.

Every stream, every module, one event log.

From camera to query in four calls.

Connect a sensor

Start the stream

Run a module

Query the events

Chain engines on real footage.

Modular by design.

Your cloud, your racks, or both.

On your own hardware

AWS or Google Cloud

Cameras on site, compute in the cloud

Point it at your cameras.

Point it at
your cameras.