Run an AI analysis on a camera image

The AI API analyses the same imagery the Dash API serves. Because it accepts a camera and a timestamp directly, you can go from "a camera" to "what is in frame" without moving image bytes around yourself.

Authenticate

The AI API accepts the same JWT the Dash login endpoint returns — but note the base URL and the alternative header:

Base URL   https://data.evercam.io/v2
Header     Authorization: Bearer <jwt>
    or     x-api-key: <jwt>

Detect objects

You can identify the image three ways — camera_exid with a timestamp, a snapshot_url, or a base64_img. Naming the camera is the simplest, because the platform already has the image:

curl -X POST "https://data.evercam.io/v2/ai-models/object-detection" \
  -H "Authorization: Bearer $EVERCAM_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "camera_exid": "dorset-street-i",
    "timestamp": "2026-08-01T09:00:00.000+00:00"
  }'

Timestamps use the same ISO 8601 form as the recordings endpoints.

Preview variants

Several models offer a /preview sibling that returns an annotated image rather than raw coordinates — useful when you want to show a result instead of drawing it yourself.

Draw results over the snapshot

To render boxes yourself, fetch the snapshot and the detections for the same moment, then overlay them. Because detection coordinates describe the source image, scale them to however you display it:

const scaleX = displayedWidth / naturalWidth
const scaleY = displayedHeight / naturalHeight

for (const box of detections) {
  context.strokeRect(
    box.x * scaleX,
    box.y * scaleY,
    box.width * scaleX,
    box.height * scaleY,
  )
}

Other models

The same request shape drives the rest of the catalogue:

Beyond per-image models, the AI API also aggregates over time: ANPR, gate reports and site analytics. Browse them all in the AI API reference.