Data pipeline

The Vangrid data pipeline is the governed path that every observation travels from physical capture to your application. At each stage, the pipeline enforces sovereignty controls, preserves cryptographic integrity, and filters for quality, so what arrives in your API response is not just spatially accurate, but verifiably so. Understanding the pipeline helps you predict latency, interpret response fields, and design your system to take full advantage of Vangrid’s guarantees.

Pipeline stages

Capture

Edge nodes capture spatial observations from their environment. Nodes operate independently and asynchronously, there is no central clock or coordinated capture event. Each node records the time and the location reported by the device at the moment of capture, and both travel with the observation for the rest of its life.

Edge compute

The node runs detection on every frame as it is recorded and obscures faces and vehicle number plates before the video is encoded. Only the filtered capture is transmitted, and spatial features, object positions, geometries and classifications are extracted from it.Edge compute is the mechanism behind Vangrid’s privacy guarantee: the unblurred frames never leave the hardware, so they cannot be intercepted, subpoenaed, or leaked in transit. No version of a capture in which a face or a number plate is legible exists anywhere off the device that recorded it.Running feature extraction on the node itself, so that only the derived payload is transmitted, will follow as device compute allows.

Cryptographic signing

After extracting features, the node signs the payload with its private key. The resulting provenance_hash encodes:

The node’s unique identifier
The capture timestamp
A hash of the feature payload content

The signature is generated on-device using hardware-backed key storage. Any modification to the payload after signing, in transit, at rest, or in your own systems, produces a hash mismatch that you can detect with standard cryptographic verification.

Aggregation

Signed payloads from multiple nodes covering the same area of interest are ingested by Vangrid’s aggregation layer. The aggregator:

Groups observations by spatial overlap and temporal proximity
Cross-corroborates observations from independent nodes to compute ground_truth_score
Detects and flags conflicting observations rather than silently resolving them
Merges compatible multi-view observations into unified feature records where appropriate

Aggregation adds latency proportional to the number of nodes contributing to a response. Real-time queries use a shorter aggregation window; historical queries use a larger one.

API delivery

The aggregated, scored response is returned to your application as structured JSON over HTTPS. The response body contains a features array, each element carrying geometry, timestamps, quality scores, provenance, and any domain-specific fields relevant to your query type.

What “sovereign” means

In the context of the Vangrid pipeline, sovereign means that data does not transit infrastructure outside your designated boundary without explicit, auditable consent. Sovereignty operates at three levels:

Node-level, Identifiable frames stay on the node. Faces and number plates are obscured before anything is transmitted, so they never traverse a network at all.
Infrastructure-level, Signed feature payloads are routed only through Vangrid infrastructure within your configured sovereignty boundary. No third-party cloud provider or transit network touches your data without your authorization.
API-level, Your API responses are delivered over TLS. Vangrid does not log or store query content beyond what is required for billing and provenance records.

For defense and government deployments, Vangrid supports nation-state grade sovereignty controls including air-gapped delivery modes, region-locked infrastructure, and custom key management integration. Contact hello@vangrid.io for details on these configurations.

Standard enterprise accounts operate within a single-region sovereignty boundary by default. Cross-region access requires explicit configuration and is audited in your account’s provenance log.

Latency characteristics

Real-time queries

Real-time queries return observations captured within the last few seconds to minutes, depending on node density and AOI size. The aggregation window is short, Vangrid does not wait for all possible contributing nodes before returning a response. You receive a partial but fast result, with node_count and ground_truth_score reflecting the nodes that responded within the window.Typical p50 latency for a real-time query over a small urban AOI is under 500ms from request to first byte. Larger AOIs or lower-density regions will be slower.

Historical queries

Historical queries retrieve observations from Vangrid’s provenance archive. Because the data is already aggregated and stored, historical queries are not subject to node response latency, but they do incur retrieval and decryption overhead.Historical responses include the full set of contributing nodes and their individual signed payloads, giving you a richer audit trail than real-time queries. Typical latency for historical queries is 1–5 seconds depending on archive depth and result size.

Streaming subscriptions

For applications that need continuous updates, fleet tracking, infrastructure monitoring, real-time situational awareness, Vangrid supports streaming subscriptions via WebSocket. Observations matching your AOI and filter criteria are pushed to your connection as they are aggregated, with latency comparable to real-time queries.

Response data format

Every Vangrid API response is a GeoJSON-compatible JSON object. The top-level structure includes a features array where each element represents one spatial observation.

{
  "query_id": "qry_01j9z3kfm8x2v4n7",
  "timestamp": "2026-05-22T14:32:07.412Z",
  "aoi": {
    "type": "Polygon",
    "coordinates": [[[-122.4194, 37.7749], [-122.4094, 37.7749], [-122.4094, 37.7849], [-122.4194, 37.7849], [-122.4194, 37.7749]]]
  },
  "node_count": 14,
  "features": [
    {
      "feature_id": "feat_7d3a9c2e",
      "type": "Feature",
      "geometry": {
        "type": "Point",
        "coordinates": [-122.4142, 37.7799, 12.4]
      },
      "properties": {
        "classification": "vehicle.truck",
        "velocity_mps": 4.2,
        "heading_deg": 273.1,
        "captured_at": "2026-05-22T14:32:05.198Z",
        "ground_truth_score": 0.94,
        "provenance_hash": "sha256:a3f8c1d9e4b27056f3a8c9d1e4b270a3f8c1d9e4b27056f3a8c9d1e4b270",
        "sensor_modalities": ["camera", "lidar"],
        "node_count": 3
      }
    }
  ]
}

Key response fields:

Field	Type	Description
`query_id`	string	Unique identifier for this query, used in audit logs and support.
`node_count`	integer	Total number of nodes that contributed to this response.
`feature_id`	string	Unique identifier for this observation.
`geometry`	GeoJSON geometry	Position and shape of the observed feature. Coordinates are `[longitude, latitude, altitude_meters]`.
`classification`	string	Hierarchical object classification (e.g., `vehicle.truck`, `person.pedestrian`).
`captured_at`	ISO 8601 string	Timestamp of the original sensor capture on the node.
`ground_truth_score`	float	Confidence score from 0.0 to 1.0 based on multi-node corroboration.
`provenance_hash`	string	Cryptographic signature of this observation.
`sensor_modalities`	array	Sensor types that contributed to this feature.

How Vangrid differs from cloud-only alternatives

Cloud-only spatial platforms funnel unfiltered footage into a central processing cluster before returning results. This design introduces several limitations that Vangrid’s pipeline avoids:

	Cloud-only platforms	Vangrid
Privacy	Identifiable footage transits and is stored in the cloud	Unblurred frames never leave the device
Provenance	No per-observation cryptographic proof	Every feature carries a `provenance_hash`
Sovereignty	Data crosses arbitrary cloud regions	Data stays within your configured boundary
Latency	Round-trip to central cluster	Aggregation happens close to the edge
Infrastructure cost	You pay for cloud compute and storage	Vangrid absorbs node and compute costs

When comparing Vangrid to a cloud-only alternative, ask the vendor: “Can you prove that a specific observation was not altered between capture and delivery?” With Vangrid, the provenance_hash answers that question without requiring you to trust any intermediary.

Start

Contribute

Protocol

Concepts

Data

Use cases

Legal

Pipeline stages

What “sovereign” means

Latency characteristics

Response data format

How Vangrid differs from cloud-only alternatives

​Pipeline stages

​What “sovereign” means

​Latency characteristics

​Response data format

​How Vangrid differs from cloud-only alternatives

Pipeline stages

What “sovereign” means

Latency characteristics

Response data format

How Vangrid differs from cloud-only alternatives