Getting started

Quickstart

Authenticate, pull your first records, and subscribe to the live stream in minutes.

Prerequisites

You will need an EdgeOrigin API key and any HTTP client. Your account team provisions a sandbox key against live coverage rather than a sample file, because a curated sample cannot show whether we are early on the events you care about.

Authenticate

Pass your API key as a bearer token on every request:

export EDGEORIGIN_API_KEY="eo_live_..."
curl https://api.edgeorigin.com/v1/streams \
  -H "Authorization: Bearer $EDGEORIGIN_API_KEY"

Tip · Use a service key

For production, authenticate with a service key scoped to the streams you consume. Personal keys are for local development only.

Your first query

Pull recent records from a stream. Each record is timestamped and carries the signal values our models assigned it in flight, plus its provenance:

curl "https://api.edgeorigin.com/v1/records?stream=global-news&since=-15m" \
  -H "Authorization: Bearer $EDGEORIGIN_API_KEY"

# -> [{ "id": "...", "ts": "2026-07-14T12:00:03Z",
#       "urgency": 0.94, "authenticity": 0.99, "data_quality": 0.96,
#       "source": "...", "payload": { ... } }]

Subscribe to the stream

For real-time delivery, open a streaming connection and receive records as they arrive:

const es = new EventSource(
  "https://api.edgeorigin.com/v1/stream/global-news",
  { headers: { Authorization: "Bearer " + process.env.EDGEORIGIN_API_KEY } }
);
es.onmessage = (e) => console.log(JSON.parse(e.data));

Note · Streaming vs. bulk

The stream pushes new records the moment they are observed. The REST API returns point-in-time history for the same stream - so a live desk and a backtest read the same data through one interface.

Handling what arrives

Two decisions on your side of the boundary decide whether the integration stays correct under load, and neither of them raises an error when it is wrong. Key your storage on the record identifier and make the write idempotent, because a retried record is how a stream avoids losing one - an at-least-once delivery guarantee means duplicates are your responsibility to absorb, not a defect.

Then decide how long a window waits for late arrivals before you compute on it. Records are indexed by event time, not by the moment they reached you, so a region reconnecting can deliver something that belongs in a window you already closed. Publishing your own watermark - the point up to which you consider the data complete - makes that a stated policy instead of an accident.

Was this page helpful?

Suggest an edit

Introduction