Crisis detection with a news API: know before the wires file

Corporate security, duty-of-care and crisis-management teams share one operational problem: the event you care about is known locally minutes to hours before it reaches the sources your tools monitor. Closing that gap is a data problem, and it has a data solution.

The latency chain of a crisis

A blast, a building fire, an airport closure, civil unrest: the typical path is local Telegram channels and regional outlets first, national media second, international wires last. In regions like Eastern Europe, the Middle East and Central Asia the first hop happens on Telegram almost every time — 15 to 40 minutes before wire coverage is normal, and for smaller events the wires never file at all.

Detection as a filter, not a feed to read

Nobody can read 27,000 sources. The point of structured news data is that you don't: every item is scored 0–10 for urgency by a deterministic pattern engine within about 60 seconds of collection, tagged with event_type (disaster, war, crime, politics…), country and topic. Crisis detection becomes a standing query:

GET /v1/feed?event_type=disaster&min_score=7&country=XX

…delivered as a webhook push, so your ops channel or GSOC dashboard receives events instead of polling for them:

POST /v1/webhooks  { "url": "https://your-soc.com/hook", "min_score": 7 }

Deduplication matters at 3 a.m.

The same incident generates dozens of headlines within minutes. Every item in the feed carries a cluster_id: one event, one cluster, however many sources cover it. Your alerting fires once per event with a growing source count — not forty times for one explosion, which is the difference between an alert channel people trust and one they mute.

Coverage where it counts

197 countries and 49 languages, with unusual depth exactly where generic tools are thinnest: 4,000+ public Telegram channels across Russia/CIS, Ukraine, the Middle East and Southeast Asia — official emergency channels, city channels and incident reporters, all source-labeled. For duty-of-care teams with people in these regions, that layer is the early warning.

Evaluate against your last incident

The honest test: take a real incident from your log and check what our archive had, and when. The playground runs live queries with no signup, and a free key (100 requests/day, no card) unlocks the full filter stack for a proper evaluation.