How does Dataminr work, and what a developer can build instead

How does Dataminr work? From the outside, it is a pipeline with three stages: collect public data at scale, decide which signals describe a real event, then describe and route that event to the people who need it. This explainer walks through those stages using what Dataminr says on its own site, then shows how a developer can get a smaller version of the same loop running today with AyeWatch, for topics that are not security work.

Stage 1: collect public data

Dataminr says it draws on 1M public data sources. Its pages describe multi-source data and "multi-modal" AI, which points to several kinds of input rather than a single text feed. For its cyber products it also talks about fusing internal and external data, and for travel risk it pulls data from travel management companies.

The point of this stage is coverage. An event that shows up first in a local post, a sensor feed or a video clip should still be caught.

Stage 2: detect that something happened

This is the core of the product. Dataminr says it runs more than 50 proprietary LLMs and calls the first layer of its AI "Multi-Modal Fusion AI for discovery". In plain terms, it looks across many weak signals and decides that they add up to one event: a fire, an outage, a strike, a breach.

Its First Alert product for the public sector adds a corroboration step, where AI agents scan public sources for supporting information before and after an alert goes out.

Stage 3: describe and route the event

Once an event is detected, Dataminr uses what it calls generative AI for description and agentic AI for context. The output is a live event brief that updates as the situation changes, with historical context from Dataminr's archive and a prediction of how the event may develop over the next hours.

Delivery happens through web, mobile and email apps, and through APIs and connectors that feed customers' internal systems. Access to all of it runs through a demo request; Dataminr publishes no prices.

The developer's version of the same loop

Most developers do not need event detection across a million sources. They need to know when a specific kind of thing happens: a dependency ships a breaking change, a cloud region degrades, a competitor launches, a standard is revised. That is the job AyeWatch does, and you can set it up without talking to anyone.

1. Describe the event

Write the monitor as a sentence. "A new major version of Node.js, or a security release for the LTS line, announced by the Node.js project." AyeWatch reads new items and alerts only on matches, so the description does the filtering that a keyword list and a pile of regexes would otherwise do.

2. Watch the pages that matter

Changelogs, status pages and release notes are often the first place a change appears. Add them as webpage monitors and AyeWatch summarizes what changed on the page.

3. Choose a speed

Auto mode on paid plans speeds up while a topic is active and slows down when nothing is happening. If you prefer control, set a custom interval per monitor, so a release feed and a quiet policy page each run at the pace they need. Paid plans start at $9 a month.

4. Route to your stack

Every paid plan includes the REST API, so you can create, list, update and delete topics from code, and webhooks signed with HMAC-SHA256 that carry the AI summary. There is an MCP server too, so an agent in your editor can manage monitors for you. Slack and Discord cover the team side.

5. Try it before you pay

The Free Preview gives three lifetime runs across up to three topics, checked daily or slower, with no card. That is enough to see whether one monitor earns its place.

Where the analogy stops

AyeWatch is not a smaller Dataminr for security. It has no threat scoring, no security integrations, and it should not be used for life-safety alerting. It also does not claim Dataminr's speed. If you build for a security team, a crisis desk or emergency services, Dataminr or another risk intelligence platform is the right base. For everything else, the AyeWatch write-up on Dataminr alternatives covers the other options.