Skip to main content

Overview

A monitor turns a set of searches into a feed. It stores the searches, re-runs them on a cadence, remembers every document it has already handed you, and emits only what is new. Four things make up the model:
  • A monitor is the saved configuration: a name, a cadence, a list of search requests, a status, and an optional webhook. An active monitor runs on its cadence; a paused one keeps its configuration and its records but does not run.
  • A run is one execution of the monitor. It has a status, counts of how many requests succeeded, and a per-request breakdown of what each search returned.
  • A record is one new document the monitor found. Records belong to the run that produced them, and to the monitor as a whole.
  • A webhook is an HTTP callback the monitor fires when a run finishes, so you do not have to poll. Deliveries carry the run, not the records, and each one is signed.

Cadence

The cadence is a count and a single unit — 6h, 2d, 90m — with a floor of one hour. It promises one run per cadence period, completing within it. Where in the period the run lands is not guaranteed. See Reference for the full grammar and the bounds.

Records and deduplication

A monitor delivers each document once, and does not deliver it again while it keeps the run that first delivered it. On the second run you get the new results, not the whole result set again. Visibility is a separate clock: runs and their records remain readable for a limited window and then disappear. Ageing out does not make a document deliverable again, so persist anything you need to keep as you read it. See Reference for the window. Records are ordered oldest first and are paged with the record id itself as the cursor. See Read records.

Next steps

To learn more about Seltz Monitor, check out the following topics: