Skip to Content
DetectorsHow Detectors Work

How Detectors Work

Detectors don’t run on their own — they run as part of a source scan, on the content of each asset that scan produces. This page explains that flow, how a detector is matched to the right kind of content, and how you choose which detectors run on which source.


Detectors live on the source

Every source carries its own list of detectors. A detector only runs on a source if it’s switched on there — so two sources can scan for completely different things. You configure this in the source’s setup (see Configuration & Fields):

On a source you can…Result
Enable a pre-built packThat pack runs on every applicable asset in the scan
Disable a packIt’s skipped for that source
Select custom detectorsYour own detectors run alongside the built-in ones

Because it’s per-source, you can be strict on a customer-facing system and lenient on an internal sandbox, without affecting each other.

Don’t want to manage this by hand? The Config and Detector agents can enable the right packs on a silent source and even author new detectors — see Autopilot.


Content-type routing

Not every detector makes sense on every asset. A secrets scanner wants text; an image classifier wants pixels. Each asset has a content type, and each detector declares which content types it can handle. Classifyre only runs a detector on assets it actually applies to.

Content typeExample assetsDetectors that apply
TextDocuments, pages, messages, codeSecrets, PII, code security, text-based custom detectors
TableDatabase rows, spreadsheetsPII, secrets, custom rules over column values
URLLinksBroken-link checks
ImagePhotos, screenshots, diagramsImage classification, object detection
Audio / VideoRecordings, clipsText detectors, after transcription
Binary / OtherFiles needing extractionText detectors, after OCR

This is why OCR and transcription matter: they turn images and media into text, which unlocks the text-based detectors on content that would otherwise be skipped.


What happens during a scan

  1. For each asset, Classifyre selects the detectors enabled on the source that fit the asset’s content type.
  2. Each detector analyses the content independently.
  3. Matches become findings — one per match, each with a severity and a confidence score. No match means no finding, which is the healthy default.
  4. Results reconcile against earlier scans, so the same issue is tracked over time rather than duplicated (see Findings & Results).

Pre-built and custom, together

Pre-built and custom detectors run through the same machinery and produce the same kind of findings — they differ only in where they come from:

So a custom EU-IBAN rule and the built-in PII pack land their results in the same place, side by side, with the same severity scale and lifecycle.

Next: see exactly what a detector produces in Findings & Results.

Last updated on