How Sources Work
A source is a connection to a system you already run — a database, a data lake, a collaboration tool, a content platform — that you want Classifyre to look inside. Once connected, Classifyre scans the source, pulls out individual items (an asset per table, file, page, message, or video), runs detectors over them, and records any findings.
Every source type has its own dedicated reference page generated from its schema — see the Catalog for the exact fields each one accepts. This guide explains the ideas that apply to all of them.
The shape of a source
However different two systems are, every source you configure is built from the same parts:
| Part | What it is | Learn more |
|---|---|---|
| Connection | The mandatory details to reach the system, plus secrets like tokens and passwords | Configuration & Fields |
| Optional settings | Per-system tuning — filters, limits, scope | Configuration & Fields |
| Detectors | The built-in and custom detectors that run on the content | Detectors |
| Sampling | How much data is read each run, and in what order | Sampling Strategies |
| Content extraction | Whether to read images (OCR) and audio/video (transcription) | OCR & Transcription |
| Resources | Compute limits for large scans (advanced) | This page, below |
| Schedule | An optional recurring scan time | Testing & Scheduling |
Only two parts are mandatory for every source: the required connection fields and a sampling strategy. Everything else has sensible defaults.
From connection to findings
Here’s the journey a source takes, end to end:
- Configure — provide the connection details and choose your options.
- Test — confirm Classifyre can actually reach the system before committing to a full scan. See Testing & Scheduling.
- Scan — kick off a run by hand, on a schedule, or let it happen as part of your workflow.
- Discover & read — the scanner lists everything in the source and reads a slice of it as assets, governed by your sampling strategy.
- Detect — detectors run over each asset’s content.
- Findings — anything a detector flags becomes a finding, ready for triage and investigation.
Want the detail on what a scan does between “running” and “done” — including how repeat scans diff against the last one? That lives in the Flow section. What happens to findings after they appear is covered in Investigations and Autopilot.
Assets: the unit of scanning
Whatever the system, Classifyre normalises its contents into assets — one per meaningful item:
| Source kind | One asset is… |
|---|---|
| Database / warehouse | A table (read in row batches) |
| File / object storage | A file |
| Collaboration tool | A page, message, or document |
| Content platform | A video, post, or article |
Each asset carries metadata (its name, location, and source-specific details) and content (the text — or, with OCR/transcription, the extracted text). Both are available to detectors. Every source’s reference page lists exactly which metadata it extracts.
A note on resources (advanced)
Large scans can be tuned with resource settings — CPU and memory limits, a runtime timeout, and how many items are processed in parallel. The defaults are fine for most sources; you only need these when scanning very large systems or when running on constrained infrastructure. Each setting is documented on the per-source reference pages.
Next, get specific about the fields you fill in: Configuration & Fields.