Castellan
Overview · Clerk
The Clerk, which makes your screenshots and scans searchable
What the Clerk reads (screenshots, scanned documents and PDFs), how it reads them on this PC, what its captions add for pictures with no text, and its settings.
- Article
- 2001
- Applies to
- Clerk 0.5.16
- Last reviewed
- For
- For everyone
What the Clerk does#
The clerk wrote the manor's rolls and accounts. The Clerk reads the text in your screenshots and scanned documents and keeps it, so "the screenshot with that error", or "the letter from the bank", can be found again by what it says.
Every 15 minutes it looks for new and changed files, reads the text in them, and adds them to its rolls. Its page searches them. Nothing leaves this PC.
It holds its office on an everyday PC. On a gamer's PC, the Scribe holds it, keeping your game screenshots by game, and the Clerk works behind it.
The Clerk comes with the Household license.
What it reads#
| Where | What |
|---|---|
| Your screenshots | The Screenshots folders Windows and OneDrive use. |
| Your scans | Documents\Scanned Documents and Pictures\Scans, where the Windows scanning apps save. |
| More PDF folders | Any you add, in Settings. |
It reads pictures (PNG, JPEG, WebP, BMP and GIF) and PDFs, the first 5 pages of each. Subfolders are read too.
It only reads them. It never moves, renames or changes a file.
How it reads them#
The text is read with Windows' own text recognition, on this PC, the same every time. No AI is involved.
Pictures with almost no text (a photo, a diagram) get a short caption from a small AI model on this PC, "A router on a desk with its lights on", say, so they can be found by what they show. Each caption says where it was written: "note from the NPU, unverified". A caption is a guess, and can be wrong.
Secrets are hidden. Anything that looks like a password, key or token in a screenshot is replaced with [redacted] before it's kept, and that picture is never shown to the AI.
At most 300 files are read a round, newest first. On a first look at a big folder, the rest follow over the next rounds.
Its page#
The Clerk's page is at http://clerk.localhost:18787/, and Open on its card in Castellan goes there.
- The search box, and the results: each with its thumbnail, the passage with your words highlighted, where the file is, its date, and Show in Explorer.
- The last round: how many files are in the rolls, how many are waiting, and how the round went.
- What it reads: the folders it read last.
- Run now and Settings are in the title bar.
See Find a screenshot or a scan.
Its settings#
| Setting | Default | What it does |
|---|---|---|
| Screenshot folders | Automatic | The folders of screenshots it reads. Automatic: wherever Windows and OneDrive keep them. |
| Scanned-document folders | Automatic | The folders of scans. Automatic: Scanned Documents and Scans, where they exist. |
| More PDF folders | None | More folders whose PDFs it reads. |
| Caption pictures | On | A short caption for pictures with almost no text. |
| Files read per round (Advanced) | 300 | The rest wait for the next round. |
| Largest file read (Advanced) | 100 MB | Bigger files are listed, not read. |
| Re-rank by meaning (Advanced) | On | Puts the closest matches first, when the local AI's search model is already running. |
| A round every (Advanced) | 15 minutes | 5 minutes to a day. |
What it can't read#
- OneDrive files that are only online. Reading one would download it. Right-click the folder and choose Always keep on this device to include it.
- Handwriting, which reads poorly, and very small text in a big screenshot.
- TIFF, HEIC and Office documents, in this version.
- More than 5 pages of a PDF.
Related articles
Is this page right?
If something on it is wrong or out of date, tell us and we'll fix the page.
Still stuck? Write to support@castellan-software.com and mention article 2001. Every version of Clerk, and what changed in it, is in its release notes.