Download
Browse the docs

Sources

Last updated 20 October 2026

Browse every image, PDF, video and recording in your vault in one gallery, see how far each has been turned into text, and clear out originals you no longer need.


Click Sources in the sidebar, or press ⌥⌘A. The gallery shows every image, PDF, video and audio file in your vault, from screenshots, voice takes and screen recordings to files attached to notes, as a grid grouped by month. Audio and video tiles show their length. Hidden folders and your chat archive are left out, and a file iCloud hasn’t downloaded to this Mac yet appears once it has. If Sources is missing from the sidebar, turn it on in Settings ▸ Features.

Click to select, ⌘-click to add a file, ⇧-click for a run, and ⌘A for everything on screen. Esc clears the selection. With several selected, right-click for Show in Finder, Extract Text, Clear Out Originals and Move to Trash.

Right-click one file for:

  • Open and Open in Default App.
  • For a file a note uses: Extract Text Now or Transcribe Now (if it hasn’t been read yet), Pull Key Frames (for a video), Mark as Used in a Note or Mark as Not Used, Clear Out Original (Keep Text), and Open the Note.
  • For a picture: Annotate, Pin to Screen and Copy Text in Image.
  • Show in Finder, Copy Markdown Link (paste it into a note to show the file there), Copy Path and Move to Trash.

The viewer

Double-click a tile to open it: pictures zoom, PDFs page and videos play. ← and → step through what is on screen, and Esc closes the viewer. Its info button shows the file’s text or transcript, its details, and the notes that link to it.

On a picture, ⌘C copies it, and right-click offers Copy Image, Copy Text in Image, Copy File Path, Share…, Open in Default App, Show in Finder, Annotate and Pin to Screen.

A recording names the note it belongs to and offers Open Note and Transcribe (Re-transcribe once it has words). A recording no note owns, such as a menu bar voice take, offers Transcribe into a New Note, which makes a note for it in the same folder. On the Transcript tab (Text for pictures and PDFs), the swap button gives the words the big pane and moves the file into the panel; it stays that way until you swap back.

  • Kind. The four squares show only Images, PDFs, Videos or Audio. Click the lit one again to show everything.
  • Stage. Shows one of the three stages below.
  • Sort. Newest first, Oldest first, Name or Largest first.
  • Search. Matches file names and folder paths only. It doesn’t look at the text inside files. To find a screenshot by its words, use /ss in the command bar; see Screenshots.

The three stages

A file attached to a note carries a badge saying how far the note’s AI can use it:

StageMeaning
OriginalNothing has been pulled out of it yet, so the AI can’t read it.
Text extractedIts text has been pulled out, and the AI can read it.
Used in a noteIts content has been folded into a note.

Files added to a note are read automatically: documents (PowerPoint decks included, slide by slide with their speaker notes and pictures) for their text, pictures with text recognition, audio and video transcribed, and a video’s most text-heavy frames, such as slides, saved as pictures too. An Excel workbook (.xlsx, .xlsm or .xltx) becomes one table per sheet under the sheet’s name, with hidden sheets marked; a long sheet stops at 300 rows and 40 columns. A PDF page that holds a picture, such as a chart, or is mostly picture is also kept as an image under a Pictures heading, for up to forty pages. If a file hasn’t been read, right-click it and choose Extract Text Now or Transcribe Now. A source counts as used when you mark it (Mark as Used in a Note), when you add its text to the note, or when the note’s AI chat edits the note with that source ticked. Files no note uses have no badge and count as Original.

A note’s Sources panel

Each note has its own list of sources: click the sources count at the top of the note, or press ⌘.. Click a row to open it beside the list; with one open, ↑ and ↓ walk up and down the list, opening each in turn.

Add sources opens a drop area, “Drop files and links here”, and Find sources opens the web search (see Finding sources on the web, below). Drag files, emails or web pages anywhere onto the panel, or use the buttons under Add sources:

ButtonWhat it does
Upload filesAdds any files you pick.
RecordRecords into this note. The take joins its sources when you stop.
Upload audioMakes an audio file you pick this note’s recording, and transcribes it.
Paste textAdds the clipboard’s text as a source of its own, titled from its first line.
Add a linkOpens “Paste a web address”. MindLake fetches the page and keeps its text, and an article’s figures.
Add a notePick one of your notes: a copy of its text as it is now joins this note’s sources, named after it.

A row’s right-click menu ends with Remove from Note, which leaves the file where it is, and Delete Source File…, which moves the file to the Trash with its extracted text. Delete asks first: Delete File, Remove from Note Only or Cancel. The ⋯ above the list has Select All, Remove All from Note and Delete All Source Files…, which says how much space it frees.

⌘-click or ⇧-click rows to pick several. The line above the list then reads how many are selected, with Remove and Delete…, and right-clicking one of them offers Remove N from Note and Delete N Source Files….

Clearing out originals

Once a source’s text has been extracted you can let the original go and keep the text, so the note keeps its grounding and the AI can still read it.

  • One file. Right-click it and choose Clear Out Original (Keep Text).
  • Many. Filter the gallery, for example to Used in a note, then open the stage menu: it offers to clear out every original on screen and says how much space that frees. Or select several and right-click.

Originals go to the Trash, so you can put them back. A source whose text was never extracted can’t be cleared, so the last copy of its content is never lost.

Finding sources on the web

To gather material for a note, open its Sources panel, click Find sources and type your topic where it asks “What are you trying to understand?”

  1. Up to five notes you already have on the topic are listed first, under In your vault.
  2. A shortlist follows under From the web: up to six pages with Fast Research, up to twelve with Deep Research. Each result has a line on what it is and why it answers your topic, and names any notes of yours that cover the same ground. Tick the ones you want, or use the cross to drop one.
  3. Save fetches the ticked pages and keeps their text, and an article’s figures with their captions, as sources. 5 more asks for more, and Discard clears the list. Nothing is added until you press Save.

To start from something you already have, right-click a source whose text has been extracted and choose Find More Like This. MindLake works out what it is about and runs the same search, leaving out the source itself.

The pills under the box choose which indexes to search and how hard to look. Fast Research, the default, runs up to three searches and reads up to six pages; Deep Research runs up to eight and reads up to twelve, which takes about a minute longer. Under the box and any results, Search indexes shows each index as a chip, filled while it is searched and outlined while it is off: click one to switch it. A chip with a key glyph still needs a key or a server address, and clicking it opens the settings. The ⓘ says what each index is. The sliders button opens the settings too, and the eye folds the chips away until you click it again.

Results that have little to do with your topic are dropped on your Mac before anything is fetched, so you may see an empty list rather than near misses. Pages already saved in your vault aren’t offered again.

Wikipedia, OpenAlex, Hacker News and Stack Overflow work with no setup. For general web results, add a key for Brave Search, Tavily, Exa, Kagi or Google Programmable Search, or a SearXNG server, in Settings ▸ AI ▸ Finding sources. With an AI model set up, your topic becomes several searches and each result gets a written reason; without one, it is a plain search. It only runs when you ask; see the privacy policy for what it sends.