Your second brain only speaks text. That's a problem.

Most knowledge tools treat PDFs, images, audio, and video as attachments. Fabric treats them as content it can actually understand, search, and reason about.

Most knowledge tools were built to handle text. Notes are text. Wiki pages are text. Documents are text. The entire architecture, from storage to search to the AI layer, assumes the content is words on a page.

Everything else is a second-class citizen.

A PDF is an "attachment" you can view but the app can't search inside. An image is a file you can see but the search can't describe. A voice memo is an audio blob the app stores but can't transcribe. A video is a large file that sits in your library taking up space with its contents invisible to every other feature. Handwritten notes exist as photographs of paper that the app treats as images rather than text.

This text-only model made sense when knowledge tools were built. In 2026, with AI that can read documents, transcribe speech, describe images, and recognise handwriting, treating non-text content as opaque attachments is an architectural limitation, not a technical one.

What you're losing

PDFs are invisible. Research papers, client briefs, contracts, reports, manuals: all arrive as PDFs. If your second brain can't search inside them, every PDF in your library is unfindable by content. You can find it by filename (if you remember the filename) but you can't find it by what it says. The research paper about attention mechanisms is invisible to a search for "attentional control" because the search can't read the PDF.

Voice is lost. The voice memo you recorded after a meeting, the verbal idea on your commute, the phone conversation you wanted to remember: all stored as audio files that sit in your library as unsearchable blobs. The thought you captured perfectly in your own words is as unfindable as if you'd never recorded it.

Images are opaque. The screenshot of a design reference. The photo of a whiteboard. The diagram from a presentation. Each one contains information that's visible to you when you look at it and invisible to the search, the AI, and every other feature of the tool.

Handwriting is invisible. Notebook pages, sticky notes, sketches with annotations: photographed and stored but unreadable by the system. The handwritten notes from the workshop are in your library as images that the search can't process.

Video is a black box. Meeting recordings, lectures, tutorials, presentations: hours of content that sits as large files with no searchable index of what's discussed at which timestamp.

Each format gap means a category of knowledge that your second brain stores but doesn't understand. A second brain that only speaks text is a second brain that's deaf, blind, and illiterate to everything that isn't typed words.

Universal content understanding

Fabric treats every format as content to understand rather than files to store.

PDFs are read, indexed, and searchable by meaning. The research paper about attention mechanisms is findable through "attentional control" because the semantic search reads inside the PDF. Annotations on PDFs are searchable alongside the source text.

Audio and video are transcribed automatically. The voice memo from your commute becomes searchable text. The meeting recording becomes a transcript with timestamps. The lecture becomes a navigable document where you can search for the moment the professor discussed a specific concept.

Images and screenshots are understood by the AI. Screenshot sync captures your phone screenshots automatically. The semantic search can find images by description of their content.

Handwriting is scanned and converted to searchable text. Photograph your notebook and the handwritten content becomes part of your searchable library.

Ebooks are read and indexed like any other document. The book you're reading is searchable by concept alongside your notes, articles, and research papers.

The AI assistant works across all of these formats simultaneously. "What have I saved about customer retention?" synthesises from your text notes, the PDF report, the annotated article, the voice memo, and the meeting recording where the topic was discussed. One question, every format, one answer with citations.

A second brain that understands everything you feed it, regardless of format, is a fundamentally more capable tool than one that only understands text. Your knowledge doesn't arrive in one format. Your second brain shouldn't be limited to one format either.

Frequently asked questions

Does the AI really understand images? The AI can describe image content and find images by conceptual description. It's not pixel-level image understanding (it won't identify specific typefaces in a screenshot) but it handles descriptive search well: "minimalist packaging design" or "whiteboard diagram about user flow" will find relevant images.

How accurate is the voice transcription? The transcription uses frontier speech-to-text models and handles most accents and speaking styles well. Technical terminology may occasionally need correction. The transcripts are editable.

Can I search across formats simultaneously? Yes. A single search query spans notes, PDFs, web clips, voice memo transcripts, images, ebooks, and connected tool content. Results are ranked by relevance regardless of format.

What about video files? Those are huge. Video files can be stored and their audio track is transcribed for search. You don't need to watch a two-hour recording to find the five minutes where pricing was discussed. Search the transcript and jump to the timestamp.

Does handwriting recognition work in all languages? The handwriting scanner works best with Latin scripts (English, Spanish, French, etc.). Support for other scripts varies. The recognition improves as the underlying AI models improve.

How does PDF search compare to dedicated reference managers? Reference managers (Zotero, Mendeley) handle citation formatting and bibliographic data. Fabric's PDF search handles semantic content search: finding PDFs by what they discuss rather than by author, title, or keyword. Many researchers use both: reference manager for citations, Fabric for conceptual search across their full library.

Can the AI answer questions that require information from multiple formats? Yes. If the answer requires combining information from a PDF, a text note, and a voice memo transcript, the AI assistant synthesises across all three and cites each source.

What about formats that aren't supported? Specialised formats (CAD files, raw database exports, proprietary app formats) may be stored but their internal content isn't indexed for search. The search covers text-based documents, PDFs, images, audio, video, ebooks, and common office formats. The list of supported formats expands with updates.

Does this work with scanned documents (not just digital PDFs)? Scanned documents that have an OCR layer are fully searchable. For scans without OCR, Fabric can process the document to extract text. The quality depends on the scan quality and the clarity of the text.

Related reading: Your notes app can't search your files, Your second brain should think, How to build a second brain in 10 minutes. Related pages: Search, Audio and video transcription, Handwriting scanner, For your PDFs, For your photos, For your videos.