Uploading research

Everything you upload to Crux is parsed and indexed so Ask and search can use it as evidence. Uploads appear as File items in the research graph.

Supported sources

• Documents — PDF, Word (.docx), PowerPoint, plain text and markdown.
• Spreadsheets — Excel (.xlsx), CSV. Tabular files can also become research tables.
• Recordings — audio and video of interviews and sessions (MP3, M4A, WAV, MP4, MOV, and more). Transcription is a paid-plan feature; recordings can be up to 400 MB, roughly 45 minutes of video.
• Transcripts — interview and meeting transcripts as text or documents.
• Links — add a URL and Crux captures the page content.
• Google Drive (paid plans) — see Google Drive & Figma.

What happens on upload

1. Parsing — Crux extracts the text, including PDF text layers and Office document content.
2. Chunking — long documents are split into overlapping chunks.
3. Embedding — each chunk gets a vector embedding so semantic search and Ask retrieve exactly the relevant passages.
4. Preview — a preview is generated for quick reading inside the app.

Indexing and transcription run in the background — you can keep working while a long interview uploads and transcribes. On paid plans, Crux can also extract text from images (screenshots, scans, photographed slides) so they become searchable too: a vision model reads the image and the result behaves like any other text layer, ready to highlight, cite, and search.

Resyncing extracted text

If a file's text looks wrong or incomplete — an early parse that missed a section, or a source that has since improved — open the file, go to its Text tab, and use Resync. It re-reads the text from the source without deleting your work: existing highlights are re-anchored to their passages wherever they can still be matched, and the toast tells you how many were kept and how many need re-anchoring by hand.

Storage limits

Deleting a project removes its files, previews, and embeddings.

Uploading a large recording

Files over about 12 MB skip the usual request path and upload straight from your browser to Crux's file storage in parts, which is what makes a 300 MB interview recording practical. Two things follow from that:

• Keep the tab open until the upload finishes. Transcription and indexing carry on server-side afterwards, so you can leave once the file appears in Research.
• A dropped connection resumes as a fresh upload rather than a corrupt file.

"This file is too large"

The message names the file, its size, and your plan's cap. It means the per-file limit was hit, not that anything is broken:

• On Free, the cap is 15 MB — interview-length recordings need Pro or a Teams plan. If a trial has just ended, the account is back on Free and large uploads that worked yesterday will be refused today. See Plans & billing.
• On a paid plan, the cap is 400 MB, roughly a 45-minute recording. For anything longer, export audio only (M4A or MP3 rather than MP4) — it is a fraction of the size and transcribes just as well — or split the recording.
• If the whole project's storage allowance is full rather than the single file being too big, the message says so instead; delete files you no longer need or upgrade.

Tips for better AI output

• Prefer text-first files. Scanned image-only PDFs rely on image text extraction, which is less precise than a real text layer.
• Split monster documents. Several focused files retrieve better than one 300-page compendium.
• Name files meaningfully. File names appear in search results and citations.

Next

• Highlights & labelling
• Search across your project