Skip to main content
Shipped 2026-03-20 · Capture Asking users to choose between “note” and “attachment” on every import was friction with no upside — researchers shouldn’t have to think about Bower’s data model to capture an observation. Bower now picks for you, based on what the file actually is.

​
How it decides

File typeWhat Bower does
Audio (voice recording)Always becomes a note (transcript). The original recording stays attached.
Word document, text, markdownAlways becomes a note. Original file attached.
ImageBird reads the text. High or medium confidence → note (with the image attached). Low confidence → stays as an image attachment.
PDFStays as a PDF. Bower extracts the text in the background so the PDF is searchable, but it’s not editable.
Spreadsheet, CSVStays as an attachment. Renders inline in read-only mode.
Everything elseStays as an attachment.

​
What stays in your control

  • Manual override — any image attachment or PDF can be converted to a note from the artifact menu. Useful when you want to edit a PDF protocol or annotate an image whose text Bower wasn’t sure about.
  • Image descriptions for everything — even when an image stays as an attachment, Bower generates a short description of what’s in it so you can find it semantically (“the photo of the shaker with the red cap”).
  • AI titles by default — voice notes and photo notes get generated titles unless you set one explicitly.

​
Why this matters

The split between notes and attachments was a mental model only Bower cared about. Removing that decision from the user makes capture faster, and the automatic image-to-note conversion (when confidence is high) is the difference between “snap a photo” and “snap a photo and get a structured note out of it.” That’s the workflow that makes mobile capture worth using.