Voice — push-to-talk

The mic icon next to the chat input records audio in your browser, sends it to OpenAI Whisper for transcription, and drops the text straight into the composer for you to review and edit before sending. Works on both /frank and the bubble.

• First time: your browser will ask for microphone permission. Granting it once persists.
• Recording: tap the mic to start. The icon turns red and pulses while listening. Tap again to stop.
• Transcribing: a spinner shows while Whisper processes — usually 1-2 seconds for a 30-second clip.
• Editing: the transcript lands in the input as plain text. Edit anything that came out wrong, then send.

Voice cost is small (~$0.006/min) and counts against your daily Frank quota.

Documents — paperclip + drag-drop

Attach a PDF or image and Frank reads it with vision. Useful for:

• Bank statements — extracted as a transaction list with totals
• Super statements — fund, balance, contributions, fees
• Receipts — merchant, total, GST, line items
• Payslips — gross / net / tax / super, YTD figures
• Anything else — Frank falls back to a generic extraction (key facts + summary)

Two ways to attach:

1. Paperclip — click the icon next to the chat input, pick a file
2. Drag and drop — drop the file anywhere on the chat surface

After parsing, Frank renders a result card with the headline figures and three follow-up prompt chips you can click to drive the next step ("Categorise these transactions", "Log this expense", "Update my super balance"). Limits: 10 MB per file, formats: PDF, JPEG, PNG, WebP, GIF.

Document parsing uses Claude vision and counts against your daily Frank quota — about $0.05 per typical statement.

Privacy

Audio clips and document images are sent to the AI provider for processing and not retained beyond the request. Documents you upload via the receipt/import paths in Money are stored separately in your document library — that's a different surface; see Document storage.