Speak it. We type it. We seal it.
Voice notes don't belong in a notes app. They belong in a typed record, entity-resolved against everything else in your ledger, and sealed by the same provenance chain that protects your statements and receipts. Captured on the device, transcribed locally where possible, the audio is never the source of truth — the typed record is.
Voice is for the moment.
Records are for forever.
You're standing next to the car at the tire shop. You don't want to fish out a notes app and type "got new tires for $401, Costco." You speak it.
Most voice apps store the audio + transcript and call it done. Lossless does what you actually wanted — types a Vehicle · Service record, attaches it to the Y, watches your bank for the matching charge, and links them. The audio is the easy bit. The record is the hard bit.
Two weeks later, when your CPA asks "what did you pay for tires?", the answer is sitting there with the receipt attached and a seal you can verify.
How voice notes turn into useful, sealed data.
"Just got new tires on the Y, four hundred at Costco."
→ Typed as Vehicle · Service, linked to Model Y, awaiting Plaid match. Found a $401.27 Costco charge two days later — auto-linked, sealed.
"Plumber quoted eight-fifty for the Pine Ave water heater."
→ Typed as Maintenance · Quote, linked to 14 Pine Ave, status: quote received. When the invoice arrives by email, it closes the loop.
"Dr. Lee said increase Lipitor to 40."
Linked to provider, typed as Prescription change, awaiting Rx fill record.
"Promised Maria I'd cover dinner Sat."
Linked to Maria, calendar Sat. Becomes a chat answer if you ever ask.
"Drove to client at Pine Ave, 14 mi."
Tagged business mileage. Cross-checked Tesla trip log. Reconciled, sealed.
Audio is processed locally where possible.
iOS and macOS use on-device transcription via Apple's Speech framework. The audio never leaves your device unless you explicitly choose cloud processing for accuracy on a specific note. The transcript itself is sealed at the same moment the record is — same Ed25519 chain, same audit log.
On-device transcription
Apple Speech (iOS / macOS). Audio + transcript stay on your device until you explicitly sync.
Cloud (zero-retention)
Whisper-v3 via zero-retention API. Audio purged after transcription. Per-note opt-in.
"I dictate while I'm walking from the car. Two weeks later it's a clean expense report. No filing, no shoebox, no mental load — and when my CPA asks for proof, the seal is right there."
— Beta user, real-estate investor · Bay Area
You've seen the input. Now see how it's organized.
Next pillar: the Memory Stack. Structured records at the bottom, embeddings in the middle, an entity graph on top — all in the same RLS-protected Postgres.