Talk instead of type. Press the mic button, say what you need, let go. Nalo transcribes and processes it as if you’d typed it.
How it works
- In WhatsApp/SMS: hold the mic button, speak, release. Same as sending a voice note to anyone.
- In Nalito web: mic icon on the chat input, tap to start, tap again to stop.
Nalo transcribes the audio with OpenAI Whisper, then runs the transcribed text through the normal chat pipeline. Everything that works by text works by voice.
When voice beats typing
At the gas pump
“Eighty-five bucks on gas at Chevron” → Fuel expense logged
Your hands are full. The pump’s still running. Voice is the only sane option.
Walking between job sites
“Quote for Mr Patel, replace garbage disposal, $320 plus install $80, his address is 4415 Pine Ridge Drive”
30-second voice note → quote generated + PDF ready + client info saved, all before you reach the next house.
End of day in the truck
“Log expense 400 at Home Depot, 85 gas, and a 12 lunch, also Mrs Lewis paid cash, and schedule Mr Chen for tomorrow at 9”
5 things in one voice note. Nalo splits and executes them serially.
Supported flows
Every Nalo capability works by voice:
- Create quote
- Create invoice
- Schedule appointment
- Log expense
- Mark as paid
- Plan route
- Rename crew
- Add employee
- Edit client
- Send to client
- Update business profile
- Get reports / ask questions
Accuracy tips
Whisper is very good, but here are the patterns that fail:
Common failure modes
- Background noise at job sites — drill, traffic, wind. Hold the phone closer.
- Very fast speech — slow down slightly. Whisper prefers natural pace.
- Names vs sound-alike words — “Mrs Lewis” sometimes becomes “Mrs Louis”. Correct once; Nalo learns via memory.
- Numbers spoken funny — “two seventy-five” becomes “275”. “Two seven five” might not.
- Mixed languages mid-sentence — works but occasionally gets a word from the wrong language. Speak in one language per message.
Sweet spot
- 5–30 seconds of voice
- Speaking in the direction of the phone’s mic
- Complete thoughts with clear pauses between items
- One language per message
What Nalo transcribes
Visible to you
When you send a voice message, Nalo displays the transcription back in chat so you can confirm it got the right words:
🎤 “Quote for Mrs Lewis broken toilet two seventy-five”
✅ Quote Q-ABC12345 for Mrs Lewis — Broken toilet repair: $275.00
If the transcription is wrong, tap to re-send or say “actually no, it was two fifty”. Nalo adjusts.
Stored
Voice messages themselves aren’t stored — only the transcribed text is saved in your conversation history. After 24 hours the audio is deleted from temporary storage.
Multi-language voice
Whisper detects language automatically. So:
- English voice → transcribed as English → processed in English
- Spanish voice → transcribed as Spanish → processed in Spanish
- Mixed sentence → best-effort transcription, usually fine
If the business’s document language is fixed (say, English), and you voice-message in Spanish, Nalo chats in Spanish but generates the PDF in English. Same rules as typing.
Troubleshooting
“The voice note didn’t get transcribed”
Possible causes:
- Internet cut out mid-send — try again
- Audio was too short (<1s) — hold the button longer
- OpenAI Whisper service was momentarily down — retry
Nalo will always confirm back: if you don’t see a transcription echo within 10 seconds, re-send.
“Transcription is garbled”
- Record in a quieter spot
- Get closer to the phone
- Slow down slightly
“Right words but Nalo misunderstood the intent”
This is rare. Read the transcription, then re-say with text for the critical detail. Then: “Remember that when I say X, I mean Y” — Nalo saves it to memory.
Cost note
Transcription uses OpenAI Whisper, billed by audio duration. Covered in all plans. You won’t see a line item.
Privacy
- Audio files transit HeyNalo → OpenAI Whisper → transcript
- OpenAI doesn’t train on data sent through its API
- Transcripts are stored in your conversation history (same as typed messages)
- Audio files are deleted within 24h of transcription
Tips for busy days
- Batch voice notes — one 60-second voice at lunch handling 5 tasks > 5 typed messages spread across the morning
- Voice + correction text — send the voice, then tap to type a single correction (“the price was actually 275 not 225”)
- Tricky names — spell them the first time by voice (“M-R-S L-E-W-I-S”), then Nalo remembers