Speak naturally · Vendix listens Speak the order. It's an order.
The fastest data-entry method ever invented is a human voice - and the messiest place to use a touchscreen is a hot, crowded outlet with one hand on a clipboard. Vendix lets field reps speak transactions the same way they speak to a customer: naturally, mid-sentence, in whatever language fits.
Heard
"Sunil's grocery-eke, twelve cases Panadol 500, hari hari mata last invoice eka-ta same price."
Sinhala-English code-switch · 1.2s STT · 4 tool calls executed
Listening...
Tap to stop · or just say "done"
Live transcript
"Twelve cases Panadol 500..."
Designed for the way reps actually work.
A field rep balances samples, a tablet, an outlet owner, and traffic. Tapping through drop-downs is the fifth thing on their mind. With voice, the rep speaks the order while they're still in front of the buyer - no “hold on, let me enter this” moment where the customer's attention drifts.
- One-thumb activation - press-and-hold the mic button, speak, release. No screens to navigate first.
- Confirmation before commit - Vendix repeats back the parsed order; rep says “yes” or corrects.
- Works through earbuds - drivers stay legal; reps in noisy outlets stay accurate.
Time to capture · 1 invoice, 4 line items
4× faster. Measured on the same tenant data with experienced reps. Voice-skeptical reps converted within two days - and never went back.
English, Sinhala, Tamil, Portuguese, Arabic. Mixed.
Real field talk isn't pure anything. A rep will switch from one language to a brand name in English to a district name in another language in one breath. Most voice systems break at the first language switch. Vendix doesn't - the AI parses the mixed-language utterance naturally, and product codes and outlet IDs are language-agnostic anyway.
- Speech-to-text uses the phone's built-in engine - no extra cloud trip, no extra latency.
- App and chatbot interface fully localised in English, Sinhala, Tamil, Portuguese, and Arabic.
- Confirmation prompts use the tenant's primary language by default.
Voice is the chatbot, just spoken.
Voice doesn't bypass the platform - it amplifies it. The transcript runs through the same intent classifier, hits the same 41-tool catalog, dispatches the same Mediator commands, lands in the same audit trail. There is no “voice exception” where a rep can do something they can't do via the chatbot or the portal. The platform is one platform, all the way down.
- Role-aware scoping enforced server-side - voice has no privilege escalation.
- FluentValidation runs on every command voice triggers.
- The same audit row stores the transcript, the intent, the tools, the cost.
Three input modes · one tool layer
Typed (chatbot)
Web portal · Android Compose UI
Voice
Android · STT → same prompt
API
Mobile JWT endpoints · system-to-system
All three converge on
41 Vendix tools · 1 audit log
Built for actual fieldwork.
Noise-resistant
Generator hum, motorbikes, idol music in the next aisle - STT engines on modern Android handle it. Vendix shows live confidence so the rep knows when to repeat.
Offline buffering
Captured audio + transcript queue locally if the chatbot endpoint is unreachable. Replays when the rep gets coverage. Same offline-first pattern as the rest of the app.
Confirmation loop
Vendix never commits a write tool from voice without verbal confirmation. Read tools (price lookup, stock check) execute immediately.
No special hardware
Works on any Android phone with a mic. No headset required. No firmware. No "approved devices" list to maintain.
Privacy by default
Audio never leaves the device. Only the transcript is sent. Tenants can disable voice per role or per user.
Audited identically
Voice transcripts and tool traces sit in the same ChatMessages table as typed messages. One query gives finance the full picture.
Hear the difference voice makes.
On a demo we'll hand you a phone pre-loaded with a sandbox tenant, ask you to speak three orders, and show you the audit trail before you put the phone down.