Voice & Media Understanding
Your AI agent understands incoming voice notes, photos, and video — and on WhatsApp it can reply with a spoken voice note too.
Two Capabilities
Waslo handles rich media in two distinct ways. It helps to keep them separate:
- Inbound understanding — customers send a voice note, photo, or video and the AI understands it, then replies in text.
- Voice replies (Growth) — the AI answers with a spoken voice note, not just text.
The first works on every major channel. The second is a Growth feature and, today, sends spoken replies on WhatsApp. Details below.
Inbound Media Understanding
Customers don't always type. They record a quick voice note, snap a photo, or send a short clip. Waslo's AI is multimodal — it can look at and listen to what a customer sends and answer based on it, in text.
Live on:
- Telegram
- Messenger
What customers can send
| Media | Example use |
|---|---|
| Voice note | "Do you have this in stock?" spoken instead of typed |
| Photo | "Is this the right part?" / a photo of a product or model |
| Video | A short clip showing a damage report or a fault |
Real-world uses: a customer photographs a product and asks if it's available; someone sends a picture from an appointment; a client records a voice question in their own language; a buyer films a defect for a warranty claim. The AI takes the media in and replies in text, grounded in your Knowledge Base and Catalog just like a typed message.
Replies stay text
For inbound media understanding, the reply is text on all four channels. That's deliberate — understanding the media and speaking back are two separate features (see Voice Replies below).
Voice Replies (Growth)
On the Growth plan your AI can answer with a real spoken voice note — a natural-sounding audio reply, in the customer's language.
The mirror rule
Waslo never surprises a customer with an unexpected voice note:
- A customer sends a voice note → the AI replies with a voice note.
- A customer sends a typed message → the AI replies with text.
So voice out only happens when the customer chose voice first. A text copy of every spoken reply is also saved to the conversation, so your team can read what was said.
Where voice replies work today
| Channel | Understands incoming media | Replies with voice |
|---|---|---|
| Yes | Yes | |
| Telegram | Yes | Text only |
| Yes | Text only (API can't send audio) | |
| Messenger | Yes | Text only |
Spoken replies currently go out on WhatsApp. On Telegram, Instagram, and Messenger the AI still understands incoming voice, photos, and video — it just answers in text for now. Instagram's API can't send audio at all, so voice replies aren't possible there.
Enabling Voice Replies
- Be on the Growth plan with an active subscription.
- Open AI Agent → Voice Replies.
- Turn on the voice-replies toggle for your organization.
- Pick a voice — choose per language and gender, and preview it before saving.
- Save. From then on, voice-in gets voice-out on WhatsApp, following the mirror rule.
Credits
Media costs more than plain text because it uses multimodal AI:
- A text reply is the baseline cost.
- Voice, image, and video events (understanding incoming media, or synthesizing a spoken reply) cost more per event than a text reply.
On PAYG these come out of your credit balance; on Growth, messaging is included. See Billing & Plans for the current rates.
Honest Limitations
- Voice replies are WhatsApp-first. Telegram and Messenger understand incoming media but reply in text today; Instagram can't send audio at all.
- Arabic voice is Modern Standard Arabic. Spoken Arabic replies use MSA (
ar-XA) — there are no regional dialect voices at this time. - Voice replies require Growth. Inbound understanding is broader, but spoken replies need an active Growth subscription plus the org toggle on.
- Media has size limits. Very large files may not be processed.
FAQ
Can my customer send a voice note and get a spoken reply back? Yes, on WhatsApp with Voice Replies enabled (Growth). The mirror rule means a voice note in gets a voice note out. On Telegram, Instagram, and Messenger the AI understands the voice note but replies in text.
Will typed customers suddenly get voice notes? No. Text in always gets text out. Voice out only happens when the customer sent voice first.
What language does the voice reply use? The customer's language. Arabic uses Modern Standard Arabic.
Can the AI understand a photo of a product? Yes — it's multimodal. It can reason about the image and answer in text, using your Knowledge Base and Catalog.
Do I need Growth for media understanding? Inbound understanding of voice, photos, and video is available on the supported channels. Sending spoken voice replies is the Growth feature.
Is there a text record of spoken replies? Yes. A text copy of every voice reply is saved to the conversation.
Next Steps
- AI Agent Configuration — set up Voice Replies and the Product Catalog
- Knowledge Base (RAG) — ground every reply in your own content
- Billing & Plans — credits, packs, and the Growth plan