How it works
From sign-up to the first selling stream in five steps. Under the hood there are seven layers: dashboard, avatar manager, AI Brain, catalogue, comments, orders and the Live Engine.
- 01
Create an AI seller
Sign up with email or Google, create a workspace and choose a country. The country defines currency, payment provider and data residency.
- Email + password of 10+ characters, or Google
- Plan in RUB via YooKassa or in USD via Stripe
- Russian stores' data stays in the Russian cluster
- 02
Choose an avatar
Open the showcase, pick a seller for your niche and tune the voice: language, rate, emotion. The preview plays your phrase in the avatar's voice.
- 6 system avatars across four niches
- Several TTS providers behind one interface
- Your own avatar on Business after moderation
- 03
Upload products
CSV template with preview and per-row validation; Shopify and YML later. Descriptions are chunked into the knowledge base, stock lives in its own table with reservations.
- Title, price, sizes, colours, stock, SKU, shipping, returns
- Row errors shown before import
- Store documents: FAQ, terms, warranty
- 04
Connect social accounts
YouTube connects through a single OAuth screen. The platform creates the broadcast and reads chat through the official Live Streaming API.
- YouTube Live + chat today
- TikTok and Instagram Live via partner access
- One stream to many platforms through the RTMP gateway
- 05
Go live
First a two-minute unlisted test stream with a chat check. Then scheduling, the operator console and a report after the stream ends.
- Test minutes are not metered
- Console: preview, comment feed, answer queue, current product
- Mobile monitor mode to keep an eye on the stream from a phone
The avatar can interrupt itself
While the presentation runs, a high-priority question interrupts speech: the avatar processes it, answers and returns to the script. It is a six-state machine.
- IDLE
- SPEAKING
- INTERRUPTED
- PROCESSING
- RESPONDING
- RESUME
Who gets an answer at a thousand comments a minute
Every comment is classified and queued by priority. Identical questions within a minute collapse into one answer.
| Comment type | Priority |
|---|---|
| "I'll take it", purchase intent | 10 |
| Product or price question | 9 |
| Availability, size, colour | 8 |
| General question | 5 |
| Small talk | 1 |
| Spam, abuse | 0 |
Tools instead of hallucinations
The model does not place an order in prose — it calls functions. Every function works only with your store's data.
- get_product()
- check_stock()
- find_similar_products()
- get_shipping()
- create_cart()
- create_order()
- payment_link()
- get_customer_memory()
- apply_discount()
Process questions
How long does launch take?
With a ready CSV, about an hour: sign-up, avatar, catalogue import, YouTube connection, test stream. Most of the time goes into preparing product descriptions.
Do I need an operator during the stream?
No, the avatar runs autonomously. The operator console lets you step in: answer manually, switch the product, stop the stream. A monitor mode is available on phones.
What does the shopper see?
A normal YouTube broadcast and a short payment link in chat. The checkout page opens on the phone, no registration, receipt by email.
How does the avatar remember a shopper?
Every shopper has a profile: size, interests, past orders, language. In the next stream the avatar uses it in recommendations.
Ready to try?
The test stream is free and trial minutes do not expire before the first charge.