01
Grounded answers from the public site plus TXT, MD, CSV or JSON
AiUse Voice AI
A Voice AI agent for your website and phone system
Two voice channels with one operating logic: a visitor speaks to the widget in the browser, while the SIP version handles telephone calls. The agent answers from an approved knowledge base, gathers the facts your team needs and hands over usable context.
- Grounded answers from the public site plus TXT, MD, CSV or JSON
- Enquiry qualification and contact capture without manual transcription
- Telegram summary and a full conversation sent to email or an agreed CRM
A WebRTC widget and a SIP agent are different products
A live conversation in the visitor’s browser over WebRTC — no phone number, SIP or PBX. A separate implementation for inbound and controlled outbound telephone calls.
| AiUse Voice AI Widget | AiUse Voice AI for SIP |
|---|---|
| Visitor’s web browser | Telephone network over SIP |
| A live conversation in the visitor’s browser over WebRTC — no phone number, SIP or PBX. | A separate implementation for inbound and controlled outbound telephone calls. |
| Learn more: AiUse Voice AI Widget | Learn more: AiUse Voice AI for SIP |
What the business receives
The manager receives qualification, contact details and a concise Telegram summary; the full conversation goes to email or an agreed CRM.
02
Enquiry qualification and contact capture without manual transcription
03
Telegram summary and a full conversation sent to email or an agreed CRM
How launch works
Provide the website URL and define the purpose of the conversation.
We create a draft RAG knowledge base and agree facts, role and prohibited claims.
Your team tests a temporary WebRTC demo; a SIP pilot is set up separately.
After acceptance, you receive the widget embed or the configured calling workflow.
Who this fits and when
Voice AI is useful when the same first-line questions recur every day, response speed affects the sale, and a manager needs context before joining personally. It is not a replacement for a negotiator: the agent handles first contact, routine explanation, qualification and a deliberate handoff.
By default the workflow retains a transcript rather than raw audio. Retention, recipients and deletion are agreed before launch; transcripts, phone numbers, emails, keys and RAG content are excluded from advertising analytics.
What happens inside the system
The voice interface is only the visible layer. Behind it sit speech recognition, the dialogue model, retrieval from the approved RAG knowledge base and speech synthesis. Each component can fail differently, so acceptance uses real questions and tests factual accuracy, response delay, language switching, contact capture and human handoff separately.
A knowledge base is not an indiscriminate copy of the website. We separate verified facts from promotional claims, remove stale prices, establish source priority and state what the agent does not know. Instructions found inside uploaded material are treated as untrusted content, never as commands that can silently change the agent’s policy.
After launch, raw conversation count is not the success metric. We inspect whether the visitor understood the offer, whether the required qualification fields were captured, whether usable context reached the manager and whether a failed session can be explained. Transcripts support controlled improvement while personal data stays out of advertising systems.
Control, security and honest limits
Quality depends on the microphone, browser, knowledge base and chosen model. The AI must not invent prices, terms or promises; complex cases need a human handoff.
Scope and price are confirmed after the call flow, integrations, languages and expected conversation volume are checked.
Questions and answers
Does the website widget need telephony?
No. It runs over WebRTC in the browser. SIP is only required for the separate telephone product.
Can the agent speak several languages?
Yes. Languages, the opening language and switching rules are set during configuration.
What happens after a conversation?
The manager receives structured qualification and a summary; the full transcript only goes to approved destinations.
Who reviews the knowledge base?
AI prepares the draft, but an administrator reviews facts, sources, allowed answers and human-handoff rules before launch.
How is usage spend controlled?
Daily sessions, concurrent conversations, rate limits, maximum duration and pause are configured for each widget or calling workflow.
Build a demo from my website
Two voice channels with one operating logic: a visitor speaks to the widget in the browser, while the SIP version handles telephone calls. The agent answers from an approved knowledge base, gathers the facts your team needs and hands over usable context.