Siberia Voice Agent

Siberia Voice Agent turns any business's assistant into a real-time voice agent, built entirely on the AssemblyAI Voice Agent API. A visitor clicks "Talk" and simply speaks: the agent answers questions about prices, opening hours, services, address and promotions using that business's own knowledge base (voice RAG), and captures leads mid-conversation — all with natural, low-latency speech. The entire real-time voice stack is AssemblyAI: speech-to-text (Universal-3.5), text-to-speech, semantic turn-taking and barge-in, and audio transport. We add the "business brain" through two AssemblyAI HTTP tools — search_knowledge (retrieval over each tenant's documents) and capture_lead (records the caller's name and contact, emails the owner and notifies the dashboard). Agents are provisioned automatically per business via REST (POST /v1/agents), with a voice-first system prompt, a language-matched native voice, and keyterms extracted from the knowledge base to boost transcription accuracy. The browser connects with a single-use session token, so the API key never leaves the server. It is multi-tenant, multilingual (Portuguese, English, Spanish, French) and embeddable — one row equals one voice assistant with your own knowledge and lead capture. In short, it gives a real voice to any business, without a call center. A live demo and an open-source (MIT) reference implementation are included.