Here's the gist: but obviously there is much more to activating this service. The service is dependent on whether a business wants BOTH inbound & outbound calling to occur. Assuming the use case is for incoming calls as an example so as to not miss calls or have an Ai agent take on the initial inquiry and provide useful responses, then the following should be applicable. Among some factors to be considered depends on whether the phone line is a copper wire land line or a digital line such as a cellphone carrier. Many copper wire lines may only allow immediate forwarding to another phone number for ALL incoming calls. ALL calls get forwarded. There are no "conditions" that can be applied. Digital lines on the other-hand usually can have specific codes applied that let the incoming call get sent to the Voice AI Agent seamlessly to step in under specific conditions like after specific hours of operation when the business may not be available, forward all calls to the virtual number to be handled by the Ai agent. Both types of lines can be used for a voice Ai agent. It is the forwarding options that might be affected by the line type. It may be possible that the businesses that you are in contact with may not be grasping how Voice AI would need to be implemented and the full benefits. Voice Ai is an "automation option" with benefits which can use their current business phone number but to use the automated aspect, provisioning a secondary "virtual" business number through an AI voice platform (like Retell AI, Vapi, eleven labs, livekit or Twilio.. many others) needs to be involved. This virtual number makes use of the actual AI Voice Agent software and automation options. Platforms like n8n, Make or zapier can be used to manage the specific automating of how the Voice Ai manages calls including the trigger points of what to do when a call is received and the information to provide. If a business does not want a return call automatically or automated interactions to occur such as booking an appointment, or an immediate option for an Ai agent to answer customer questions, and respond to inquiries, or simply PREVENT A MISSED CALL by forwarding live leads to be handled by a Voice Ai agent, they probably should stick with a physical receptionist, hire a call center for live human interaction or simply "set-up" and use voicemail.

Hey Mohammad, if you're open to a collaboration, I'll spill all I know on both real estate & HVAC + a few more. ~cheers Rc

@Sumeyye Bozkus - For realtime STS voice, I started with this config and recently have been testing for latency improvements. [ use case ex - restaurant order works well ] This may sound a like a bit of over-think but it works. Here is the gist of the flow. It is a modular option I can use/swap for multiple uses. Environment Stack: OS: Ubuntu 24.04 LTS (VPS / 8GB RAM). Orchestration: Docker Compose (Isolated Services). Primary Logic: n8n (v2.12.1 at the time) via HTTP Request nodes. A separate setup for RAG voice using a Telegram bot Website-to-LiveKit setup first with n8n logic to handle a "Voice-to-Voice" interaction Docker Stack: n8n, Redis, Postgres, ChromaDB (Core 0) AI Engine: llama.cpp + Llama-3.2-1B (Core 1) Voice Agent: Python LiveKit Worker + Faster-Whisper + Piper RAG Logic: n8n Webhook + Markdown Ingestion. Monitoring: Telegram Expert-in-the-loop Audio Networking: LiveKit Cloud (Free Tier) handles the WebRTC switchboard, The Worker (VPS): Python-based agent connects to LiveKit Cloud. It performs STT (Faster-Whisper Tiny), LLM (Llama-3.2-1B via llama-cpp), and TTS (Piper). The RAG Brain (n8n): The Agent Worker sends a text query to an n8n Webhook. n8n searches the Markdown files (Vector Store - Chromadb) and returns the context The n8n RAG Workflow - How it works: Created a workflow in n8n with these nodes: Webhook Node: Listens for POST requests from my Agent Worker VPS. Vector Store Node: Use a local engine (e.g., chromaDb) to index Markdown files. Code Node: Formats the retrieved context into a prompt. Response Node: Sends the context back to the Webhook. *** There is a python script call on the VPS for the LiveKit Agent Worker (Python). This script runs as a service on my VPS. It bridges the voice stream to my n8n workflow. This script connects to my LiveKit Cloud account for the audio stream and calls my n8n Webhook to retrieve Markdown context (RAG). The script uses Faster-Whisper for transcription and Piper for speech to decrease latency.

@Sumeyye Bozkus You can use slack, whatsapp, gmail, .. swap out and place your communication app of choice ~cheers Rc

you can use n8n and MCP with LiveKit ~cheers Rc