Summary
Bart Slodyczka demonstrates how to build a 24/7 phone-answering voice agent using open-weight models — specifically Kimmy K2.6, Qwen, and GLM — hosted through the Talinex platform. The video opens with a live demo call to a fictional carpet-cleaning business, where the agent named Sam handles a stain inquiry, checks a Google Calendar for availability, books an appointment, and confirms the details — all without human intervention.
The build tutorial covers the full Talinex configuration: choosing between self-hosted open-weight models and closed providers like Anthropic or OpenAI, with concrete latency and cost comparisons (Kimmy K2.6 at $0.053/minute vs. GPT at roughly double the latency), writing the single-prompt agent instructions, and deciding between inline knowledge bases and external RAG tools for business-specific information.
The integration section shows how to connect Google Calendar via a custom MCP server built in N8N, exposing two tool calls — one to check the next seven days of availability and one to create a booking. Slodyczka finishes by assigning a real phone number through Talinex’s calling tab so the agent is reachable immediately. The tutorial is well-paced for developers and small-business builders looking to deploy a working voice agent on open-weight models without standing up their own infrastructure.
📺 Source: Bart Slodyczka · Published September 02, 2026
🏷️ Format: Hands On Build
