Sovereignty · 2026-09-16 · 1:35

A phone line that answers with your own AI

Name: A phone line that answers with your own AI
Uploaded: 2026-09-16
Duration: 1 min 35 s
Description: A promising base for anyone building a private voice assistant — the honest work is replacing the cloud voice pieces and reading the licence before it faces clients.

Tel-Agent is a self-hosted gateway that connects phone (SIP) calls and 23+ messaging channels to AI agents, with recordings and transcripts kept on your own infrastructure.

In favour

Runs on your hardware; conversations and recordings stay in-house.
Model-agnostic, including local models via Ollama.

Worth watching

Voice defaults to cloud speech services — swap in local speech-to-text and text-to-speech for full sovereignty.
AGPL-3.0 and still alpha; treat as a reference architecture, not production yet.

Our takeA promising base for anyone building a private voice assistant — the honest work is replacing the cloud voice pieces and reading the licence before it faces clients.

Source: Tel-Agent (Dpro GmbH, GitHub)

← All Focus posts

AI-narrated · AI-generated content

Transcript

Before you connect a phone line to someone else's AI, one question is worth asking. Could that line answer with an AI you run yourself? Here is one answer. A self-hosted gateway that connects phone calls, and 23+ messaging channels, to AI agents. So the point is where the conversations end up. The recordings and the transcripts are kept on your own infrastructure, not on someone else's servers. What that buys you. The gateway runs on your own hardware. The conversations stay in-house, and so do the recordings, on a machine you control. It is also model-agnostic, including local models on your own machine. So the agent that answers the phone does not have to be a cloud model. Two things to watch. By default, the voice goes to cloud speech services. For full sovereignty, swap in local speech-to-text and local text-to-speech. And the licence is AGPL-3.0, and the project is still alpha. Treat it as a reference architecture, not as production. Not yet. So this is a promising base for anyone building a private voice assistant. The honest work is replacing the cloud voice pieces, and reading the licence before it faces clients.