Tel-Agent

Closed voice-agent and omnichannel platforms have made it straightforward to wire an AI agent to a phone number, at the cost of sending every call and every API key through someone else's infrastructure. Tel-Agent, built by Dpro GmbH in Vienna, offers a self-hosted alternative: connect any phone line to any AI model, bring your own keys, run it on your own hardware. Incoming calls arrive over SIP and are routed by rule, to a human, blocked, or handed to an AI agent that can transfer calls, take messages, check calendars, and call arbitrary HTTP endpoints, with every interaction recorded and transcribed.

The distinctive part is breadth of channel coverage rather than telephony alone. The same agent configuration handles phone calls plus more than 24 other channels, including web chat, SMS, email, WhatsApp, Telegram, Messenger, Instagram, Discord, and Slack, all through credentials you supply yourself. The stack is a real, multi-service application: a Python and LiveKit Agents voice pipeline, a FastAPI backend, a Next.js and React frontend, SQLite or PostgreSQL for storage, Redis for caching and queues, and Caddy as reverse proxy, wired to providers like Deepgram for speech recognition and ElevenLabs for speech synthesis. The project targets under 800 milliseconds from the end of caller speech to the first audio response.

The natural fit is organizations that want an AI receptionist or omnichannel support agent but are unwilling to route customer calls and messages through a third-party SaaS platform, whether for data-residency, compliance, or cost reasons. Because Tel-Agent is backed by a registered company, Dpro GmbH (FN 631492s, Vienna), rather than an anonymous repository, it carries more institutional weight than a typical side project, and it ships Docker images and five interface languages to support real deployment rather than a demo build.

The caveats center on maturity and licensing. The project carries an explicit alpha badge, and only two pre-releases exist so far, v0.1.0 and v0.1.1, both from the same week in September 2026. The license is AGPL-3.0, which means running a modified version as a network service obligates you to publish those modifications, a real constraint for anyone building a closed commercial product on top of it, though Dpro GmbH sells a separate commercial license for exactly that case. The sub-800-millisecond latency figure is the project's own target for its pipeline, not an independently measured guarantee across real-world network conditions.

The verdict: Tel-Agent is a credible, actively developed answer to a specific gap, a self-hosted, own-keys alternative to closed voice-and-chat agent platforms, made more trustworthy by a real company standing behind it and a working dual-license model. Adopt it if you want to pilot a self-hosted omnichannel AI agent and are comfortable with alpha-stage software and AGPL's network-service obligations, or are prepared to buy the commercial license. Hold off if you need a finished v1.0, closed commercial redistribution without a paid license, or latency guarantees beyond the project's own stated target.