Dpro-at/Tel-Agent

Tel‑Agent – Open‑source AI phone‑assistant gateway

What it is – Tel‑Agent is a self‑hosted gateway that connects a traditional SIP phone line (or any SIP‑compatible PBX) to an AI agent. Incoming calls are routed according to user‑defined rules: they can be passed to a human, blocked, or answered by an AI that talks in real‑time, can call external HTTP endpoints, take messages, check calendars, etc. The same AI agent can also converse over dozens of messaging channels (web chat, SMS, email, WhatsApp, Telegram, Discord, Slack, …) using the user’s own credentials.

Why it matters – Commercial “AI receptionist” services exist, but they are closed‑source, cloud‑only, and keep recordings and API keys in the vendor’s hands. Tel‑Agent lets you run the whole stack on your own hardware, choose any LLM or TTS/STT provider (or run local models), and keep all data under your control.

Core features

Multi‑channel inbound routing – SIP voice plus 24+ messaging platforms (WhatsApp, Telegram, Discord, Slack, email, etc.).
Real‑time AI conversation – streaming speech‑to‑text, LLM generation, and text‑to‑speech; first audio response under 800 ms.
Tool calling – the agent can invoke arbitrary HTTP endpoints, enabling actions like calendar lookup, ticket creation, call transfer, etc.
Full recording & transcription – every interaction is stored, searchable, and GDPR/EU‑AI‑Act compliant.
Self‑hosted deployment – native installers for Windows/macOS/Linux, Docker‑Compose, and Caddy‑based HTTPS reverse‑proxy.
Pluggable back‑ends – default uses Deepgram (STT), ElevenLabs (TTS) and a cloud LLM; local models (Ollama, Whisper, Piper) are planned for v1.1.
Multi‑language support – UI and agents currently in English, German, Arabic, Spanish, Dutch; i18n contributions welcomed.

Architecture at a glance

Layer	Technology
Voice/agent runtime	Python + LiveKit Agents
HTTP API	FastAPI (Python)
Front‑end dashboard	Next.js + React
Persistence	SQLite (default) – optional PostgreSQL
Cache / queue	Redis
TLS termination / reverse proxy	Caddy
Packaging	Windows installer, macOS pkg, Linux DEB/RPM, Docker‑Compose

Quick start (Docker)

git clone https://github.com/Dpro-at/Tel-Agent.git
cd Tel-Agent
cp .env.example .env               # set ENCRYPTION_KEY (e.g. openssl rand -hex 32)
# edit .env to add your STT/LLM/TTS keys and channel credentials
docker compose up -d --build       # starts API, dashboard, LiveKit, Redis, etc.

Dashboard: http://localhost:38471
API docs: http://localhost:38472/docs
Use the released docker-compose.release.yml to skip the build step, or run a native installer from the releases page.

Typical use cases

AI receptionist – automatically answer business calls, provide information, schedule appointments, and hand over to a human when needed.
Omni‑channel support – a single AI assistant handles voice, SMS, WhatsApp, Slack, etc., giving customers a consistent experience.
Internal tools – let the agent trigger internal APIs (e.g., ticketing, inventory) via the generic HTTP tool.
Data‑owner compliance – keep recordings and transcripts on‑premises to satisfy GDPR or EU‑AI‑Act requirements.

License & commercial option

AGPL‑3.0 – any network‑served modifications must be published.
A commercial closed‑source license is available from Dpro GmbH for enterprises that need it.

Where to learn more

Full roadmap: docs/ROADMAP.md
Contribution guide: CONTRIBUTING.md (includes an onboarding tutorial)
Security policy: SECURITY.md
API specification: docs/SPEC.md

TL;DR – Tel‑Agent is an open‑source, self‑hosted bridge that lets you run an AI‑powered phone receptionist (and many other messaging channels) on your own infrastructure, with full control over data, models, and integrations.