Tel-Agent: Build Your Own Open-Source AI Phone Receptionist

By Prahlad Menon 4 min read

If you’ve ever watched a small business miss a call — or pay a receptionist SaaS a monthly fee per seat just to have someone (or something) answer the phone — you already understand the gap Tel-Agent is filling.

Well-funded closed products already do “AI receptionist for business.” What didn’t really exist until now is a good open, self-hosted one — where you own the recordings, choose the models, and decide which callers ever reach the AI at all.

Tel-Agent is that: an open-source gateway (AGPL v3, ~500 GitHub stars and climbing, maintained by Dpro GmbH in Vienna) that sits between a phone line and an AI agent.

How It Works

A call arrives over SIP. Tel-Agent checks the caller against your routing rules and does one of three things:

  1. Passes it through to a human
  2. Blocks it
  3. Hands it to an AI agent that speaks with the caller in real time

When the AI takes the call, it can invoke tools mid-conversation:

  • Transfer the call to a person
  • Take a message / voicemail
  • Check a calendar
  • Call any HTTP endpoint (your CRM, booking system, internal API)

Every call is recorded, transcribed, and searchable. Crucially, it all runs on your own hardware, on your own LAN, with your own API keys.

The “Bring Your Own Everything” Model

This is the part that makes Tel-Agent interesting for anyone tired of vendor lock-in:

  • Bring your own model — OpenAI, Claude, Gemini, Mistral, Ollama, Perplexity, DeepSeek, Grok, Qwen, Groq, OpenRouter and more, connected over MCP
  • Bring your own keys — no shared application, no middleman on your usage
  • Bring your own channels — each platform (WhatsApp, Telegram, Slack…) uses your credentials, not a pooled account
  • Own your data — recordings and transcripts stay on your box
❌ Closed SaaS: their models, their servers, your recordings on their disk, per-seat billing
✅ Tel-Agent:   your models, your hardware, your recordings, your keys

One Agent, Every Channel

The same agent that answers the phone also handles web chat, SMS, email, WhatsApp, Telegram, Messenger, Instagram, Discord, and Slack. The official scope now spans 24 channels including the phone, with more built on an extension contract.

This matters because a customer doesn’t care which door they walked through — they want the same context and the same answers whether they called, texted, or DM’d. A single agent across channels is how you get that consistency without wiring nine separate bots.

What It Is — and Isn’t

Being honest about scope is part of why the project is credible:

  • Not a PBX replacement — it connects to your existing PBX (3CX, Asterisk, FreePBX) as an extension.
  • Not a workflow automation platform — webhooks and a generic HTTP tool reach n8n and Home Assistant, which do that job better.
  • Not a CRM — it acts on your CRM through the HTTP tool.
  • Not analog-capable — bridge an analog line with an ATA; Tel-Agent only speaks SIP.

The mental model the maintainers use is clean: a channel is where the conversation happens; an integration is a system the agent acts on. Tel-Agent owns the first and reaches the second through the HTTP tool.

Project Status

It’s alpha, but usable alpha. Native installers and container images ship today — a Windows installer, macOS packages for Apple Silicon and Intel, DEB/RPM for Linux, and a docker-compose.release.yml. Conversations are answered end to end: replies stream token by token, can be interrupted mid-sentence, are stored, and raise a notification. The phone path includes streaming speech recognition and synthesis, interruption handling, caller-ID routing, and stored transcripts via SIP/LiveKit. It’s GDPR/DSGVO and EU AI Act compliant, and speaks English, German, Arabic, Spanish, and Dutch.

Why This Matters

Voice AI receptionists are one of the clearest near-term wins for small businesses — a missed call is often a lost customer. But the closed SaaS versions ask you to hand over your call recordings and lock you into their models and pricing.

Tel-Agent is the self-hosted answer: connect any phone line to any AI model, keep your data, and route calls exactly the way you want — human, blocked, or AI — before the agent ever says hello.

If you run a business phone line and you’re comfortable with a bit of SIP setup, it’s worth a weekend.

Repo: github.com/Dpro-at/Tel-Agent · Site: tel-agent.com