Self-Hosted Voice AI: How to Deploy Dograh AI as a Vapi Alternative

Voice AI agents are becoming a standard interface for customer support, appointment scheduling, lead qualification, and internal operations — but most solutions (Vapi, Retell, Bland AI) run on proprietary cloud infrastructure. Your audio, transcripts, and customer data pass through third-party servers.

**Dograh AI** changes this. It is an open-source (BSD 2-Clause), self-hostable voice AI platform that lets you build production voice agents with a drag-and-drop workflow builder and bring your own LLM, speech-to-text, and text-to-speech providers. It went viral on Hacker News and GitHub this week as the self-hosted alternative the voice AI space needed.

What Dograh does

Dograh provides a complete voice agent pipeline:

• **Drag-and-drop workflow builder** — define conversation flows visually without writing code
• **Multi-provider support** — bring your own STT (Whisper, Deepgram, Azure), LLM (OpenAI, Anthropic, local models via Ollama), and TTS (ElevenLabs, Azure, OpenAI)
• **Phone integration** — inbound and outbound calling via Twilio, Plivo, or Telnyx
• **REST API** — programmatic agent creation and management
• **Analytics dashboard** — call logs, transcriptions, cost tracking, and quality metrics

The project is maintained by YC alumni and exit founders who committed to keeping voice AI open and self-hostable.

Dograh vs Vapi vs Retell

| Feature | Dograh | Vapi | Retell |

|---------|--------|------|--------|

| **License** | BSD 2-Clause (open source) | Proprietary | Proprietary |

| **Self-hostable** | ✅ One Docker command | ❌ SaaS only | ❌ SaaS only |

| **Pricing** | Free (self-host) | Per-minute SaaS | Per-minute SaaS |

| **BYOK providers** | Any LLM/STT/TTS | Limited integrations | Limited integrations |

| **Source access** | Full customisation | Closed | Closed |

| **Data residency** | Your infrastructure | Their cloud | Their cloud |

One-command Docker deployment

Dograh ships as a Docker Compose stack. Deploy it on any Linux server, VPS, or homelab with Docker installed:

```bash

git clone https://github.com/dograh-hq/dograh.git

cd dograh

Start with default configuration

docker compose up -d

```

This spins up the Dograh web UI, API server, workflow engine, and database. The first-launch setup wizard guides you through connecting your STT, LLM, and TTS providers.

For production deployments, use the included production compose file:

```bash

docker compose -f docker-compose.yaml -f docker-compose.local.yaml up -d

```

For understanding the Docker fundamentals behind this deployment, read Docker Setup for Local AI Tools.

Configure speech and language providers

Dograh lets you mix and match providers at each layer of the voice pipeline.

Speech-to-text (STT)

• **Whisper (local):** Use faster-whisper for completely offline transcription. Set the Whisper model path in the Dograh admin panel.
• **Deepgram:** Set your `DEEPGRAM_API_KEY` for Nova-2 quality
• **Azure Speech:** Enterprise-grade with custom vocabulary support

Language model (LLM)

• **OpenAI:** Set `OPENAI_API_KEY` for GPT-4o or GPT-4o-mini
• **Anthropic:** Set `ANTHROPIC_API_KEY` for Claude models
• **Local models:** Point Dograh at any OpenAI-compatible endpoint (Ollama, vLLM, LocalAI). See How to Run Llama 3 Locally with Ollama for Ollama setup.

Text-to-speech (TTS)

• **ElevenLabs:** Set `ELEVENLABS_API_KEY` for the highest quality voices
• **Azure TTS:** Neural voices with SSML support
• **OpenAI TTS:** Quick setup with `OPENAI_API_KEY`

Build your first voice agent

1. Open the Dograh web UI at `http://your-server:3000`

2. Click **Create Agent** and name it

3. Use the visual workflow builder to define the conversation flow:

• **Greeting node:** "Hello, you have reached [business name]. How can I help?"
• **Intent classification:** Branch based on keywords or LLM classification
• **Action nodes:** Look up data, transfer to human, schedule appointments
• **Fallback:** Handle unrecognised inputs gracefully

4. Connect your phone number (Twilio/Plivo) under **Settings → Phone**

5. Test via the built-in voice playground or call your number

Production hardening

For a production voice agent deployment:

• Place Dograh behind a reverse proxy (Caddy, Nginx, Traefik) with TLS
• Use environment-specific `.env` files for secrets
• Configure database backups for call logs and agent configurations
• Set up monitoring on the Docker containers — see Local AI Monitoring Stack for guidance
• Consider data retention policies for call recordings and transcriptions

For general security practices, see How to Secure a Self-Hosted AI Server.

Conclusion

Dograh AI fills a real gap in the voice AI ecosystem. Until now, building a voice agent meant either paying per-minute SaaS fees to Vapi or Retell (and sending all your audio data through their infrastructure) or stitching together a dozen open-source components yourself. Dograh provides a complete, production-ready platform that you can run on your own hardware with your own provider keys.

For teams that handle sensitive customer conversations, operate in regulated industries, or simply want control over their voice AI stack, Dograh is worth serious evaluation.

Source

• GitHub: dograh-hq/dograh
• Documentation: docs.dograh.com
• HN Discussion: Dograh — self-hosted open-source Vapi alternative