Dograh — Open-Source Self-Hosted Voice AI Platform

Introduction

Dograh provides the infrastructure for building voice AI agents that handle phone calls, voice interactions, and real-time conversations. It runs on your own servers, giving you full control over voice data and eliminating per-minute cloud API costs.

What Dograh Does

• Processes inbound and outbound voice calls with AI agents
• Provides a visual no-code workflow builder for designing call flows
• Supports speech-to-speech and LLM-driven conversation pipelines
• Integrates with telephony systems via SIP and WebRTC
• Connects to local LLMs and TTS/STT models for on-premise deployment

Architecture Overview

Dograh is built in Python with a modular pipeline architecture. Voice input flows through configurable stages: speech-to-text, language model processing, and text-to-speech. The visual workflow builder generates pipeline configurations that the runtime engine executes. Telephony integration uses Asterisk ARI for SIP connectivity, while WebRTC handles browser-based voice interactions.

Self-Hosting & Configuration

• Deploy with Docker Compose for a complete stack setup
• Configure telephony providers through the web dashboard
• Connect your own STT, LLM, and TTS models or use integrated defaults
• Set up SIP trunks for production phone number integration
• Scale horizontally by adding worker nodes for concurrent call handling

Key Features

• Visual workflow builder for designing voice agent logic without code
• Native MCP integration for connecting AI agents to external tools
• Support for both cloud and local AI model backends
• Real-time speech-to-speech mode for low-latency conversations
• Telephony-ready with SIP, WebRTC, and PSTN connectivity

Comparison with Similar Tools

• Vapi — cloud-based voice AI API with per-minute billing; Dograh is self-hosted with no usage fees
• Retell AI — managed voice agent platform; Dograh provides equivalent features as open-source software
• LiveKit Agents — real-time communication focused on WebRTC; Dograh adds telephony and visual workflow design
• Pipecat — Python framework for voice pipelines; Dograh adds a no-code builder and telephony integration on top

FAQ

Q: Can Dograh handle production phone calls? A: Yes. It integrates with SIP trunks and supports inbound and outbound calling through standard telephony protocols.

Q: What AI models does it support? A: It works with any OpenAI-compatible API, local models via Ollama, and specialized STT/TTS engines like Whisper and Piper.

Q: How many concurrent calls can it handle? A: Concurrency depends on your hardware and model choices. Worker nodes can be scaled horizontally for higher throughput.

Q: Is it suitable for customer service automation? A: Yes. The workflow builder supports conditional logic, knowledge base lookups, and handoff to human agents.

Sources

• https://github.com/dograh-hq/dograh