Dograh
Dograh is an open-source voice-agent infrastructure platform that lets teams build, deploy, and run production-grade voice agents on their own infrastructure. It supports both speech-to-speech and STT→LLM→TTS cascade pipelines, with pluggable models for inbound channels, speech recognition, language models, and text-to-speech. The platform emphasizes data sovereignty, offering self-hosted, managed cloud, and VPC deployment options with enterprise-grade compliance certifications.
- Artificial Intelligence
- AI Agents
- Developer Tools
- Software Only
Funding
Founders
Product
Problem
Building production-grade voice agents typically requires stitching together multiple hosted services for speech recognition, language models, and text-to-speech, creating latency, cost, and data-privacy concerns. Teams in regulated industries often cannot use hosted voice-agent vendors at all because their data would leave their perimeter, creating vendor compliance risk and breaking existing certifications.
Solution
Dograh provides an open-source voice infrastructure stack that teams can fork, customize, and deploy anywhere—on their own servers, in a managed cloud, or inside their VPC. The platform supports both a speech-to-speech pipeline (audio in, audio out with a single model) and a cascade pipeline (STT → LLM → TTS), letting users plug in their preferred models for each component. Dograh ships a Model Context Protocol (MCP) server so agent runtimes like Claude Code, Cursor, or OpenClaw can spin up and deploy full voice agents directly from the IDE. The platform also supports a hybrid voice approach that mixes pre-recorded human voice clips with TTS in the same cloned voice, reducing cost and improving conversational quality.
Target Audience
Primary customers are engineering teams and enterprises building voice agents for customer support, sales, or operations, particularly those in regulated industries requiring data residency and on-prem deployment. The platform also targets developers who want an open-source alternative to hosted voice-agent vendors like Vapi or Retell.
Features
- Speech-to-speech pipeline using models like Gemini 3.1 Flash Live or GPT Realtime 2, with real turn-taking, interruption handling, and 70+ language support
- Cascade pipeline with pluggable STT (Whisper, Voxtral, Canary Qwen) and TTS (Kokoro, Chatterbox, custom voice clones) models
- Hybrid voice mode that mixes pre-recorded human audio with TTS in the same cloned voice, cutting costs up to 3× and improving perceived naturalness
- Model Context Protocol (MCP) server for building and deploying voice agents from IDEs like Claude Code, OpenClaw, Cursor, or Codex
- Self-hosted deployment via Docker Compose with BSD 2-Clause license, plus managed cloud and VPC deployment options
- Enterprise-ready data residency with on-prem deployment, auditable open source, and compliance with HIPAA, GDPR, SOC 2, ISO 27001, and CCPA