Dograh

Dograh is fully open-source under the BSD 2-Clause license and self-hostable via a single Docker command. Unlike Vapi or Retell, every line of code is auditable, you bring your own LLM/STT/TTS providers, and there is no platform markup layered on top of model costs. Your data never leaves your infrastructure unless you opt into the managed cloud.

Dograh supports any OpenAI-compatible LLM endpoint, plus integrations for providers like xAI Grok and MiniMax on the LLM side. For speech, it supports Google STT/TTS, ElevenLabs, Ultravox, and other major providers. You can also override base URLs for custom or local model endpoints.

Yes. Dograh's telephony layer natively supports call transfer to human agents mid-conversation. This is handled through workflow nodes connected to your telephony provider (Twilio, Vonage, Telnyx, etc.), allowing for escalation logic to be built directly into the visual workflow builder.

The QA Node is a built-in workflow component that runs automated quality analysis across the prompts in your voice agent workflow before production deployment. It evaluates prompt effectiveness and flags potential weaknesses, reducing the need for manual prompt audits during iteration cycles.

Dograh is MCP-native, meaning voice agents can be driven and configured directly from MCP-compatible clients such as Claude Code or Codex. Dograh also ships an official setup plugin for Claude Code that handles installation, configuration, and troubleshooting of the platform end-to-end.