Dograh
Open-source voice-agent workflow builder with visual orchestration and private deployment.
About Dograh
Dograh is an open-source workflow builder for voice agents with visual node orchestration: click a node to connect providers for inbound channels, STT, LLM, TTS, and telephony, or skip the cascade and switch the full pipeline to speech-to-speech. It supports local or private-cloud VPC deployment, with AI models fully air-gapped so call, recording, transcription, and inference data stays within your own boundary. Its built-in Model Context Protocol server lets agent runtimes such as Claude Code and Cursor create, modify, and deploy voice agents inside an IDE. Clone the repository and run docker compose up to start it; it is permanently free under the BSD 2-Clause license.
Best for
Best use cases
Build an inbound telephone voice agent
Deploy voice AI services in an enterprise intranet or private cloud
Let an agent create, modify, and deploy voice workflows automatically inside an IDE
Skip the cascaded pipeline and adopt a speech-to-speech solution directly
Compare before choosing
The official comparison page calls Dograh the open-source Vapi alternative (The Open Source Vapi Alternative), focused on forkable voice agents.
Features
Visual workflow orchestration
Click any node to connect a provider and select inbound channels, STT, LLM, TTS, and telephony as needed.
Speech-to-speech mode
Skip the cascaded structure and switch the full pipeline to speech-to-speech.
Private and air-gapped deployment
Deploy locally or in your own cloud VPC; calls, recordings, transcriptions, and model inference can stay fully within your own boundary.
MCP server
Connect runtimes such as Claude Code and Cursor to create, modify, and deploy complete voice agents inside an IDE.
Pros
- Open source (BSD 2-Clause) and permanently free when self-hosted; voice agents can be forked
- Supports local or VPC deployment; AI models can be fully air-gapped, keeping data within your own boundary
- Pluggable voice pipeline components (inbound channels, STT, LLM, TTS, telephony), plus speech-to-speech mode
- Built-in MCP server lets agents create and deploy voice agents inside an IDE
Cons
- The official documentation says LLM switching is still in beta.