Dograh is an open-source, self-hostable voice agent platform, officially positioned as an OSS alternative to Vapi & Retell.
Purpose and positioning It is aimed at teams that need self-hosted voice agents, especially in regulated industries and jurisdictions with strict requirements for data residency, compliance, and vendor review. The official emphasis is that it can run on users' own servers, within their own cloud VPC, or even in a fully air-gapped environment, so that calls, recordings, transcripts, and model inference do not leave the user's boundary.
Main capabilities
- Open-source workflow builder: used to build voice agents. You can click any node to view the providers that can be connected, and select the inbound channel, STT, LLM, TTS, and telephony system; you can also skip the cascaded pipeline and switch the entire pipeline to speech-to-speech.
- Bring your own models: you can connect models such as Voxtral, Kokoro, and Whisper that can run within your own VPC. The official claim is that "any model you can host" can be connected. The official statement says STT and TTS are production-ready, and LLM replacement is in beta.
- MCP-driven: Dograh provides a Model Context Protocol server that can point Claude Code, OpenClaw, Cursor, Codex, or any agent runtime to it, so that you can create, modify, and deploy a complete voice agent without leaving the IDE.
- Hybrid voice approach: mix human voice clips with TTS in the same cloned voice. The LLM uses the pre-recorded line when it is appropriate and falls back to TTS when it is not. The official claim is that this can bring unprecedented latency improvements, reduce costs by up to 3x, and "sound more human."
- Real-time audio-to-audio: audio input, audio output, without the transcription cascade and without a text round trip. The official claim is that this can bring real turn-taking, real interruption handling, and ultra-low latency. You can use Gemini 3.1 Flash Live (or Vertex AI) or GPT Realtime 2.
- Data residency and compliance: data stays within the user's boundary, the user's existing authentication and controls continue to apply, and there is no need to audit third-party vendors. The official claim is that no HIPAA review, GDPR signing, or SOC 2 is needed for Dograh, because Dograh cannot see user data; the code is open, and security teams can audit it line by line before the first call is made.
Deployment and usage methods The official offering provides three modes:
- Run it yourself: clone the repository and run docker compose up. Permanently free, BSD 2-Clause license, using your own server and data.
- Managed cloud: Dograh runs the same stack, with zero operations; the official team handles hosting, patching, and scaling, while users build and publish agents.
- In-VPC deployment: Dograh deploys the entire stack within the user's cloud, data stays local, and the official team handles orchestration, upgrades, and reliability.
Applicable scenarios Fintech, healthtech, defense, EU/GDPR, and other scenarios with high requirements for data sovereignty and compliance; teams that need regional language support, local deployment, and an auditable open-source solution. The official statement says self-hosting is one of the most requested features in the repository, and also the reason many teams cannot use managed voice agent vendors at all.
Limitations and additional notes The official statement clearly says LLM swapping is in beta. In addition, this material does not provide specific pricing, billing plans, SLA metrics, a complete list of supported telephony systems, the number of languages, or performance benchmark data; this information is not currently available. review, GDPR signing, or SOC 2, because Dograh cannot see user data; the code is open, and security teams can audit it line by line before the first call is made.
Deployment and usage methods The official offering provides three modes:
- Run it yourself: clone the repository and run docker compose up. Permanently free, BSD 2-Clause license, using your own server and data.
- Managed cloud: Dograh runs the same stack, with zero operations; the official team handles hosting, patching, and scaling, while users build and publish agents.
- In-VPC deployment: Dograh deploys the entire stack within the user's cloud, data stays local, and the official team handles orchestration, upgrades, and reliability.
Applicable scenarios Fintech, healthtech, defense, EU/GDPR, and other scenarios with high requirements for data sovereignty and compliance; teams that need regional language support, local deployment, and an auditable open-source solution. The official statement says self-hosting is one of the most requested features in the repository, and also the reason many teams cannot use managed voice agent vendors at all.
Limitations and additional notes The official statement clearly says LLM swapping is in beta. In addition, this material does not provide specific pricing, billing plans, SLA metrics, a complete list of supported telephony systems, the number of languages, or performance benchmark data; this information is not currently available.
