Dograh

Dograh is a self-hostable voice agent platform built for teams that can't hand call recordings, transcripts, or customer PII to a third-party vendor. It's the open-source alternative to Vapi and Retell, licensed under BSD-2-Clause, and designed to run entirely within your own infrastructure perimeter.

The core idea is a modular pipeline you assemble yourself. Pick an inbound channel, a speech-to-text provider, an LLM, and a text-to-speech engine. Swap any node for a different provider, or skip the cascade entirely and run a speech-to-speech model like Gemini Flash Live or GPT Realtime for single-hop, audio-in/audio-out conversations. Lower latency, real interruption handling, no text round-trip.

Key capabilities:

• Bring your own models across STT (Whisper, Voxtral, Deepgram, AssemblyAI), LLM (OpenAI, Groq, Gemini, OpenRouter, Azure, AWS), and TTS (ElevenLabs, Cartesia, Kokoro, Chatterbox, your own voice clone weights)
• Telephony integrations with Twilio, Telnyx, Plivo, Vonage, Vobiz, Asterisk, and Cloudonix
• MCP server so Claude Code, Cursor, or any MCP-compatible agent runtime can spin up and configure voice agents without leaving the IDE
• Hybrid voice mode that mixes pre-recorded human clips with TTS in the same cloned voice, cutting costs up to 3x and reducing perceived latency
• On-prem or VPC deployment with fully air-gapped model serving, so no data crosses your boundary
• Observability via Langfuse integration

For regulated industries like healthtech, fintech, insurance, and government, the pitch is direct: your existing HIPAA, GDPR, and SOC 2 controls keep covering the data because Dograh never sees it. The code is auditable, so your security team can review every line before a call goes out. No vendor compliance review needed.

Compared to LiveKit or hosted platforms, Dograh's differentiator is the combination of a visual workflow builder, MCP-driven agent creation, and genuine data sovereignty, all in one stack you can fork and run yourself. Three deployment modes cover the range: self-host on your own servers, use their managed cloud, or have them operate the stack inside your VPC.