Dograh

Dograh is open source infrastructure for building and running AI voice agents. It provides a workflow builder for voice assistants in which each stage of the pipeline can be swapped: the inbound channel, speech to text, the LLM, text to speech and telephony. Alternatively the whole pipeline can switch to a speech to speech model such as Gemini Live or GPT Realtime for audio in, audio out conversations with natural turn taking and interruption handling across more than 70 languages. It targets teams building phone and voice agents, particularly in regulated sectors such as fintech, healthtech, insurance, legal intake and government. Dograh can be deployed on premises or inside a customer's own cloud VPC, with models served within the perimeter or fully air gapped so calls, recordings and transcripts never leave. Open weight models such as Whisper, Voxtral and Kokoro can be self hosted; STT and TTS swapping is production ready and LLM swapping is in beta. A Model Context Protocol server lets Claude Code, Cursor, Codex and other agents create, modify and deploy voice agents from the IDE. Agents can also mix pre recorded human clips with TTS in the same cloned voice. Pricing is published. Pay as you go carries a platform fee of 1 cent per minute, with BYOK or Dograh Models billed by usage, 10 concurrent calls included and self serve credits from $5. Custom volume offers discounted committed minutes and priority support. Enterprise, covering on premises deployment, SSO, audit logs and custom SLAs, is custom priced. The open source edition is free to self host under a BSD license.