Dograh is an open-source voice AI platform for building production phone agents — a self-hostable alternative to hosted services such as Vapi and Retell. You design call flows in a visual builder, connect your own speech and language providers, and deploy agents that answer or place real calls. The project has about 5,900 stars on GitHub, is BSD 2-Clause licensed, and was actively developed in 2026.
Dograh is built on Pipecat and supports both cascaded pipelines (speech-to-text, LLM, text-to-speech) and direct speech-to-speech models, with telephony handled through integrations such as Asterisk ARI. Everything runs bring-your-own-key: you plug in your preferred LLM, STT and TTS providers and keep full control of cost and data, on-premises if you like. It is also MCP-native, so AI coding assistants can help design and edit call workflows. A hosted cloud at app.dograh.com is offered for teams that do not want to run the stack themselves; the project was a Product Hunt #1 Product of the Day.
Dograh positions itself on transparency and economics: because the platform is open source and provider-agnostic, your per-minute cost is simply what your chosen speech-to-text, LLM and text-to-speech providers charge, rather than a marked-up platform rate. For teams evaluating voice agents, the one-command self-host setup makes it possible to place a real test call within an afternoon.
Freemium. The platform is open source and free to self-host; Dograh also offers a hosted cloud version for teams that prefer a managed service, and you pay your own telephony and model providers either way.
Businesses and developers building AI calling agents — support lines, reminders, lead qualification — who want Vapi-class capabilities without giving up self-hosting and provider choice.