Dograh: Open-Source Visual Voice AI Builder

Dograh's Architecture

Dograh combines a voice engine, visual workflow builder, and platform layer. The voice engine connects telephony providers, speech-to-text (STT), large language models (LLMs), and text-to-speech (TTS). Developers map agent logic visually: add prompt nodes, qualification steps, API tool calls, branches, and transfers without custom orchestration code. The platform provides testing, tracing, recordings, analytics, and state inspection. Users bring their own providers, LLMs, and TTS since Dograh is open-source.

Local Setup and Agent Demo

Developers clone the GitHub repo, navigate to the folder, and run docker compose up to spin up containers. Access the UI to build a lead qualification agent: prompt node asks what the caller wants to build, qualification step queries company size and budget, API tool call creates or updates a CRM lead, branch checks qualification, and transfer node routes qualified leads to a human.

Test calls show full transcripts, traces of state changes and tool calls, and audio recordings. For example, a simulated call from 'Sarah from Inbound Calls' responds to questions about AI phone agents for inbound demo requests, company details, and 20,000 minutes usage.

Positioning Against Alternatives

Hosted platforms like Vapi, Bland, and Retell offer fast dashboards, APIs, transcripts, and testing but lock users into pricing changes, limits, and infrastructure. Raw frameworks like Pipecat, Vocode, and LiveKit provide flexibility and control without UI workflow editors, requiring glue code for orchestration.

Dograh targets developers wanting visual flow design without sacrificing self-hosting, provider swaps, or runtime inspection. Write code only where needed, use the builder for flows, and inspect failures with evidence.