Dograh

Product Introduction

1. Definition: Dograh is an open-source, self-hostable voice agent infrastructure platform. Technically, it is a visual workflow builder and orchestration engine for creating, deploying, and monitoring AI-powered voice agents. It enables the construction of telephony and voice interaction systems by chaining together modular components for speech recognition (STT), large language models (LLM), and speech synthesis (TTS), or utilizing direct speech-to-speech (S2S) models.
2. Core Value Proposition: Dograh exists to provide complete data sovereignty and architectural freedom for voice AI applications. Its primary value is being a fully auditable, open-source alternative to closed, hosted platforms like Vapi and Retell, allowing enterprises to deploy voice agents on their own infrastructure without vendor lock-in, ensuring data never leaves their security perimeter.

Main Features

1. Visual Flow Builder & Modular Stack: Dograh provides a node-based visual interface where users can design conversation workflows. Each node in the cascade (Inbound, STT, LLM, TTS, Telephony) is swappable, allowing users to "bring your own model" (BYOM) across 30+ integrated providers like OpenAI, Gemini, Groq, ElevenLabs, and Twilio, or use local, on-premises models.
2. Speech-to-Speech (S2S) Pipeline: A key technical feature is the support for end-to-end audio models like Gemini 3.1 Flash Live. This bypasses the traditional STT→LLM→TTS cascade, enabling audio-in/audio-out processing. This reduces latency, enables more natural turn-taking and interruption handling, and supports over 70 languages natively.
3. Model Context Protocol (MCP) Server: Dograh ships with an integrated MCP server, allowing AI coding assistants like Claude Code, Cursor, or OpenClaw to directly interact with the platform. Developers can programmatically create, modify, and deploy voice agents through natural language commands in their IDE, streamlining development workflows.
4. Hybrid Voice Output (Pre-recorded + TTS): To improve perceived humanity and reduce cost/latency, Dograh's LLM can select from a library of pre-recorded human voice clips for common responses. It seamlessly falls back to TTS (in the same cloned voice) for dynamic content. This hybrid approach can reduce TTS costs by up to 3x while significantly improving audio quality.
5. Enterprise-Grade Deployment Options: The platform supports three deployment models: 1) Self-Hosted (OSS): Full BSD 2-Clause licensed stack deployable via docker compose. 2) Managed Cloud: Dograh-hosted, fully-managed service. 3) Private Managed Cloud: Dograh operates the complete stack within the customer's VPC, ensuring data residency while handling ops.

Problems Solved

1. Pain Point: Data Residency and Compliance Risk in Regulated Industries. Traditional hosted voice AI vendors require sending sensitive call audio, transcripts, and customer PII to third-party servers, creating compliance hurdles for HIPAA, GDPR, SOC 2, and financial regulations.
2. Target Audience: Engineering and product teams in Fintech, Healthtech/Telemedicine, Insurance, Banking, Legal, Defense, and Government sectors who require voice AI but cannot use hosted SaaS due to data sovereignty laws or internal security policies. Also targets developers seeking an open-source, customizable alternative to avoid vendor lock-in.
3. Use Cases: Building compliant AI calling agents for patient intake (HIPAA), EMI payment reminders and collections (PCI DSS), legal intake hotlines, customer support lines with full data control, and internal enterprise voice applications where all data must remain on-premises.

Unique Advantages

1. Differentiation: Unlike Vapi or Retell, which are closed-source, hosted services, Dograh is fully open-source and self-hostable. This eliminates the "vendor compliance review" process, as the customer's existing security certifications apply to their own infrastructure. It also offers deeper cost control and the ability to integrate any on-premises AI model.
2. Key Innovation: The combination of true open-source licensing (BSD 2-Clause) with a production-ready, modular voice agent stack and a built-in MCP server is unique. This allows for both complete infrastructure control and a modern, AI-native developer experience, bridging the gap between enterprise compliance needs and cutting-edge AI tooling.

Frequently Asked Questions (FAQ)

1. Is Dograh truly free and open-source? Yes, Dograh is licensed under the permissive BSD 2-Clause license. The entire codebase is available on GitHub, allowing for free self-hosting, modification, and distribution without any feature gating or mandatory fees.
2. How does Dograh handle data privacy for GDPR or HIPAA? By being self-hostable, Dograh ensures all voice call data, recordings, transcripts, and AI model inferences can be contained within your own data center or cloud VPC. Since no data is sent to Dograh's servers, you maintain full control, and your organization's existing compliance protocols cover the system.
3. Can I use local models like Llama or Whisper with Dograh? Absolutely. A core feature is the ability to swap in any STT, LLM, or TTS provider. You can configure Dograh to use locally hosted open-source models (e.g., Whisper for STT, Llama for LLM, Kokoro for TTS), enabling a fully air-gapped, on-premises voice AI deployment.
4. What is the advantage of the Speech-to-Speech (S2S) mode? S2S mode using models like Gemini Live reduces latency by eliminating the sequential processing of traditional cascades. It leads to more fluid, human-like conversations with realistic interruption handling and is inherently multilingual, improving user experience in real-time voice interactions.
5. How does the MCP integration benefit developers? The integrated Model Context Protocol server allows AI-powered coding tools to directly manipulate your Dograh deployment. Developers can use natural language (e.g., in Claude Code) to command the creation of new agents, modification of workflows, or connection of CRM webhooks, dramatically speeding up development and iteration cycles.