Dograh is a self-hostable, open-source voice AI platform under the BSD 2-Clause licence that lets teams build production voice agents with a visual workflow builder while keeping every line of source code, model choice, and call data on infrastructure they control.

What it is

Dograh is an open-source voice AI platform written in Python, published as a self-hosted alternative to the proprietary voice-agent products Vapi and Retell. It runs as a full stack deployed with Docker, driven by a browser-based visual workflow builder for designing inbound and outbound voice agents. The project lives in the conversational AI and telephony ecosystem: it integrates with speech-to-speech pipelines, or with separate LLM, speech-to-text, and text-to-speech providers, and it ships an MCP server so coding agents such as Claude Code and Codex can design and edit agents directly inside a Dograh workspace. It also supports telephony providers including Twilio, Vonage, and Telnyx, alongside an Asterisk ARI topic in its repository.

The concrete problem it solves is vendor lock-in and the closed nature of hosted voice-agent services. Teams evaluating Vapi or Retell get per-minute SaaS pricing, proprietary code, and call data resident in someone else's cloud, with customization limited to what the vendor's integrations expose. Dograh replaces that arrangement with a platform the operator hosts, where the LLM, TTS, and STT layers are interchangeable and the source itself is modifiable. Data residency is a property of the deployment rather than a contractual promise, and there is no per-minute tariff attached to the self-hosted path.

Key capabilities

• Visual workflow builder for agent design, starting from an Inbound or Outbound bot template and a short use-case description.
• BYOK across two architectures: end-to-end speech-to-speech, or a composable LLM / STT / TTS stack, with any provider or Dograh's bundled stack.
• Auto-generated API keys and a built-in LLM, TTS, and STT stack, so no external API keys are required to get a first agent running.
• In-browser testing through Test Audio for live voice conversations and Test Chat for text iteration, including editing or replaying user turns so the agent regenerates replies and node transitions from that point.
• MCP server integration, letting Claude Code and Codex work inside the workspace; installable via /plugin marketplace add dograh-hq/dograh-plugins and /plugin install dograh@dograh.
• Telephony support for outbound and inbound calling through providers such as Twilio, Vonage, and Telnyx, with Asterisk ARI listed among the project's topics.
• Deployment as a Docker Compose stack, downloadable as docker-compose.yaml with a start_docker.sh launcher script.

Who uses it and how

• Teams that must keep call recordings, transcripts, and prompts inside their own infrastructure for data residency or compliance reasons, deploying the Compose stack rather than consuming a hosted API.
• Startups and small product teams qualifying leads or screening submissions with an inbound or outbound agent, for example a bot named "Lead Qualification" that screens insurance form submissions for purchase intent.
• Engineering organisations that want source-level control, modifying the Python codebase and swapping LLM, STT, or TTS providers without waiting on a vendor roadmap.
• Developers using AI coding assistants, who install the Dograh setup skill into Claude Code or Codex and let the agent detect the OS, choose the deploy path, run the setup scripts, and verify the result.
• Operators running on-premise voice agents where the on-prem constraint rules out SaaS-only platforms entirely.

Getting started

Download docker-compose.yaml from the Dograh repository, fetch start_docker.sh, make it executable, and run it; then open http://localhost:3010 to create the first voice assistant. A hosted option is also available at https://app.dograh.com, and remote-server deployment is documented at docs.dograh.com.

How it compares

Against Vapi and Retell, both proprietary and SaaS-only, Dograh is BSD 2-Clause licensed and self-hostable, so the code is modifiable at source level rather than configurable within fixed integrations. Pricing differs in kind: Dograh self-hosted is free with a usage-based cloud option, while Vapi and Retell charge per minute. Data residency and vendor lock-in follow from that: on Dograh the operator's infrastructure holds the data and there is no lock-in, whereas on both competitors the data sits in their cloud.

When to use it — and when not to

A self-hoster must run and maintain the Docker Compose stack, including the bundled model stack, and should expect a two-to-three minute first startup while images download. Telemetry is collected by default and must be disabled explicitly by setting ENABLE_TELEMETRY=false before running the startup script, which is a real consideration for privacy-sensitive deployments. Teams that want a fully managed service with no operational surface, or that cannot run Docker at all, should look elsewhere.

curl -o docker-compose.yaml https://raw.githubusercontent.com/dograh-hq/dograh/main/docker-compose.yaml && curl -o start_docker.sh https://raw.githubusercontent.com/dograh-hq/dograh/main/scripts/start_docker.sh && chmod +x start_docker.sh && ./start_docker.sh