Dograh is a voice AI platform for building phone agents that handle inbound and outbound calls. You design an agent in a visual workflow builder, test it in the browser by voice or in a text chat, and connect it to a telephony provider to take real calls. It runs self-hosted through Docker or as a managed cloud service.
The workflow builder composes a call from start nodes, agent nodes, global instructions, tools, transitions, and end-call outcomes. The Test Chat panel lets you edit or replay a user turn, after which the agent's replies and node transitions are regenerated from that point. A default LLM, text-to-speech, and speech-to-text stack ships with auto-generated keys, and you can swap in your own providers for any of those layers, for speech-to-speech models, or for telephony.
An MCP server exposes the workspace to coding agents such as Codex, Claude Code, and Cursor, so an assistant can inspect existing agents, search the documentation, fetch node schemas, create workflows, and save draft edits from a natural-language brief. A separate setup skill for Claude Code and Codex handles installation and configuration.
Features
- Visual workflow builder: start, agent, tool, transition, and end-call nodes with global instructions, plus QA nodes, knowledge bases, webhooks, and embeds
- Browser testing: Test Audio for talking to the agent by voice and Test Chat for iterating on prompts in text, with turn editing and replay
- Telephony integrations: Twilio, Vonage, Telnyx, Plivo, Vobiz, Cloudonix, and Asterisk ARI, with call transfer for human handoff on supported providers
- Bring your own providers: LLM, TTS, STT, speech-to-speech, and telephony are all replaceable; artifacts go to bundled MinIO or S3-compatible storage
- MCP server: coding agents create and edit workflows, fetch node schemas, and search docs inside the workspace
- SDKs: Python and Node packages for programmatic agent creation and outbound calls
- Deployment: one-command Docker self-hosting, remote server deployment with HTTPS, or the managed cloud
- Telemetry opt-out: anonymous usage collection can be disabled with an environment variable before startup
