Dograh

Tool snapshot
Best fitChatbots · Audio and Voice · Automation
In one lineDograh builds voice-agent workflows using selectable speech, model and telephony components. It offers open-source self-hosting and managed routes; external provider usage and...

Dograh is an open-source platform for building voice agents. Its official overview presents a workflow builder with selectable call, speech and model components. Teams can operate the stack themselves, use Dograh's managed cloud or discuss a managed deployment inside their own cloud.

Choose the conversation pipeline

A cascade combines speech-to-text, a language model, text-to-speech and telephony. Dograh also describes a speech-to-speech route in which a compatible model receives and returns audio directly. These are different configurations, so compare the supported providers and operational requirements for the specific agent rather than assuming every listed option can be swapped without adjustment.

The page describes mixing prerecorded voice lines with generated speech, allowing the model to choose a prepared line when one fits. Review the wording and permissions for the voice material. Claims about conversion, latency and cost reduction are publisher claims, not measurements of a workflow we ran.

Define a narrow call purpose, the information the agent may collect and when it should hand control to a person. Confirm how a failed connection or ambiguous caller request affects the next step before giving it a broad business role.

Keep deployment and data flow aligned

Dograh describes self-hosted, managed-cloud and private-cloud routes. Self-hosting does not by itself prove that the whole workflow stays within one perimeter: a chosen external speech or model provider can change where content travels. Inspect the actual component configuration and connected accounts before making a local-only claim.

The homepage says self-hosted speech recognition and synthesis are production-ready, while language-model swapping is in beta. Preserve that distinction when planning an all-local setup.

Its MCP server lets compatible coding agents create, modify and deploy voice workflows. Limit those permissions to the intended environment and require review before changing a production calling agent. An available tool connection does not establish safe authority for every action.

Separate platform fees from operations

The pricing page lists a pay-as-you-go platform fee of one cent per minute, with bring-your-own-key or Dograh-model usage options and credit top-ups. That headline fee is not a complete estimate of telephony, model or hosting costs.

The open-source route is described as the full platform with no Dograh subscription limit. It still requires infrastructure and an owner for deployment, upgrades and monitoring. Managed private-cloud terms are custom. Review licensing, provider bills and current plan scope before committing call volume. No installation, phone call or hands-on voice-quality test was performed for this overview.