Self-hosted open-source voice AI removes the $14,400-$20,160/year per-minute billing trap that kills most voice agent experiments before ROI runs. One Docker command deploys everything. Runs on a $5 VPS at $0 ongoing cost.
What is Dograh and why is it gaining traction this fast?
Dograh is a self-hosted, open-source voice AI platform that replaces paid services like Vapi and Retell. It hit 3.6k GitHub stars with a trending badge active for 48 hours as of May 29, 2026. Version 1.32 shipped the same day, adding Pipecat 1.3.0 support, transcript URLs, and multi-architecture speed improvements. The entire stack deploys in a single Docker command and runs indefinitely on a $5 VPS. The voice layer tax that kills most agent projects in month two disappears the moment operators own the stack instead of renting it.
How does Dograh compare to Vapi and Retell on real operator costs?
Vapi charges $0.05 per minute and Retell charges $0.07 per minute. A 10-person team running routine calls burns $14,400 to $20,160 per year on the voice layer alone, before any other stack costs. Dograh’s self-hosted architecture brings that number to the cost of the VPS already running. The BSD-2 license places zero restrictions on usage and eliminates vendor lock-in risk entirely. Community traction confirms the math: an IndieHackers post on the cost savings crossed 1M impressions, Reddit cost breakdowns hit 10k+ views, and the HN Show thread drew significant developer response. The per-minute billing model Vapi and Retell run is structurally designed to compound against operators who scale, and Dograh’s self-hosted architecture makes that compounding irrelevant from day one.
Should small business owners running voice automation care about this signal?
Any operator currently running or planning voice automation for lead qualification, outbound sales, customer support, or appointment reminders should evaluate this stack before signing a per-minute contract. The platform handles drag-drop workflow nodes for QA scoring, CRM fetch, human handoff, and retry logic. Speech-to-speech mode cuts inference cost by a reported 3x. Pre-recorded plus TTS fallback improves conversions on repeated script lines. All customer data stays on the operator’s own box. Track the latest underdog signals that survive the Hype Check filter at The AI Profit Wire signal feed. If a voice automation budget exists or is being planned, self-hosting on Dograh before signing a per-minute contract is the single decision that determines whether the experiment survives month two.
Exact operator execution steps
1. Run: curl -o docker-compose.yaml https://raw.githubusercontent.com/dograh-hq/dograh/main/docker-compose.yaml
2. Run: ENABLE_TELEMETRY=false docker compose up –pull always
3. Open localhost:3010. Name the bot and describe the use case in plain language.
4. Hit Web Call. Test live with no API keys on the first run.
5. After first call clears: add your own LLM, STT, and TTS keys under settings.
6. Connect Twilio: add the webhook node. Set inbound or outbound. Add human transfer button.
7. Build the first workflow using drag-drop nodes. QA scoring and CRM fetch are available out of the box.
The deploy takes 90 seconds and the first live call confirms whether the use case works before a dollar of per-minute billing enters the picture.
What is the move on Dograh for operators running voice automation?
Operators who need voice automation have one decision to make before the next contract renewal: own the infrastructure or rent it. Dograh removes the technical barrier that previously made self-hosting impractical. The Docker deploy is one command, the visual builder removes the need to write agent logic from scratch, and the BSD-2 license means the stack is permanently free at the infrastructure level. The cost math is not close. Deploy Dograh on an existing VPS tonight, run one lead qualification flow before the next invoice arrives, and compare the line item at month end against what Vapi or Retell would have charged for the same call volume.
Source: GitHub (dograh-hq/dograh)
