Dograh vs Retell AI: Run Your Own Voice Platform or Rent One

Dograh is an open-source (BSD-2-Clause) voice agent platform that anyone can self-host for free, with no per-minute platform fee. Retell AI is a closed-source hosted platform that charges $0.055 a minute for voice infrastructure, plus separate per-minute charges for models and phone. Retell offers on-prem only through Enterprise sales, with no published price.

Key Takeaways

• Retell's $0.055 a minute comes before the model and phone charges.
• Dograh is open source and free to self-host. Retell's on-prem is an Enterprise-only deal.
• Retell's AI QA (quality assurance) costs $0.10 a minute after 100 free minutes.

This post is part of our guide to Retell AI alternatives. The guide surveys hosted and open-source options. Here we put just two of them on the same rows, for teams already choosing between Dograh and Retell.

Dograh vs Retell AI: the key differences

The main split is who can read and run the platform code. Dograh's code is open, and Retell's is closed. The table puts both products on the same rows, at regular list prices.

Read down the rows and they split into two groups. Open code, self-hosting, the platform fee, carrier choice and per-step QA favour Dograh. Retell's published certifications and its free BAA favour Retell.

On the Retell AI HIPAA BAA question, Retell signs a BAA at no extra fee and requires one before patient data passes through it. Dograh publishes no BAA.

Where the platform runs and where call data goes

With Dograh you can run the whole platform on servers you control. With Retell, calls run on Retell's cloud unless you sign an Enterprise deal.

If you searched for a Retell AI on-premise deployment, Retell does offer one: its security overview lists cloud, VPC and on-premises options, air-gapped included. It is sold through Enterprise sales, with no published price, hardware requirements or install package, and the code stays closed.

On Retell AI data retention, Retell keeps call data indefinitely by default until you set a retention period for each agent. It handles GDPR through AWS, and it does not currently operate services within the European Union.

Dograh, an open-source, self-hostable voice agent platform, is released under the BSD-2-Clause licence. The Dograh source code on GitHub has more than 5,800 stars, and anyone can self-host it for free with Docker Compose. Because the code is open, you can read it before you trust it with calls, and change it when you need to.

Dograh is the orchestration layer: it runs the call flow, tools and telephony, and the speech and language models are a separate layer you choose. Running the platform needs an ordinary server with at least 8 GB of RAM and 4 vCPUs, and no GPU when you use model providers through API keys.

Call data stays on your servers only when you self-host Dograh and host the models too. With your own API keys, you hold the contract, but the audio still reaches that provider. The phone carrier carries the call either way. Our post on the open-source alternative to Retell AI covers what changes when you own the platform.

Open Source Alternative to Vapi / Retell

Self-hosted voice agent platform — no per-minute fees

dograh-hq/dograh

Star on GitHub

How Dograh and Retell AI are priced

Both products bill the AI models and the phone line apart from the platform, so compare the platform line first.

Retell's pricing page lists "Retell Voice Infra" at $0.055 a minute. The model and the voice are billed per minute on top, then the phone line, and Retell's own range for the total is $0.07 to $0.31 a minute. Retell lists no separate speech-to-text charge, so its $0.055 appears to cover speech-to-text as well. If it does, the gap with a pure platform fee is smaller than it looks.

Paid add-ons sit on top of that, such as $0.005 a minute for a knowledge base and $0.01 for removing personal details.

At 1,000 minutes a day, Retell's voice infra line alone comes to $20,075 a year ($0.055 x 365,000 minutes), before any model or phone charges. That is our arithmetic from the list price.

Self-hosted Dograh has no per-minute platform fee. You still pay for your servers and for any paid model and phone accounts. At the same volume, Dograh Cloud's platform fee comes to $3,650 a year, before models, by the same arithmetic.

We don't put one all-in price per minute on either product, because the total moves with the models you pick. Our breakdown of Retell AI pricing works through full bills, line by line.

Models, phone lines and call transfer

Retell bills the models it runs for you, while Dograh lets you pay providers directly or host the models yourself.

Retell offers text models from OpenAI, Anthropic and Google, priced per minute, and Retell bills the model itself. Its custom LLM option connects your own model server over a WebSocket, while Retell still runs the phone call and the speech on both sides. Retell also sells speech-to-speech, using OpenAI's realtime models only, at $0.07 to $0.38 a minute.

Dograh gives you three ways to fill the model layer: Dograh's own models, your own key for any supported provider, or open-weight models you host. With your own LLM keys, you can set a backup LLM that takes over if the main one errors or goes quiet. Speech-to-speech runs out of the box through Realtime mode, for example with Gemini 3.1 Flash Live. A hand-off to another agent and the spoken transfer introduction need a cascade agent.

For Retell AI custom telephony, Retell sells US and Canada numbers only, at $2 a month ($5 for a US toll-free number), and accepts your own SIP trunk at no charge, from carriers such as Twilio and Telnyx or contact-centre systems such as Genesys. Dograh sells no numbers or minutes, so your carrier contract stays yours. It works with Twilio, Vonage, Plivo, Telnyx, Cloudonix, Exotel, Vobiz and Asterisk ARI, among the telephony providers Dograh supports.

For Retell AI call transfer, you get cold, warm and AI-screened warm transfer to a phone number or SIP address, on phone calls only. Dograh transfers a call to a person on Twilio only, with an optional spoken introduction that is off by default. Test every Dograh transfer over a real phone call, because a browser test call cannot transfer.

Building agents and checking calls

Both products have a visual flow builder, so the differences are in how a step picks its next move and how calls get graded.

Retell offers single-prompt agents and Conversation Flow, a canvas of nodes joined by edges with transition conditions. Its logic nodes can branch on a fixed rule. In Dograh, the call is a canvas of steps joined by pathways, each with a written condition, and the AI decides which pathway to take. There is no fixed "if amount > 100" branch. Through Dograh's MCP server, an AI assistant such as Claude Code or Cursor can also create and edit agents for you.

For Retell AI quality assurance, Retell's AI QA scores a sample of calls for hallucination, knowledge base accuracy, latency, sentiment and tool use. It is free for the first 100 minutes of analysed call time, then $0.10 a minute.

Dograh's QA runs from a QA node on editor test calls and on live calls. Each step of the call gets its own 1-10 score with its own tags and sentiment, so you can see which step went wrong. Dograh's QA runs on the AI model you choose and is billed as normal model usage, with no separate per-minute QA fee. You can analyze only a sample of calls to keep costs down. In production, we suggest QA on a random 5-10% of calls.

Our post on Automated QA for Voice Agents covers QA setup in depth.

Which one fits your team

Each product suits a different kind of team, so match it to how you want to run your calls.

Retell fits teams that want no servers at all and need the vendor to hold the certifications. Retell reports SOC 2 Type 1 and Type 2 certification plus HIPAA and GDPR compliance, and it signs a BAA at no extra fee. It also fits teams that run fewer than 20 calls at once and want one bill that covers the models and the phone line. Flows that depend on fixed rule-based branches or an AI-screened warm transfer also suit Retell, since Dograh has no direct match for either.

Dograh fits teams that want to own the code and decide where their calls run. You can read the code before you trust it with calls, self-host it with no per-minute platform fee, bring your own model keys or host open-weight models, and see a QA score for each step of a call. Self-hosted, you control the servers and choose when to upgrade. Dograh Cloud removes the servers, and Dograh Enterprise adds on-prem deployment and support with SLAs if you need it. If Vapi is on your shortlist as well, Vapi vs Retell AI compares those two hosted platforms.

If owning the code and the call data path matters to you, test the open-source route first. Rebuild one Retell flow on it and run that flow over a real phone call before you move a number. Our step-by-step Retell migration guide walks through the rebuild.