Dograh

Dograh vs Vapi: Run Your Own Voice Agent Platform or Rent One

Dograh vs Vapi: Run Your Own Voice Agent Platform or Rent One
ComparisonOctober 13, 2026·10 min read

Dograh vs Vapi: Run Your Own Voice Agent Platform or Rent One

Pritesh Kumar
Pritesh Kumar·Co-founder, Dograh AI

Building Dograh, the open-source alternative to Vapi. OSS and Voice AI. Exit founder and YC alum.

Dograh is an open-source (BSD-2-Clause) voice agent platform. You can self-host it, on-prem included, or use Dograh Cloud at 1¢ a minute. Vapi is closed source and hosted only, with a $0.05 a minute hosting fee before models and phone lines. Pick Vapi if you want a hosted API with published certifications and uptime SLAs.

Key Takeaways

  • Self-hosted Dograh has no platform fee; Vapi charges $0.05 a minute.
  • Vapi does not support on-prem; Dograh can run inside your own network.
  • Dograh QA gives each step of a call its own score.

This post is part of our guide to Vapi alternatives. The guide ranks every option, and this post is for readers already down to two, Dograh and Vapi, compared row by row.

Dograh vs Vapi: the key differences

The biggest difference is who runs the platform and who can read its code. The table below puts the two on the same rows. All prices are regular list prices.

FeatureDograhVapi
Source codeOpen source, BSD-2-ClauseClosed source
Where it runsYour own servers (Docker Compose), or managed Dograh CloudVapi's cloud only, in the US or its EU region
On-premYes. Self-host the full platform on servers inside your own network, or get on-prem and private-cloud deployment with support on Enterprise. With open-weight models you host, call audio never goes to a model providerNo. Vapi does not support on-premise deployments; it runs only on Vapi's cloud
Platform feeSelf-hosted: none. Dograh Cloud: 1¢/min. Models and phone are billed separately$0.05/min hosting, prepaid, on every package. Models and phone are billed on top
Monthly packagesNone required. Custom volume pricing and Enterprise plans on requestCore $29/mo. Pro 10% of hosting, $999/mo minimum. Premier by contract
Concurrent callsSelf-hosted: 10 at once by default, adjustable with the DEFAULT_ORG_CONCURRENCY_LIMIT setting; your servers and phone provider set the practical ceiling. Cloud: 10 included4, 10 or 30 by package, then $10 per extra line a month
AI modelsDograh's models, any provider with your own key, or open-weight models you hostProviders billed through Vapi at cost, your own provider keys, or a custom LLM server
Phone numbersNot sold. You keep your numbers in your own carrier account1 free US number, inbound only (5 on Core, 10 on Pro). Imported numbers from other carriers
Carriers and SIPTwilio, Vonage, Plivo, Telnyx, Cloudonix, Exotel, Vobiz, Asterisk ARI. Ready-made SIP on Cloud; Asterisk ARI or your own Cloudonix account when self-hostedVapi's own numbers, or numbers imported from Twilio, Telnyx or Vonage. You can also bring your own SIP trunk
Call transferTo a person on Twilio today, tested over a real phone call. Can also hand the call to another Dograh agentBlind and warm transfer to a phone number or SIP address. Handoffs between assistants in a Squad
BuilderVisual canvas of steps joined by pathways, with a global prompt and per-step instructions. AI assistants such as Claude Code or Cursor can also build agents through its MCP serverAssistants and Squads in the dashboard or code, with a visual prompt editor and the Composer AI builder. Visual Workflows flow builder retired August 2026
QA after callsQA node on editor test calls and live calls. A separate score for each step, plus tags and sentimentScorecards, structured outputs and monitors, but scorecards grade the whole call rather than each step. Simulations and Evals before launch
Call tracesA View Trace button on each call opens its trace in Langfuse: speech-to-text, AI model calls, tool calls. Full source access when self-hostedCall logs with per-turn latency by stage, recordings and cost. No source access
Compliance claimsNo certifications or BAA published. Self-hosting keeps call data off Dograh's systems entirelySOC 2 Type II, HIPAA and GDPR claimed. HIPAA add-on $2,000/mo with a BAA
Vendor SLACustom SLAs on Enterprise. Community support for the open-source version99% uptime on Pro and 99.9% on Premier. No uptime SLA on lower packages

The rows split into two groups. On-prem, the platform fee, carrier choice, per-step QA and access to the source favour Dograh. Vapi's published compliance claims and written uptime terms favour Vapi.

On the Vapi HIPAA BAA question, Vapi signs a business associate agreement (BAA) as part of a $2,000 a month add-on, and HIPAA mode cannot be combined with zero data retention. Teams that self-host Dograh with local models keep the call audio on their own servers.

Open source and self-hosting: who holds the code and the call data

With Dograh you can run the whole platform where your data already lives. With Vapi, every call passes through Vapi's cloud.

If you searched for a Vapi on-premise deployment, there isn't one: Vapi does not support on-premise deployments. For Vapi data residency, the option is its EU region. HIPAA mode still stores call recordings and transcripts in Vapi's storage.

Dograh is released under the BSD 2-Clause licence. You can change the code and run it anywhere, as long as you keep the copyright notice. The Dograh source code on GitHub has more than 5,800 stars. Access to the source matters most later: when something breaks, you can open the code and see why.

Keep two choices apart. Running the platform yourself 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. Running the AI models yourself is a separate choice. Dograh is the orchestration layer that runs the call, and it is not a model. Most teams start with Dograh's models or their own provider keys. Teams with strict data rules host open-weight models themselves.

Only self-hosted models keep the audio entirely on your servers. With your own key, you hold the contract, but the audio still reaches that provider. Your phone carrier also carries the call either way.

Where one call runs: on Vapi it passes through Vapi's cloud, while self-hosted Dograh keeps the platform and open-weight models inside your own servers, with the phone carrier outside in both
If your data rules forbid outside providers, plan for open-weight models on your own servers before you compare platform fees. The phone carrier still carries the call either way.

Our guide to the open-source alternative to Vapi covers the real work of self-hosting in more detail.

Dograh

Open Source Alternative to Vapi / Retell

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

dograh-hq/dograh

Star on GitHub

What you pay: a per-minute platform fee, or none

Both products bill the AI models and the phone line separately, so the difference is the platform line.

Vapi charges a prepaid $0.05 a minute for hosting on every package. Optional plans, which Vapi calls Success Packages, sit on top: Core at $29 a month, and Pro at 10% of hosting with a $999 monthly minimum. HIPAA is a $2,000 a month add-on. At 1,000 minutes a day, that is about $18k a year in hosting fees alone, before models and phone ($0.05 x 365,000 minutes = $18,250). Customers who joined before Vapi's mid-September 2026 plan change may still be on older terms.

The open-source Dograh platform has no per-minute platform fee. You still pay for your servers, plus any paid model and phone accounts. If you would rather not run servers, Dograh Cloud charges a 1¢ a minute platform fee, with 10 concurrent calls included. At the same 1,000 minutes a day, that is $3,650 a year in platform fees, by our arithmetic.

We don't give one all-in price per minute for either product, because the total swings with the models you pick. Vapi Pricing Explained works through full bills for three setups, line by line.

Phone numbers, SIP trunks and call transfer

Dograh never sells you phone numbers or minutes, so the phone line stays your contract and moves with you.

Dograh works with Twilio, Vonage, Plivo, Telnyx, Cloudonix, Exotel, Vobiz and Asterisk ARI, among the telephony providers Dograh supports. To bring your own SIP trunk to a voice agent, Dograh Cloud gives you a ready-made SIP connection. Self-hosted setups connect through Asterisk ARI or their own Cloudonix account.

Call transfer to a person works on Twilio today. Test every transfer over a real phone call, because a browser test call cannot transfer.

Vapi is quicker on day one. It gives you a free US number, though that number takes inbound calls only and cannot be ported out. Beyond that, you import numbers from carriers such as Twilio and Telnyx, or connect a SIP trunk. Its transfer tool handles blind and warm transfers, and Squads hand a call from one assistant to another.

Building the agent and checking every call

Dograh keeps a visual flow builder and grades each step of every call, while Vapi now builds agents as assistants.

Vapi retired its visual Workflows flow builder on 18 August 2026, and existing workflows stopped running the next day. Its agents are now Assistants and Squads, set up in the dashboard or in code, with help from a visual prompt editor and an AI builder called Composer. In Dograh, the call is a canvas of steps joined by pathways. Through Dograh's MCP server, an AI assistant such as Claude Code or Cursor can also create and edit agents for you. How to Migrate From Vapi to Open-Source Voice AI walks through the rebuild.

Dograh's post-call QA runs from a QA node in the flow. It reviews the transcript and gives each step its own score, with tags, sentiment and a short summary. There is no single score for the whole call, so you can see which step went wrong. It runs on test calls from the editor as well as on live calls, so you catch problems before launch. "Use Workflow's LLM" is on by default, and a Sample Rate below 100 picks calls at random. In production, we suggest QA on a random 5-10% of calls.

For debugging, every Dograh call has a View Trace button that opens the call's trace in Langfuse. It lists each speech-to-text result, each AI model call with its prompt and tools, and each tool call, so you can find the layer that failed. Vapi's call logs show per-turn latency for each stage, plus recordings and cost, and Vapi adds scorecards, which grade the whole call rather than each step, plus monitors and pre-launch Simulations. Vapi can send its traces to Langfuse too. Our post on Automated QA for Voice Agents covers QA setup in depth.

Join the Dograh Community

Dograh is an OSS alternative to Vapi. Join our Slack community for queries, releases, best practices & community interactions.

When Vapi may suit your team better

Vapi is the better pick for some teams, and it is better to say so here than to have you find out later.

Choose Vapi if you want no infrastructure at all and need the vendor to hold the certifications. Vapi reports SOC 2 Type II, HIPAA and GDPR compliance, and its Pro and Premier packages carry 99% and 99.9% uptime SLAs. It also fits if you need a US number in minutes with no carrier account, or if your engineers already run Squads in code.

Self-hosting Dograh is real work. You run Docker, apply upgrades and debug the server yourself, and the open-source version comes with community support only. Dograh Cloud removes the servers, and Dograh Enterprise adds on-prem deployment with custom SLAs.

If owning the code, the data path or the phone line matters to you, test the open-source route first: install it, rebuild one flow and run it over a real phone call before you move a number.

Glossary

Success Package
Vapi's name for its optional paid plans: Core, Pro and Premier. A package sets support, concurrent calls and data retention on top of usage; it does not change the $0.05 a minute hosting fee.
Bring your own key (BYOK)
Using your own account with a model provider inside a platform. You hold the contract and the bill, but the call audio still goes to that provider.
Asterisk ARI
The REST interface of the open-source Asterisk phone server. Self-hosted Dograh uses it to control calls on your own SIP trunk.
Call trace
The record of one call's speech-to-text results, AI model calls and tool calls, used to find which layer of the agent failed.

Frequently Asked Questions

Get started with Dograh

Build, deploy, and scale AI agents with Dograh. Join the community of developers building the future.