Vapi vs LiveKit: Hosted Voice API or Open-Source Framework

Vapi vs LiveKit is a closed hosted API against an open framework with its own cloud. Vapi runs only in its cloud, at $0.05 a minute for hosting plus models. LiveKit Agents is open source and runs on your servers or on LiveKit Cloud. Dograh, an open-source platform, is a ready product you can self-host or run on Dograh Cloud.

Key Takeaways

• Vapi includes 4 concurrent calls without a package, and LiveKit Cloud Build includes 5.
• LiveKit runs on your servers or its cloud, while Vapi runs only on its own.
• Vapi is quicker to start, but LiveKit gives developers the code.

This post is part of our guide to The Best Open-Source Voice AI Platforms Compared, which weighs Vapi against the open options. Here we take Vapi against LiveKit on building, phone calls, concurrency and price, then cover Dograh.

A closed API, or open code with a cloud

Vapi and LiveKit differ first in who owns the code. Vapi is closed source and calls itself "the developer platform for building voice AI agents". You configure agents through its dashboard and API, and every call runs on the company's cloud.

LiveKit Agents is an open-source framework under Apache-2.0. It runs your agent in a room on its open-source media server, which you can host yourself or leave to LiveKit Cloud. One license detail matters. Its turn detection models ship under the separate LiveKit Model License.

So the choice is between renting a finished product and building on open code with an optional managed cloud. Both connect to the same kinds of speech and language models, so the models, more than the platform, set call quality and most of the bill.

How you build the agent

In Vapi, an agent is an Assistant with its own prompt and tools, and a Squad hands a call between several assistants. You can bring your own model keys for any part of the pipeline or point it at your own language model through a custom LLM URL. Vapi retired its visual Workflows on 18 August 2026, so call logic now lives in assistants and squads.

In LiveKit, the agent is code in Python or Node. Handoffs between agents and multi-step workflows are written in code, and a built-in test framework lets you write tests with LLM judges. LiveKit's Agent Builder prototypes simple agents in the browser on the company's cloud, but it does not support handoff workflows or tests. Its Download code button hands you a Python project to carry on in code.

Phone numbers, SIP and transfers

Both put agents on the phone. Vapi provides its own numbers, and its telephony and SIP carry no extra charge. You can also bring numbers from carriers such as Twilio, which bill you directly. Premier customers can get dedicated SIP.

LiveKit speaks SIP natively for inbound and outbound calls. Its telephony docs list tested carriers, including Twilio, Telnyx, Exotel and Plivo, and cover cold transfers and agent-assisted transfers. LiveKit Cloud sells US local and toll-free numbers too, so you do not need an outside carrier just to get a number.

Concurrency and price

Concurrency is the clearest contrast. With no package, Vapi's pricing page lists 4 concurrent calls and 1 phone number. Core costs $29 a month for 10. Pro costs 10% of your hosting fee with a $999 monthly minimum and gives you 30, and extra lines cost $10 each a month. On top sits "$0.05/min Vapi hosting", with models passed through at cost. HIPAA is a $2,000 a month add-on.

LiveKit Cloud's pricing is tiered by plan. Build is free, with 1,000 agent session minutes and 5 concurrent sessions. Ship starts at $50 a month with 5,000 minutes and 20 concurrent sessions, and Scale starts at $500 with 50,000 minutes and up to 600. On both paid plans, extra minutes cost $0.01 each. US local numbers cost $1 a month after the first free one.

Self-hosted LiveKit has no platform fee at all. Your servers set how many calls run at once, and you pay for those servers and for model and phone usage.

Open Source Alternative to Vapi / Retell

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

dograh-hq/dograh

Star on GitHub

Vapi vs LiveKit side by side

The "where it runs" row matters most for regulated calls. Vapi's FAQ says "No, Vapi does not support on-premise deployments." LiveKit can run on your own servers, though call audio stays fully in-house only when the speech and language models run there too.

Pick Vapi if you want a phone agent running this week and have no one to run servers. Pick LiveKit if your developers want the code and expect to outgrow a few concurrent calls. It also suits products that need video or in-app voice.

Dograh: a hosted product's ease, open code

Dograh sits between the two. Like Vapi, it is a finished product for phone agents. Like LiveKit, its code is open, under BSD 2-Clause, so you can self-host it or run it on Dograh Cloud.

You build in a visual workflow builder, or let a coding agent such as Claude Code build through Dograh's MCP server.

Calls run through your own carrier, with eight telephony providers built in. Transfer Call works on Twilio, Telnyx, Plivo, Cloudonix and Asterisk ARI. Dograh calls the destination first and connects the caller only when someone answers. On Twilio, it can play a short spoken introduction to both people before connecting them.

Outbound lists run as Campaigns, and a QA node scores each step of a call after it ends. You can use the platform's own models or bring your own keys, and any model served behind an OpenAI-compatible endpoint works too, fully local ones included. Self-hosted Dograh has no per-minute platform fee (servers, model keys and phone minutes still cost). Dograh Cloud has a 1¢-a-minute platform fee.

Our Dograh vs Vapi and Dograh vs LiveKit comparisons go deeper.

If cost is what pushes you off Vapi, our post on the open-source alternative to Vapi works through the numbers.