Contents
Fast answer
LiveKit publishes the lowest agent-minute rate in voice, and it covers the least. As of 20 August 2026 Build is free with 1,000 agent session minutes, 5 concurrent sessions, 1 deployment and $2.50 in inference credits; Ship is $50 a month with 5,000 minutes, 20 concurrent sessions, 2 deployments and $5 in credits; Scale is $500 a month with 50,000 minutes, up to 600 concurrent sessions, 4 deployments and $50 in credits; Enterprise is custom. Overage on paid plans is $0.01 per agent minute, which is also exactly what the included minutes work out at: $50 divided by 5,000 and $500 divided by 50,000 both give a cent. Around that sit separate meters: third-party SIP at $0.004 a minute on Build and Ship or $0.003 on Scale, downstream data at $0.12 a gigabyte on Ship or $0.10 on Scale, and additional phone numbers at $1.00 a month after the free local number. The language model, transcription and speech synthesis are yours to buy.
Facts last verified 2026-08
Quick facts
- Build
- $0/mo, 1,000 agent minutes, 5 concurrent sessions (checked 2026-08-20)
- Ship
- $50/mo, 5,000 agent minutes, 20 concurrent sessions
- Scale
- $500/mo, 50,000 agent minutes, up to 600 concurrent sessions
- Overage
- $0.01 per agent minute, identical to the included rate
- Third-party SIP
- $0.004/min on Build and Ship, $0.003/min on Scale
- Downstream data
- $0.12/GB on Ship, $0.10/GB on Scale
- Phone numbers
- One free local number, $1.00/month per additional number
- Not included
- Language model, transcription, speech synthesis and telephony
On this page
What this page covers
A comparison visitor should understand the tradeoff, the best-fit scenario, and the next diagnostic tool to confirm the choice.
- 01What it costs
- 02The other meters
- 03Worked example
- 04Where the meter bites
- 05Cheapest honest path
- 06When to leave
Why does this matter now?
A cent a minute looks like it undercuts every other voice platform by an order of magnitude, and it does, because it is a different product. LiveKit sells real-time infrastructure, not a working agent. The comparison people should be making is not LiveKit's minute against a bundled vendor's minute, it is LiveKit's minute plus the model plus transcription plus speech plus telephony plus the engineer, against that bundled vendor's minute. Done properly the gap narrows sharply, and for a team without an engineer it inverts.
Internal path
Where to go next from this page
These links are part of the A8gent learning and conversion path. Use them to move from concept, to diagnosis, to workflow build, to course.
What you should be able to do after this
- See what the $0.01 minute does and does not cover
- Add the meters LiveKit does not bundle
- Size concurrency rather than volume
- Judge self-hosting against the managed tier
How do you do it, step by step?
1. Understand what the agent minute covers
The $0.01 agent session minute is LiveKit's own orchestration and media handling. It does not include the language model, speech-to-text, text-to-speech or telephony. Those are separate suppliers you choose and pay directly, which is a genuine advantage for control and a genuine liability for anyone expecting a single bill.
2. Add every other meter before comparing
Third-party SIP is $0.004 a minute on Build and Ship, $0.003 on Scale. Downstream data is $0.12 a gigabyte on Ship and $0.10 on Scale. Extra phone numbers are $1.00 a month each beyond the included free local number. Inference credits of $2.50, $5 and $50 are starter allowances, not an inference budget.
3. Size on concurrent sessions, not minutes
Build allows 5 concurrent sessions, Ship 20 and Scale up to 600. Since overage on minutes costs exactly the same as included minutes, the tier you need is decided by simultaneous calls at peak, not by monthly volume. Work out the most calls you have had at once in a busy hour and buy for that.
4. Cost the engineering, honestly
LiveKit is infrastructure. There is no dashboard where you configure an agent and go live, and making one reliable is real engineering work followed by real operational work. If the alternative is a bundled platform at ten or fifteen cents a minute, a few thousand minutes a month will not repay a fortnight of build plus ongoing maintenance for a long time.
5. Only consider self-hosting if you want the operational job
The Agents framework and the media server are both open source, so you can run the whole thing on your own infrastructure and pay LiveKit nothing. Real-time media at scale is a serious operational undertaking, not a weekend project. Choose it when regulatory or data-location requirements demand it, not to save $50 a month.
What a cent a minute actually covers
LiveKit publishes the lowest per-minute figure in voice AI by a wide margin, and it covers the least of any platform in the category. The $0.01 agent session minute pays for LiveKit's real-time media handling and session management. It does not include the language model, speech-to-text, text-to-speech or telephony. Those are four separate suppliers you choose, contract with and pay directly, which is a genuine advantage for control and a genuine liability for anyone expecting one invoice.
The tier structure confirms that the rate is fixed rather than discounted. Build is free with 1,000 agent session minutes, 5 concurrent sessions, 1 deployment and $2.50 in inference credits. Ship is $50 a month for 5,000 minutes, 20 concurrent sessions, 2 deployments and $5 in credits. Scale is $500 a month for 50,000 minutes, up to 600 concurrent sessions, 4 deployments and $50 in credits. Divide Ship and Scale by their minutes and both give exactly $0.01, which is also the published overage rate.
That means the ladder buys concurrency and deployments, not a better unit price, exactly as ElevenLabs Agents does at $0.08 a minute. The buying rule is the same in both cases: pick the smallest tier whose concurrent session limit covers your busiest hour and let minutes spill into overage, because overage costs precisely what included minutes cost. Sizing up to pre-buy minutes is paying earlier for the identical thing.
The inference credits are a starter allowance rather than a budget and should not be modelled as one. $2.50, $5 and $50 respectively will cover early testing and will not cover a production month of real model usage. Once they are consumed, inference bills per model at the provider's rates, and that line will typically dwarf the agent minute charge. Treating the credits as an inference plan is the fastest way to build a forecast that is wrong by an order of magnitude.
The meters around the edges
Beyond agent minutes and inference, three further meters apply and each catches people out. Third-party SIP is $0.004 a minute on Build and Ship and $0.003 on Scale, which is the telephony transport and is entirely separate from your telephony provider's own charges. Downstream data is $0.12 a gigabyte on Ship and $0.10 on Scale. Additional phone numbers are $1.00 a month each beyond the one free local number included.
Bandwidth is the meter people have least intuition for, because voice engineers think in minutes and finance thinks in dollars and neither thinks in gigabytes. It scales with session count and duration rather than with anything on your business dashboard, and it is the line most likely to appear larger than expected on a first real month. Model it explicitly rather than assuming it rounds to nothing, particularly if you are running many concurrent sessions.
Put the whole picture together for a realistic small deployment and the cent stops being the headline. At 1,500 agent minutes a month you owe LiveKit $15 if you are on overage, or $50 for the Ship tier that gives you 20 concurrent sessions. Add SIP, add bandwidth, add a model, add speech-to-text, add text-to-speech, add your telephony provider. The platform slice is a small and predictable part of a total that is mostly other people's invoices.
None of that is a criticism of the pricing, which is transparent and cheap for what it is. It is a warning about the comparison people actually make, which is LiveKit's $0.01 against a bundled vendor's $0.14 and a conclusion that LiveKit is fourteen times cheaper. The correct comparison adds every meter above plus the engineering, and once you do that honestly the gap narrows sharply and for some teams reverses entirely.
This is infrastructure, and that is a cost line
There is no dashboard on LiveKit where you configure an agent and go live. It provides the real-time plumbing and expects you to supply everything else, including the person. That is a deliberate and honest positioning rather than a gap, and it is the most important fact on this page, because it means the largest number in your total cost of ownership does not appear on the pricing page at all.
Price it properly. Assembling a working voice agent on real-time infrastructure is a serious engineering exercise, and making it reliable, handling interruptions gracefully, dealing with silence, managing barge-in, recovering from provider hiccups, is a considerably larger one. Against a bundled platform at $0.14 a minute, LiveKit's rate advantage on 1,500 minutes a month is well under $200. A fortnight of an engineer's time is worth many months of that difference, and the maintenance never stops.
The self-hosting option deserves the same scrutiny. The Agents framework and the media server are both open source and can run on your own infrastructure with nothing paid to LiveKit at all. Running real-time media at scale is a substantial operational undertaking rather than a weekend project, involving capacity planning, network tuning and being on call for something customers hear in real time. Choose it when a regulatory or data-location requirement demands it, not to avoid a $50 platform fee.
The clean test for whether LiveKit is right is a staffing question rather than a pricing one. Is there an engineer who wants to own the voice stack, who will still be here in six months, and whose time is not already fully committed elsewhere? If yes, this is excellent infrastructure at an honest price and the control is worth having. If no, a bundled platform costs more per minute and vastly less in total, because the alternative is a half-finished integration nobody owns.
At a glance
LiveKit published plans and additional meters, read from livekit.com/pricing on 20 August 2026
| Line item | Published price | Note |
|---|---|---|
| Build | $0/mo | 1,000 agent minutes, 5 concurrent, 1 deployment, $2.50 inference credits |
| Ship | $50/mo | 5,000 agent minutes, 20 concurrent, 2 deployments, $5 inference credits |
| Scale | $500/mo | 50,000 agent minutes, up to 600 concurrent, 4 deployments, $50 inference credits |
| Enterprise | Custom | Custom minutes and concurrency, unlimited deployments, volume pricing |
| Agent minute overage | $0.01 per minute | Same as $50 divided by 5,000 and $500 divided by 50,000 |
| Third-party SIP | $0.004/min Build and Ship, $0.003/min Scale | Telephony transport, separate from agent minutes |
| Downstream data | $0.12/GB Ship, $0.10/GB Scale | Bandwidth is its own meter |
| Additional phone numbers | $1.00/month each | One free local number is included |
| Model, transcription, speech | Your own providers | Not included in the agent minute at any tier |
| Self-hosted | $0 to LiveKit | Agents framework and media server are open source |
What mistakes should you avoid?
- Comparing $0.01 an agent minute against a bundled vendor's all-in rate
- Treating the $2.50, $5 or $50 inference credits as an inference budget
- Sizing on monthly minutes when overage costs the same as included minutes
- Forgetting bandwidth, SIP and phone number meters when modelling the bill
- Choosing LiveKit without an engineer who will still own it in six months
- Self-hosting a real-time media server to save a platform fee rather than to meet a requirement
FAQ
- How much does LiveKit Agents cost?
- Build is free with 1,000 agent session minutes and 5 concurrent sessions, Ship is $50 a month with 5,000 minutes and 20 concurrent, Scale is $500 a month with 50,000 minutes and up to 600 concurrent, and Enterprise is custom, as published on 20 August 2026. Overage on paid plans is $0.01 per agent minute.
- Is $0.01 a minute really the whole cost?
- No. It covers LiveKit's orchestration and media handling only. The language model, transcription, speech synthesis and telephony are all separate suppliers you pay directly, and LiveKit adds meters for third-party SIP at $0.004 or $0.003 a minute, downstream data at $0.12 or $0.10 a gigabyte, and extra phone numbers at $1.00 a month.
- Do the higher tiers give a better rate?
- No. Ship at $50 for 5,000 minutes and Scale at $500 for 50,000 minutes both work out at exactly $0.01 a minute, which is also the overage rate. The tiers buy concurrent sessions, deployments and inference credits, so size on peak concurrency rather than on monthly volume.
- Can I self-host LiveKit for free?
- Yes. The Agents framework and the media server are open source and can run on your own infrastructure with nothing paid to LiveKit. Running real-time media reliably is a substantial operational commitment, so this is a route for teams with a specific data-location or compliance requirement rather than a way to avoid a $50 platform fee.
- Who should not use LiveKit?
- A small business that wants a working phone agent rather than a platform to build one on. LiveKit provides excellent real-time plumbing and expects you to supply everything else, including the engineer. If nobody in the business will own that, a bundled voice platform costs more per minute and much less in total.
Sources & further reading
Done for you
Want this built for your business?
Tell us the workflow. We scope, build, and ship the agent with guardrails and the numbers to prove it worked. The scoping call is free.
Free scoping call. You own the code.
Was this page helpful?

