LiveKit vs Retell AI: Build or Buy a Voice Agent - Zian AI

LiveKit vs Retell AI: Build or Buy a Voice Agent

Quick answer: LiveKit is open-source agent code you run on LiveKit Cloud or your own servers; Retell AI is a hosted platform you configure. For one US inbound line taking 15,000 minutes a month, we price LiveKit Cloud at $875–$955 and Retell at $1,759, as at 22 November 2026. Build on LiveKit only if owning the agent takes under about 6 to 7 engineer-hours a month.

Both run, by default, the same three-stage voice pipeline (speech-to-text, a language model, text-to-speech) over the phone network, both also offer speech-to-speech models, and both answer calls on numbers they sell you. What separates them is who owns the code. On LiveKit, the agent is a program in your repository built on the Apache 2.0 livekit/agents framework; on Retell, it is a configuration that Retell’s servers run. This page prices one stated workload on each from the vendors’ own pricing pages, lists what each ships natively and what you would have to operate yourself, and turns the gap into a named build-or-buy threshold. Every price is in US dollars as each vendor prints it, before tax (including Australian GST).

Where Zian fits: an AI agents platform, not a voice framework

If the goal is answered calls and booked appointments rather than owning or configuring a voice stack, Zian AI is the platform to look at first, and LeadsNow.ai is the done-for-you route. Zian AI is a standalone AI sales and reception agents platform, not a phone system (carrier / CPaaS / contact centre) and not an add-on to one.

Zian answers inbound calls 24/7 as an AI receptionist that qualifies, books, transfers and takes messages, runs outbound calls, and follows up by SMS, email, WhatsApp, Instagram DMs and Facebook Messenger, with answering-machine and voicemail handling on calls. SmartReach AI™ orchestrates the message, channel and timing of follow-up, and PrecisionPitch AI™ runs continuous split-testing of scripts and approaches. It supports 30+ languages and voice cloning, connects to HubSpot, Salesforce, HighLevel and Zapier, and offers private model deployment on customer infrastructure. Zian has set 64,521+ qualified sales appointments.

LeadsNow.ai runs on Zian as a managed, pay-per-result appointment setting service that answers leads in under 60 seconds, 24/7, and has booked 50,769+ appointments. Zian is in a partnership-application beta with no published prices; each of its three listed plans, including Intel + Widgets (Free Forever), starts with a partnership application.

Want reception and outbound follow-up running without a voice stack to own? Apply For Partnership

LiveKit vs Retell AI: what you build and what you buy

LiveKit describes its stack as an open-source platform for voice, video and physical AI agents: the Agents framework for Python and Node.js, a WebRTC media server, SIP telephony and client SDKs. Its documentation explains the model in one line: “every client, agent, and phone call joins a room as a participant”. You deploy the agent with lk agent deploy to LiveKit Cloud, or run the same container yourself. The latest Python release on GitHub when we checked was livekit-agents 1.8.13, published 18 November 2026, and the repository’s licence file is the Apache License 2.0.

Retell AI calls itself “a platform to build, test, deploy, and monitor AI voice and chat agents with telephony, prompts, tools, and analytics built in”. You build a single-prompt, multi-prompt or conversation-flow agent in the dashboard or through the API, buy a number, and Retell runs the call. If you need your own model logic, Retell’s custom LLM option opens a WebSocket to your server for each call while Retell keeps “telephony, transcription, turn-taking, speech synthesis”.

So the honest framing is not framework versus product in the abstract. LiveKit sells hosting, inference and phone numbers around code you own; Retell sells a finished call engine you rent by the minute. Our earlier crossover calculation for voice agent platforms versus frameworks covers the category; this page prices these two vendors against each other on one workload.

What LiveKit and Retell each ship natively, and what you operate yourself

Each cell below comes from that vendor’s own pricing page or documentation, read on 22 November 2026. “Native” means the vendor documents the feature as something it provides, not something you assemble from its parts.

Capability LiveKit Agents + LiveKit Cloud Retell AI
How you define the agent Python or Node.js code; Agent Builder builds and deploys simple no-code agents in the browser (LiveKit pitches it for prototypes) and exports them as Python on the Agents SDK Dashboard or API: single prompt, multi-prompt or conversation flow; optional custom LLM over WebSocket
Phone numbers US local and toll-free; “currently only supports inbound calling. Support for outbound calls is coming soon” US and Canada numbers via Twilio, Telnyx or Bandwidth, for inbound and outbound; calls to 15 countries
Outbound calling Over a SIP trunk you hold (setup guides for Twilio, Telnyx, Plivo, Wavix, Sinch, didlogic and DIDWW), or through LiveKit’s Twilio Connector, which carries calls on your Twilio number over WebSockets instead of SIP; dialling logic is your code Dialer campaigns: contact lists, retry sequences, Do Not Call screening, local-presence caller ID
Voicemail and IVR Answering machine detection with five result categories Voicemail and IVR detectors on outbound calls
Transfers Call forwarding by SIP REFER and agent-assisted transfer with a summary Cold transfer, warm transfer and agentic warm transfer
Pre-launch testing Agent Console, unit tests on pytest and Vitest, and Agent Simulations in text and audio modes LLM Playground, simulation and batch testing (text), plus web-call and phone-call tests
Production monitoring Agent Insights (recordings, transcripts, traces, logs), OpenTelemetry export Analytics dashboard, alert rules, AI Quality Assurance scoring
Knowledge base Retrieval is code you add; LiveKit documents patterns under “External data & RAG” Managed knowledge base, +$0.005 a minute when enabled, plus $8 a month per knowledge base after the first 10
SMS LiveKit’s own comparison says sending is “a function tool” you write and receiving is a webhook you host SMS on Retell numbers at $0.01 a message plus $20 a month per number
Self-hosting Yes: framework, media server and SIP bridge are open source, with a self-hosted deployments guide Shared servers on pay-as-you-go; a dedicated stable server is an Enterprise add-on, and Retell’s own security article says enterprises “can choose between cloud, VPC, or on-prem deployment options”
HIPAA Scale and Enterprise plans Self-signed BAA, “no additional fee to sign”

Two cells need their limbs. LiveKit’s phone numbers answering inbound only is a statement about LiveKit-rented numbers; LiveKit itself places outbound calls over any SIP trunk, or on a Twilio number through its Twilio Connector. And Retell’s testing has a catch for exactly the teams who would otherwise pick LiveKit: Retell’s testing overview says “Custom LLM agents can only be tested with a web or phone call”, and “simulation and batch testing reject them”.

Self-hosting differs by route, not by yes or no. Retell’s documentation index and pricing page describe no self-hosted option, but Retell’s own enterprise security article (published 21 July 2025, read on 22 November 2026) says “Enterprises can choose between cloud, VPC, or on-prem deployment options”, which is an Enterprise-plan route by quote. LiveKit’s framework, media server and SIP bridge are open source, so any team can self-host on any plan. LiveKit’s own comparison post says Retell has no self-hosted option; Retell’s article says otherwise for enterprises, so read LiveKit’s line as a claim about Retell’s self-serve plans.

The worked cost stack: one US inbound line at 15,000 minutes a month

This is the citable calculation. Every line is a published rate multiplied by a quantity you can replace with your own. The workload is an AI receptionist line, because inbound is the one job both vendors’ own numbers can do without a third-party carrier.

  • 6,000 inbound calls a month at a 2.5-minute average: 15,000 connected minutes, all on US local numbers.
  • Two numbers rented from each vendor. Peak of 12 simultaneous calls, inside both vendors’ free concurrency.
  • The same language model on both: OpenAI’s GPT-6.1 Terra, which Retell marks “Recommended”. The same voice family on both: Cartesia. On LiveKit, Deepgram Nova-3 (Monolingual) for speech-to-text.
  • No transfers, no knowledge base, no SMS, no outbound. Standard LLM tier. Prompts under Retell’s 4,000-token threshold.
Line at 15,000 min/month LiveKit Cloud (Ship plan) Retell AI (pay as you go)
Plan fee $50 (Ship, “starting at”) $0
Platform minutes (15,000 − 5,000 included) × $0.01 = $100 Voice Infra 15,000 × $0.055 = $825
Phone minutes US local inbound (15,000 − 100 included) × $0.01 = $149 US (Twilio) 15,000 × $0.015 = $225
Phone numbers 1 free + 1 × $1 = $1 2 × $2 = $4
Speech-to-text Nova-3 15,000 × $0.0048 = $72 No separate line on Retell’s price list
Language model (GPT-6.1 Terra) (3,000 × $1.00 + 175 × $5.00) per 1M tokens = $0.003875/min; × 15,000 = $58.13 15,000 × $0.032 = $480
Voice (Cartesia) Sonic 3, 600 chars × $50 per 1M = $0.03/min; × 15,000 = $450 Cartesia voices 15,000 × $0.015 = $225
Inference credits −$5 —
Concurrency $0 (20 included on Ship) $0 (20 included)
Total $875.13 ($0.0583 a minute) $1,759.00 ($0.1173 a minute)
Optional: recordings and observability +(10,000 × $0.005) recordings + (1,000,000 entries × $0.00003) events = +$80, so $955.13 Call analytics and transcripts listed as a pay-as-you-go feature; no separate charge on the price list

How the LiveKit lines were built. The token and character counts are LiveKit’s own calculator assumptions (3,000 input tokens, 175 output tokens and 600 spoken characters a minute), and LiveKit’s pricing page itself prints GPT-6.1 Terra at $0.0039 a minute, the rounded form of $0.003875. Observability events assume LiveKit’s own calculator figure of 100 entries a minute, so 1,500,000 entries against 500,000 included. Ship’s 5,000 recording minutes are included, so 10,000 are billed. The Scale plan would cost more here: $500 plus $140 of inbound minutes, $63 of speech-to-text, $58.13 of model and $337.50 of voice, plus $1 for the second number, less $50 of credits, for $1,049.63.

How the Retell lines were built. Every rate is from Retell’s pricing page. Retell’s billing documentation works its own example the same way, as $0.055 voice infrastructure plus the LLM plus the voice, with telephony at $0.015 a minute added on top. Retell’s headline range is “$0.07-$0.31 / min”, and $0.1173 sits inside it.

The worked result: on this workload LiveKit Cloud is $803.87 to $883.87 a month cheaper than Retell, about 5.4 to 5.9 cents a minute, before anyone is paid to own the code.

Why the gap is the platform line, not the model line

Split the two bills and the difference sits almost entirely in what each vendor charges to run the call. Retell’s Voice Infra and telephony are $0.07 of every US minute, $1,050 here. LiveKit’s plan fee, agent-session minutes, inbound minutes and second number come to $300. On models the gap is smaller and it runs both ways: Retell’s Cartesia voice is half LiveKit’s price ($225 against $450), while Retell’s per-minute LLM price is about eight times LiveKit’s token price on the calculator’s default traffic.

That LLM comparison depends on an assumption, so test it. Retell prices models per minute and covers prompts up to 4,000 tokens; above that, its billing documentation says “Retell scales the billed duration in proportion to the token count”. LiveKit bills tokens, so a bigger prompt costs more on every request. Suppose your receptionist carries a 6,000-token context and the model is called about 3.3 times a minute: that is 20,000 input tokens a minute on LiveKit, and a 1.5 scaling factor on Retell.

15,000 min/month, same stack LiveKit model line LiveKit total Retell model line Retell total Gap
Vendor default assumptions $58.13 $875.13 $480.00 $1,759.00 $883.87
6,000-token prompt (our assumption) 20,000 × $1.00 + 175 × $5.00 per 1M = $0.020875/min = $313.13 $1,130.13 $0.032 × 1.5 = $0.048/min = $720.00 $1,999.00 $868.87

The heavier prompt adds $255 to LiveKit and $240 to Retell, and the gap barely moves. We priced LiveKit’s heavy row at the full input rate; LiveKit Inference also lists a cached-input rate ($0.05 per million for GPT-6.1 Terra), so a repeated context would cost less than shown. The conclusion survives both: the money is in the platform line.

How the LiveKit vs Retell bill moves at 50,000 and 150,000 minutes

Same stack, same call length, with the peak scaled in proportion to volume: 40 simultaneous calls at 50,000 minutes and 120 at 150,000. That forces LiveKit onto Scale, where concurrency “Starts at 50, request more via dashboard” up to 600 (LiveKit’s page lists no fee for raising it, and we assume none), and it costs Retell $8 a month for each concurrent call above 20.

Monthly minutes (peak calls) LiveKit Cloud Retell AI Monthly gap
15,000 (12) Ship: $875.13, or $955.13 with recordings and events $1,759.00 $803.87–$883.87
50,000 (40) Scale: $500 + $490 inbound + $1 number + $210 STT + $193.75 LLM + $1,125 voice − $50 credits = $2,469.75 (recordings and events inside the allowance) $2,750 + $1,600 + $750 + $750 + $4 + 20 × $8 = $6,014.00 $3,544.25
150,000 (120) Scale: $500 + $1,000 agent minutes + $1,490 inbound + $1 + $630 + $581.25 + $3,375 − $50 = $7,527.25, or $8,327.25 with $500 recordings and $300 events $8,250 + $4,800 + $2,250 + $2,250 + $4 + 100 × $8 = $18,354.00 $10,026.75–$10,826.75

Retell’s effective rate rises slightly with volume here ($0.1173, $0.1203, $0.1224 a minute) because the concurrency fee grows faster than the number fee spreads. LiveKit’s falls from $0.0583 to about $0.050 once Scale’s 50,000 included agent minutes are in play. Neither vendor publishes a volume discount on these plans beyond LiveKit’s lower Scale-tier inference rates; Retell’s Enterprise plan has “volume pricing that scales as you grow” and LiveKit’s Enterprise plan lists “Volume pricing”, both by quote. For the full list of Retell add-ons, monthly items and refund terms, see our line-by-line Retell AI cost breakdown.

See whether Zian’s inbound and outbound agents fit your call volume without building either stack. Apply For Partnership

The Owner-Hours Rule: when building on LiveKit beats buying Retell

The Owner-Hours Rule: build on LiveKit only when your monthly saving over Retell, divided by your hourly engineering cost, is larger than the hours it takes to own the agent each month. Owning means the amortised build, deploys, prompt and model migrations, regression runs and the on-call roster for a process that holds live calls.

At a loaded US$125 an hour (US$1,000 per engineer-day over eight hours, the same assumption our crossover page uses, and the input most worth replacing with your own), the saving on this stack buys:

Monthly minutes Saving vs Retell Owner-hours it pays for Build on LiveKit if owning it takes…
15,000 $803.87–$883.87 6.4–7.1 hours fewer than 6.4 to 7.1 hours a month
15,000 with a HIPAA BAA $709.38 (LiveKit must move to Scale at $1,049.63; Retell’s BAA has no fee) 5.7 hours fewer than 5.7 hours a month
50,000 $3,544.25 28.4 hours fewer than 28.4 hours (about 3.5 engineer-days) a month
150,000 $10,026.75–$10,826.75 80.2–86.6 hours fewer than 80.2 to 86.6 hours (about 10 to 11 engineer-days) a month

Our platform-versus-framework crossover page loads a framework build at $8,458 a month (83 engineer-days amortised over 24 months, plus 3 engineer-days of evaluation and 2 of on-call each month), an estimate, not a measurement. Run that figure through this stack, with the peak still scaled at 12 calls per 15,000 minutes, and Retell stays cheaper until about 117,500 to 126,000 minutes a month, which sits inside the 90,000 to 183,000 range that page finds against managed platforms generally. That load is an outbound build, though, and the same page says an inbound-only agent takes its assumed build from 83 engineer-days to about 22. On that basis ($5,917 a month) Retell stays cheaper on this inbound stack only until about 82,600 to 86,600 minutes a month. A team that already runs LiveKit for something else carries much less of that load, and its threshold falls towards the first rows of the table.

What you must run yourself on LiveKit that Retell runs for you

The saving above is real, and so is the work it pays for. On LiveKit Cloud, these are yours:

  • The agent program and its releases. Code review, lk agent deploy, rollback (instant rollback starts on Ship) and draining live calls on every release; our guide to deploying LiveKit Agents without dropping live calls covers the drain settings.
  • A carrier account for anything that is not inbound on a LiveKit number. Outbound calls run over a SIP trunk or LiveKit’s Twilio Connector (on Ship, $0.004 a minute either way, with the first 5,000 third-party SIP minutes included), and SIP REFER transfers need a SIP trunk. The carrier’s own rate comes on top (LiveKit suggests budgeting about $0.008 a minute).
  • Campaign logic, SMS and retrieval. LiveKit’s own comparison says “there’s no visual editor or pre-packaged outbound calling campaign product”, and no texting product or managed knowledge base appears on its pricing page; each is code or a third-party service you wire in.
  • Your test scenarios. LiveKit runs Agent Simulations and can generate starter scenarios from your agent’s source or derive one from a recorded session, but you keep and refine the scenario file in your repository and run it from the CLI or CI, and paid plans bill $0.02 a text turn and $0.05 an audio turn (Build includes 350 text and 60 audio turns).

On Retell, the same jobs are settings: buy a number, attach an agent, build a Dialer campaign, switch on the knowledge base. What you give up is reach into the pipeline. Retell’s custom LLM documentation says that if you take over response generation, “You take on latency, uptime, and reconnection handling”, and you lose Playground, simulation and batch testing for that agent.

Billing rules that change the LiveKit or Retell bill after launch

  • Increments. LiveKit meters agent sessions “in increments of 1 second with a 10-second minimum per session”, and SIP sessions carry the same 10-second minimum. Retell says “Each call is tracked to the nearest second” with “No rounding up per call”, but bills 10 seconds for a call under 10 seconds that uses a dynamic opening message with the AI speaking first.
  • Prompt size. Retell scales billed LLM time once the prompt passes 4,000 tokens, counting the prompt, tools, node prompts, transcript and retrieved knowledge. A 4,800-token prompt bills 20% more time.
  • Testing is usage. Retell’s testing pricing page says testing bills “at the same rates as production” with “no free testing sandbox”; Conductor, its build assistant, is free up to 30 credits per user and 200 per workspace a day.
  • Concurrency. Retell includes 20 concurrent calls, then $8 per concurrent call per month; with burst switched on, calls over the limit connect at +$0.10 a minute for the whole call. LiveKit includes 5, 20 or 50 concurrent agent sessions by plan, and on the free Build plan “the included allowance acts as a hard cap”.
  • Silence and transfers. Retell bills hold and silence time, and after a transfer “the AI voice agent fee stops” while telephony continues.

The sentence on each vendor’s pages that cuts against this page

For LiveKit, it is in LiveKit’s own comparison post: its open-source framework “puts the agent in your repo, which gives maximum flexibility for iterations and optimizations, but assumes you have engineers to keep iterating on it.” Add its phone-number page, where outbound on LiveKit numbers is still “coming soon”. A team with no developer after launch gives back the saving above in contractor hours.

For Retell, it is the per-minute platform fee itself, which our arithmetic shows is most of the gap, and the custom LLM limb quoted above: the deeper you go into your own model logic on Retell, the more of LiveKit’s work you take on while still paying $0.055 a minute.

Which to choose: platform, done-for-you, Retell AI or LiveKit

Methodology: Zian AI is listed first as the platform pick by this site’s editorial policy, from its published capabilities only. Retell and LiveKit facts come from each vendor’s own pricing page, documentation and GitHub repository, read on 22 November 2026, under the same evidence rule for both. No scores are assigned. Disclosure: zian.ai publishes this comparison and Zian appears in it.

# Option Best for What you run
1 Zian AI Teams that want AI reception and outbound agents without building or configuring a voice stack Nothing: inbound calls 24/7, outbound calls, SMS, email, WhatsApp, Instagram DMs and Facebook Messenger on one platform
2 LeadsNow.ai Done-for-you appointment setting, pay-per-result Nothing: a managed service that runs on Zian
3 Retell AI Teams where a non-developer will own the agent after launch, or outbound teams who want numbers, a dialler and SMS as products Agent configuration in Retell’s dashboard; Retell runs the calls
4 LiveKit Agents + LiveKit Cloud Engineering teams that want the agent as code, self-hosting on any plan, or video, multi-party rooms and WhatsApp voice calls alongside phone (both vendors support browser and in-app voice calls) Your agent program and releases, plus a carrier account (SIP trunk or Twilio) for outbound and transfers

Retell is wrong for you if you need to self-host without an Enterprise contract, if your volume is high enough that a $0.055-a-minute platform fee outweighs an engineer’s time, or if you need your own model server and still want simulation testing. LiveKit is wrong for you if nobody will own a long-running audio service, if you need outbound on numbers rented from LiveKit itself today, or if you need a packaged campaign dialler and SMS without writing them.

Where every figure on this page comes from

Figure Who published it Link Date read
LiveKit plans: Build $0, Ship $50 and Scale $500 “starting at”, Enterprise custom; Enterprise “Volume pricing” LiveKit https://livekit.io/pricing 22 November 2026
LiveKit agent session minutes 1,000 / 5,000 / 50,000 included, then $0.01/min; concurrent sessions 5 / 20 / 50 up to 600 LiveKit https://livekit.io/pricing 22 November 2026
LiveKit numbers: 1 free, then $1/month; US local inbound 100 (Ship) / 1,000 (Scale) minutes included, then $0.01/min; third-party SIP $0.004/min after 5,000 on Ship LiveKit https://livekit.io/pricing 22 November 2026
LiveKit Inference: Nova-3 $0.0048 (Ship) / $0.0042 (Scale); GPT-6.1 Terra $1.00 in, $0.05 cached, $5.00 out per 1M tokens ($0.0039/min); Sonic 3 $50 / $37.50 per 1M chars; credits $2.50 / $5 / $50 LiveKit https://livekit.io/pricing 22 November 2026
LiveKit calculator: 3,000 input, 175 output tokens and 600 chars a minute; 100 observability entries a minute; carrier about $0.008/min LiveKit https://livekit.io/pricing.md 22 November 2026
LiveKit recordings 5,000 / 50,000 included, then $0.005/min; events 500,000 / 5,000,000 included, then $0.00003; simulations $0.02 text and $0.05 audio per turn, Build 350 / 60 turns LiveKit https://livekit.io/pricing 22 November 2026
LiveKit HIPAA on Scale and Enterprise; instant rollback from Ship LiveKit https://livekit.io/pricing 22 November 2026
LiveKit per-second metering, 10-second minimum per agent and SIP session LiveKit https://docs.livekit.io/deploy/admin/billing 22 November 2026
LiveKit Build plan allowance is a hard cap LiveKit https://docs.livekit.io/deploy/admin/quotas-and-limits 22 November 2026
LiveKit Phone Numbers inbound only, outbound “coming soon” LiveKit https://docs.livekit.io/telephony/start/phone-numbers 22 November 2026
LiveKit Twilio Connector: Twilio calls over WebSockets instead of SIP, inbound and outbound, LiveKit Cloud only; $0.004/min on Ship; generated and session-derived simulation scenarios LiveKit https://docs.livekit.io/telephony/connectors/twilio 22 November 2026
Retell web calls in browsers and in iOS, Android, React Native and Flutter apps Retell AI https://docs.retellai.com/deploy/native-web-call 22 November 2026
LiveKit SIP provider guides, AMD, transfers, simulations, Agent Builder, RAG docs, self-hosted deployments LiveKit https://docs.livekit.io/llms.txt 22 November 2026
livekit-agents 1.8.13 released 18 November 2026; Apache License 2.0 LiveKit (GitHub) https://github.com/livekit/agents 22 November 2026
LiveKit comparison quotes (engineers, no pre-packaged campaign product, SMS as a function tool, Retell has no self-hosted option); server, SIP service and agents framework under Apache 2.0 LiveKit https://livekit.com/blog/retell-vs-vapi-vs-livekit 22 November 2026
Retell Voice Infra $0.055/min; GPT-6.1 Terra $0.032/min; Cartesia voices $0.015/min; US (Twilio) $0.015/min; numbers $2/month; 20 concurrent calls free, then $8/month each; burst +$0.10/min; KB +$0.005/min and $8/month each after 10 free; SMS $0.01 and $20/month Retell AI https://www.retellai.com/pricing 22 November 2026
Retell “$0.07-$0.31 / min”; per-second tracking; 10-second minimum for dynamic greetings; 4,000-token scaling; silence and transfer billing Retell AI https://www.retellai.com/pricing 22 November 2026
Retell worked example: $0.055 + LLM + voice, telephony $0.015; prompt scaling formula Retell AI https://docs.retellai.com/accounts/billing-exceptions 22 November 2026
Retell numbers via Twilio, Telnyx or Bandwidth; US and Canada; calls to 15 countries Retell AI https://docs.retellai.com/deploy/purchase-number 22 November 2026
Retell testing methods; custom LLM agents audio-only for testing Retell AI https://docs.retellai.com/test/test-overview 22 November 2026
Retell testing at production rates, Conductor 30 / 200 free credits a day Retell AI https://docs.retellai.com/test/testing-pricing 22 November 2026
Retell custom LLM: you take on latency, uptime and reconnection Retell AI https://docs.retellai.com/integrate-llm/overview 22 November 2026
Retell Dialer: sequences, DNC screening, local presence Retell AI https://docs.retellai.com/dialer/overview 22 November 2026
Retell BAA self-signed, no additional fee Retell AI https://docs.retellai.com/general/compliance 22 November 2026
Retell enterprise deployment: cloud, VPC or on-prem (article dated 21 July 2025) Retell AI https://www.retellai.com/blog/enterprise-ai-calling-security 22 November 2026
$8,458 a month framework engineering load (assumption) Zian AI https://zian.ai/voice-agent-platform-vs-framework-when-to-switch/ 22 November 2026
All bills, gaps, owner-hours and break-even volumes (our arithmetic on the figures above) Zian AI (this page) — 22 November 2026

Frequently asked questions

Is LiveKit cheaper than Retell AI?

On published prices, yes. Our worked bill for a US inbound line at 15,000 minutes a month is $875.13 on LiveKit Cloud’s Ship plan, or $955.13 with recordings and observability, against $1,759.00 on Retell pay as you go, as at 22 November 2026. The gap leaves out the engineering time to write and run a LiveKit agent, which can cost more than the saving.

Can LiveKit phone numbers make outbound calls?

Not yet. LiveKit’s phone numbers documentation says LiveKit Phone Numbers currently only support inbound calling and that outbound support is coming soon. LiveKit agents can still place outbound calls over a SIP trunk from a carrier such as Twilio, Telnyx or Plivo, or on a Twilio number through LiveKit’s Twilio Connector. Retell sells US and Canada numbers that make and take calls.

Can I self-host LiveKit or Retell AI?

LiveKit, yes: the Agents framework, media server and SIP bridge are open source under the Apache License 2.0, and LiveKit documents self-hosted deployments. Retell, on its Enterprise plan only: its pricing page lists a dedicated stable server as an Enterprise add-on, and its own enterprise security article says enterprises can choose cloud, VPC or on-prem deployment, as at 22 November 2026. Its self-serve pay-as-you-go plan runs on shared servers.

Do LiveKit and Retell AI have simulation testing?

Both do. LiveKit Agent Simulations run in text and audio modes, at $0.02 a text turn and $0.05 an audio turn on paid plans. Retell runs simulation and batch tests in text at its chat rates, plus web and phone test calls at production rates, but its custom LLM agents can only be tested by web or phone call.

How do LiveKit and Retell AI bill a minute?

LiveKit Cloud meters agent sessions per second with a 10-second minimum per session, then adds separate lines for phone minutes, models and observability. Retell tracks each call to the nearest second and adds Voice Infra at $0.055, the model, the voice and telephony, billing 10 seconds for very short calls with a dynamic greeting.

When should I build on LiveKit instead of buying Retell AI?

When the monthly saving, divided by your hourly engineering cost, buys more hours than owning the agent takes. On our stack at US$125 an hour that is about 6 to 7 hours a month at 15,000 minutes, 28 hours at 50,000 and 80 to 87 hours at 150,000.

Related Blogs

Related from Zian AI