Self-hosted power vs
managed simplicity.
LiveKit gives you full infra control and a path to dramatically lower TCO at scale. Retell gets you to production in days without an ops team. Here's how to choose.
We build production voice AI on both. This is our honest take.
TL;DR
Best for: teams without dedicated voice infra engineers, healthcare (HIPAA), SMB and mid-market deployments, and anyone who needs to ship a production voice agent in days not weeks.
- ✓$0.07/min fully managed (STT included)
- ✓HIPAA BAA available
- ✓Visual drag-and-drop flow editor
- ✓Built-in analytics and transcript search
Best for: engineering-heavy teams, high-volume deployments (50k+ min/month) where self-hosting drops TCO significantly, multi-modal use cases (voice + video), and data-sovereign deployments.
- ✓Open-source, self-hostable (Apache 2.0)
- ✓Full WebRTC + SIP infrastructure control
- ✓Video + voice in one SDK
- ✓Multi-participant rooms natively
The Real Pricing Math
What You Actually Pay Per Minute
LiveKit Cloud and Retell Cloud both use BYOK (bring your own keys). But LiveKit self-hosted can eliminate the platform fee entirely — at the cost of infra ops. Here's the full math.
| Cost component | LiveKit Cloud | LiveKit Self-Host | Retell |
|---|---|---|---|
Platform fee Self-host replaces platform fee with EC2/server costs | ~$0.04–0.06 / min | $0 (infra cost) | $0.07 / min |
STT (Deepgram) Retell bundles STT; both LiveKit options charge separately | ~$0.008 / min | ~$0.008 / min | Included above |
LLM (GPT-4o mini, BYOK) Same cost on all three — depends on turn length | ~$0.02–0.05 / min | ~$0.02–0.05 / min | ~$0.02–0.05 / min |
TTS (ElevenLabs Flash, BYOK) Identical — you bring your own key | ~$0.10–0.15 / min | ~$0.10–0.15 / min | ~$0.10–0.15 / min |
Infra (EC2 / server) Self-host infra cost varies with instance type and utilisation | Included | ~$0.01–0.03 / min | Included |
Effective platform + STT Self-host LiveKit can be the cheapest at scale | ~$0.048–0.068 / min | ~$0.018–0.038 / min | $0.07 / min |
Typical all-in total Self-host wins TCO at 50k+ min/month | $0.21–0.33 / min | $0.14–0.25 / min | $0.21–0.30 / min |
5,000 min / mo
Retell wins on simplicity at low volume
50,000 min / mo
Self-host starts pulling ahead here
200,000 min / mo
Self-host can save $10–20k/mo
Estimates based on 2026 pricing. LLM cost assumes ~150 tokens/turn at GPT-4o mini rates. TTS assumes ElevenLabs Flash Turbo v2. Self-host infra assumes t3.xlarge-class instances at 60% utilisation. Actual costs vary.
Feature Comparison
What Each Platform Does and Doesn't Do
| Feature | LiveKit | Retell |
|---|---|---|
Self-hosted option LiveKit is open-source and can run on your own infra; Retell is fully managed | ✓ | ✗ |
Fully managed cloud LiveKit Cloud and Retell Cloud both available; Retell is managed-only | ✓ | ✓ |
Visual flow / no-code editor Retell has a drag-and-drop node editor; LiveKit requires code | ✗ | ✓ |
WebRTC & SIP support Both support WebRTC; LiveKit SIP is deeper and more configurable | ✓ | ✓ |
Video + voice in one SDK LiveKit handles video, screenshare, and voice in one SDK; Retell is voice-only | ✓ | ✗ |
HIPAA BAA available Both can sign a BAA; self-hosted LiveKit requires your own infra compliance posture | ✓ | ✓ |
BYOK LLM Both support GPT-4o, Claude, Gemini, custom endpoints | ✓ | ✓ |
BYOK STT Both support Deepgram, AssemblyAI, and others; Retell bundles Deepgram in platform fee | ✓ | ✓ |
BYOK TTS Both support ElevenLabs, Cartesia, PlayHT, Azure | ✓ | ✓ |
Native analytics dashboard Retell has built-in call analytics; LiveKit requires custom observability tooling | ✗ | ✓ |
Real-time interruption handling Both handle barge-in; LiveKit exposes more low-level control | ✓ | ✓ |
Multi-participant rooms LiveKit supports N-party rooms natively; Retell is 2-party (agent + caller) | ✓ | ✗ |
Knowledge base / RAG Retell has built-in document upload and retrieval; LiveKit requires custom RAG pipeline | ✗ | ✓ |
Call recording Both record calls; Retell has native transcript search UI | ✓ | ✓ |
Open-source codebase LiveKit core is Apache 2.0 licensed; Retell is fully proprietary | ✓ | ✗ |
Volume Crossover
The 50k-Minute Crossover Point
Use Retell
SMB, mid-market, healthcare, and early-stage deployments
- →No infra ops overhead — Retell manages everything
- →Visual flow editor means non-engineers can iterate on agent logic
- →HIPAA BAA out of the box — no custom compliance work
- →Built-in analytics dashboard covers most reporting needs
Evaluate LiveKit Self-Host
Enterprise scale, video + voice, and data-sovereign deployments
- →Platform fee savings can exceed $5–15k/month at this volume
- →Full WebRTC and SIP stack control for complex routing
- →Video + voice rooms in one SDK — critical for telehealth, co-browsing
- →Data residency on your own cloud account — EU/UK data sovereignty
HIPAA & Compliance
Healthcare Compliance Posture
Both platforms can support HIPAA workloads, but the path is different. Retell offers a signed BAA on managed infrastructure. LiveKit on self-hosted infra means your team owns the compliance posture — more control, more responsibility.
- →Signed BAA available on request
- →Infrastructure managed and audited by Retell
- →Built-in audit logging for compliance reporting
- →Faster path: no custom infra compliance work required
- →BAA can be signed with LiveKit Cloud
- →Self-hosted: your infra, your compliance posture
- →Open-source means you can audit every line of code
- →More setup work but more control over data residency
The Decision Tree
How We Choose Internally
This is the actual decision flow we walk clients through. Answer yes to the first question that applies.
Do you need video, screenshare, or multi-participant rooms alongside voice?
Yes
Use LiveKit
It is the only platform that handles all three in one SDK
No
Continue ↓
Is data sovereignty or on-premises deployment a hard requirement?
Yes
Use LiveKit
Self-host on your own infra; full control over data residency
No
Continue ↓
Does your team lack dedicated voice AI engineers?
Yes
Use Retell
Visual flow editor — most agents built without writing code
No
Continue ↓
Under 50k minutes / month with healthcare or SMB use case?
Yes
Use Retell
Lower operational burden, HIPAA BAA, built-in analytics
No
Continue ↓
Over 50k minutes / month and comfortable managing infra?
Yes
Evaluate LiveKit self-host
TCO can drop significantly vs managed platforms at scale
No
Continue ↓
None of the above apply?
Yes
Use Retell
Default choice — managed, lower ops overhead, faster to ship
No
Continue ↓
Final default
When in doubt, start with Retell.
Lower ops burden, faster to ship, HIPAA managed for you. Switch to LiveKit when volume justifies the infra investment.
Our Role
We Build on Both. Here's How We Engage.
Hestur AI is a platform-agnostic voice AI engineering firm. We don't resell LiveKit or Retell and have no preferred-partner agreements with either. We recommend the platform that fits your use case and volume — then build on it.
Retell engagements
Typically healthcare, insurance, and SMB deployments where speed to production matters and a non-engineering team will manage the agent ongoing. We configure HIPAA-eligible infrastructure and hand off cleanly.
Common scope: patient intake, appointment scheduling, inbound triage, customer support deflection.
LiveKit engagements
Typically enterprise clients with an in-house engineering team, high call volume (50k+ min/month), multi-modal requirements (voice + video), or strict data residency requirements where self-hosting on their own cloud is the only acceptable option.
Common scope: telehealth platforms, custom SIP routing, multi-participant call centres, EU/UK data-sovereign deployments.
Not Sure Which Platform Fits?
Book a 30-minute call. We'll walk through your use case, volume expectations, and compliance requirements — and give you a clear recommendation with no pitch attached.