.webp)
Key takeaways
• Agora is a great product, not the only product. SD-RTN delivers reliable global low-latency real-time at managed-SDK convenience. The 2026 question is not whether Agora works — it does — but whether the price, lock-in, and feature shape still match your trajectory.
• The crossover is roughly 100k–500k minutes/month. Below that, managed (Agora, LiveKit Cloud, Daily) wins on time-to-market. Above 500k/month with predictable shape, custom WebRTC on a managed-cloud SFU often saves 30–40% on a 3-year horizon.
• LiveKit changed the math in 2025. Managed SFU without Agora-class lock-in — HIPAA/SOC2 at the Scale tier, AI agent framework built in, $0.0005/min WebRTC. For most new builds in 2026 it’s the “default Agora alternative.”
• Lock-in is the underrated risk. Agora’s RTM signaling, proprietary tokens, and SDK-tight mobile bindings make migration a real project. Plan a dual-SDK abstraction layer if you want optionality — especially before you ship a regulated or government-adjacent product.
• The honest answer is hybrid. Use Agora for what it’s great at (global mobile, managed AI add-ons, fast launch); migrate or build custom for high-volume, regulated, or specialized lanes. We do both at Fora Soft and we’ll tell you which fits your case.
Why the WebRTC vs Agora choice matters in 2026
Real-time video has moved from niche to commodity. WebRTC is in every browser. SDKs from Agora, LiveKit, Daily, 100ms, Twilio, and Vonage compete on price and AI features. Self-hosted SFUs — MediaSoup, Janus, Pion, ion-sfu — are production-grade and often run on commodity hardware that costs less per gigabit than a Starbucks budget.
The decision a CTO faces in 2026 is not “which video tech?” It’s “what fraction of my product runs on a managed SDK and what fraction runs on infrastructure I control?” This article gives you the numbers, the trade-offs, and the migration recipes to answer that. Built from real engagements at Fora Soft — we ship both Agora-based products and fully custom WebRTC stacks every quarter.
Why Fora Soft wrote this playbook
Fora Soft has built real-time video products since 2005 — over 600 of them across WebRTC, Agora, LiveKit, MediaSoup, Janus, Wowza, RTMP, SRT, and now MoQ. We’re Agora-certified and an Agora development partner. We’re also a long-standing custom-WebRTC shop. We use spec-driven agent engineering to compress timelines on either path.
Three projects ground this playbook. BrainCert is a WebRTC virtual-classroom LMS at $3M revenue and 100k+ customers, four-time Brandon Hall winner. Scholarly went from prototype to 15,000 users and 2,000 concurrent live students on Go + LiveKit + Kubernetes — the AWS Most Innovative EdTech finalist. Alve Live shipped 100k+ Android downloads and per-minute monetised 1:1 video using Janus + WebRTC. We bring that breadth into every recommendation below.
Choosing between Agora and a custom WebRTC build?
Bring your monthly minute volume, latency target, and regulatory constraints. We’ll model both options and give you a delivery estimate on a 30-min call.
The 60-second answer
Agora is a closed managed real-time platform with a proprietary global network (SD-RTN), built-in AI noise suppression and transcription, and SDKs across every platform. WebRTC + custom SFU is the open-standards alternative: you run MediaSoup or LiveKit (managed or self-hosted), integrate the AI features you need, and own the infrastructure. LiveKit Cloud sits between the two: managed SFU on open standards, AI-agent-native, transparent pricing.
Pick by minute volume, regulatory needs, and how much engineering you want to spend on infrastructure. The decision-tree at the bottom of this article boils it down to five questions.
Pricing in 2026: managed providers and DIY
| Provider | Audio | Video HD | Notes |
|---|---|---|---|
| Agora | $0.99 / 1k min | $3.99 / 1k participant-min | 10k free/mo; recording, RTM, AI add-ons billed separately. |
| LiveKit Cloud | ~$0.30 / 1k min | $0.50 / 1k WebRTC min | $50/mo Build, $500/mo Scale (HIPAA/SOC2/FedRAMP). |
| Daily | ~$4 / 1k participant-min | $4 / 1k participant-min | 10k/mo free; transparent flat rate. |
| 100ms | ~$2–$4 / 1k min | $2–$4 / 1k min | AI-agent-focused. |
| Twilio Programmable Video | — | Sunset | Migration is mandatory; Twilio recommends LiveKit/Daily. |
| Custom SFU on Hetzner | N/A | ~$500–$2k / mo for 500k min | No per-GB egress; you own the ops. |
Read the table with two eyes. Per-minute pricing dominates the marketing. The hidden costs — recording at $/GB, RTM messaging, AI add-ons, regional egress — routinely double a back-of-envelope managed bill at scale. Custom SFU on Hetzner-class hardware is dramatically cheaper at the bandwidth-heavy end because outbound traffic is included; the offset is the ops engineering you have to staff.
Reach for managed when: minute volume is unpredictable, you don’t have a media-ops team, and time-to-market matters more than per-unit cost. That’s most products at launch.
The crossover point: when does DIY beat managed?
Industry consensus puts the crossover between managed and custom SFU at around 100k–500k minutes per month, depending on the cost shape of your minutes (lots of small calls vs. fewer long ones), how much you depend on the AI features that ship in the SDK, and your team’s appetite for media-ops.
| Scenario (500k HD min/month) | Monthly cost | Year-1 + ops |
|---|---|---|
| Agora HD video | ~$2,000 + add-ons | ~$24k+ |
| LiveKit Cloud (Build/Scale) | ~$300–$1,200 | ~$3.6k–$14.4k |
| Custom SFU on Hetzner + 0.5 FTE ops | ~$500–$1,000 infra | ~$50k–$80k all-in |
| Custom SFU on AWS + 1 FTE ops | ~$2k–$5k infra (egress hurts) | ~$130k+ |
Two pivots in this table. First, LiveKit Cloud often beats both Agora and DIY on TCO when AI agents and HIPAA matter. Second, the choice of provider for the underlying compute is the most important DIY decision: Hetzner’s included-traffic policy makes it the budget winner, but it’s not where you put a US-government workload. Match infrastructure to compliance, not to the dashboard.
Latency: SD-RTN vs custom regional WebRTC
Agora’s SD-RTN is a globally distributed software-defined real-time network with 200+ datacenters and proprietary routing optimisation. Agora’s claimed average global edge latency is 76 ms; worst-case under 400 ms. That’s the magic Agora is selling: predictably fast no matter where the participants are.
Custom WebRTC SFU clusters in well-engineered deployments hit sub-100 ms regional and 100–250 ms multi-region. The gap between Agora and a thoughtful custom build is therefore 50–150 ms in the worst case — meaningful for esports and live shopping, less so for collaboration or telehealth. LiveKit Cloud, which runs across multiple regions and uses standard WebRTC under the hood, hits similar numbers to a well-built custom cluster.
Reach for SD-RTN (Agora) when: participants are spread across emerging-market mobile networks where last-mile latency dominates — that’s where Agora’s routing wins materially.
Features: what comes in the box, what you bolt on
| Capability | Agora | LiveKit Cloud | Custom WebRTC SFU |
|---|---|---|---|
| AI noise suppression | Built-in | Krisp/AI-coustics partnership | Integrate Krisp / open models |
| Real-time transcription | Built-in (premium) | Speechmatics / AssemblyAI / OpenAI realtime | Whisper / vendor-agnostic |
| AI agent / LLM voice | Conversational AI Engine (2025) | Native Agents framework | Build with OpenAI / Claude / Gemini SDKs |
| Real-time messaging (RTM) | Built-in | Data channels | MQTT / Redis Pub/Sub / custom |
| Recording & storage | Cloud + on-prem (paid) | Included on Scale tier | FFmpeg pipeline + S3 |
| Compliance (HIPAA/SOC2/FedRAMP) | Available, premium | Scale tier ($500/mo) | Yours to design |
| Mobile SDKs (iOS/Android/Flutter/RN) | Mature, opinionated | Open, well-maintained | Standard libwebrtc + your code |
Vendor lock-in: the hidden line item
Agora’s strongest hand and biggest risk are the same: the SDK is opinionated. RTM signaling is proprietary. Token formats are Agora-specific. Native mobile bindings often hook into platform-specific code that doesn’t map to other providers without rewriting. The result is real switching cost — we typically scope an Agora→LiveKit migration at 4–8 weeks, faster (around 3 weeks) on simple cases with agent engineering.
WebRTC + LiveKit is the more portable choice. Standard JSEP, standard SDP, standard TURN; swap LiveKit for a self-hosted MediaSoup cluster without rewriting your client. We don’t advise founders to architect for a hypothetical migration on day one — you’ll over-invest — but we do recommend a thin signaling abstraction layer in the client so a change of backend isn’t a reskin.
Stuck on Agora pricing? Curious about LiveKit?
We’ve migrated teams off Agora to LiveKit and to custom WebRTC. Bring the use case; we’ll model the migration cost and timeline on a 30-min call.
Compliance and geographic constraints
Agora is GDPR-compliant, has US data-residency options, and historically the strongest China connectivity story among Western-listed providers. China’s Cybersecurity Law amendment (effective January 1, 2026) and Network Data Security Management Regulations strengthen in-country requirements; Agora’s in-country deployment is a differentiator if you serve mainland China at scale.
For US healthcare and government, LiveKit’s Scale tier ($500/month) ships with HIPAA, SOC 2, and FedRAMP options. Custom WebRTC inherits your own compliance posture — and that’s the right answer when residency rules forbid third-party media servers (military, classified-adjacent, certain healthcare and pharma).
Reliability and SLAs
Agora claims 99.9% uptime. Real status-page history shows occasional regional incidents (China-region RTC degradation in early 2025, brief Pakistan and EMEA dips in late 2024). LiveKit Cloud and Daily run similar managed-cloud SLAs — comparable reliability when you exclude niche regions.
Custom WebRTC reliability is bounded by your ops discipline. A multi-region MediaSoup cluster with health checks, region failover, and predictive capacity hits managed-cloud-level uptime when designed by a team that’s done it before. The risk-adjusted answer for most teams: managed for the first 12–18 months, custom only when minute volume justifies the operational hire.
Concrete benchmarks
| Metric | Agora | LiveKit Cloud | Custom WebRTC |
|---|---|---|---|
| Avg global edge latency | ~76 ms (claim) | ~100–200 ms | ~100–250 ms |
| Worst-case global | <400 ms (claim) | ~400–600 ms | ~500–800 ms |
| SDK ecosystem | All major; opinionated | All major; open | libwebrtc + your work |
| China connectivity | In-country deployment | No native presence | Plan and pay for it |
| Time-to-MVP for video calling | ~1–2 weeks | ~2–3 weeks | ~6–10 weeks |
A decision framework — choose your stack in five questions
1. What’s your monthly minute volume in 12 months? Under 100k — managed (Agora, LiveKit, Daily) wins on TCO. 100k–500k — LiveKit Cloud is usually the value sweet spot. Above 500k with predictable shape — custom or hybrid.
2. Where are your users? Heavy mainland-China presence — Agora has the cleanest in-country story. Western markets only — LiveKit, Daily, or custom is fine.
3. What compliance bar applies? HIPAA, SOC 2, FedRAMP — LiveKit Scale or custom on a compliant cloud. ITAR / sovereign — custom on infrastructure you own.
4. How much of the AI feature stack do you need shipped on day one? Turnkey AI noise suppression and STT now — Agora or LiveKit Agents. Bring-your-own model, fine-tuned for domain — custom WebRTC + open AI stack.
5. How big is your engineering team? Two founders and a contractor — managed. Five real engineers with one media-ops lead — custom or LiveKit self-hosted is reasonable.
Migration recipes
Agora → LiveKit (managed). Typical timeline: 4–8 weeks; ~3 weeks on simple cases with agent engineering. Hard parts: signaling rewrite (Agora RTM → LiveKit data channels), token regeneration, mobile SDK swap. Run side-by-side for 2–4 weeks before cutover.
Agora → custom WebRTC SFU. Typical timeline: 8–12 weeks. Stand up MediaSoup or Janus on Hetzner / GCP / AWS; design signaling with WebSocket or MQTT; integrate AI services (Whisper, Krisp); add observability and failover. Plan for a 4–6-week parallel run before decommissioning Agora.
Twilio Programmable Video → anything. Twilio is sunsetting Programmable Video, so this isn’t optional — pick LiveKit or Daily for fastest migration; pick custom if you have a media team and a strong cost case.
Reach for a parallel run when: the migration touches paying customers. Cut over a small percentage of traffic first, watch the metrics, then graduate — never flip the switch on launch day for everyone at once.
Mini case: BrainCert — virtual classroom from prototype to $3M revenue
Situation. A virtual-classroom LMS built on a generic SFU with growing reliability and scaling pain — 100k+ customers, four Brandon Hall awards in sight, but the live-class experience was hitting limits during peak hours.
What we built. We re-architected the video stack on a custom WebRTC SFU mesh, with regional cluster placement, adaptive bitrate, recording, and a thin signaling abstraction so future changes wouldn’t require client rewrites. We layered AI features (transcription, attentional analytics) using vendor-agnostic services so the LMS team could swap providers without touching the SFU.
Outcome. $3M revenue, 100k+ customers, four Brandon Hall awards. The architecture scaled with the business and the per-minute cost dropped enough to make the AI tier sustainable. Read the full BrainCert project page or book a 30-min review if you’re sitting on similar growing-pain numbers.
Five pitfalls we see teams hit
1. Buying Agora on day one without modelling year-three minutes. The price escalates with success. Forecast on the trajectory, not the launch.
2. Picking custom WebRTC because “it’s cheaper” without budgeting ops. Cheaper compute is not cheaper engineering. A 1 FTE media-ops lead is roughly $120k–$200k loaded.
3. Hard-coding to Agora-RTM for signaling. Future migration becomes a rewrite. Use a thin signaling abstraction from day one.
4. Skipping a load test before launch. Whether managed or custom, your specific use case has a unique cost curve. Test it on real traffic before the press release.
5. Treating AI features as commodities. Quality of noise suppression and transcription varies a lot by domain. Benchmark on your audio — not on the vendor demo.
Reach for a hybrid stack when: different lanes of the product have different needs — Agora for the conversational AI agent shipping in 4 weeks; custom WebRTC for the high-volume webinar lane that ships in 8.
KPIs to measure
Quality KPIs. P50 / P99 glass-to-glass latency by region. Mean opinion score on audio (target above 4.0 on the 5-point scale). Subjective video quality score after each session.
Business KPIs. Cost per minute all-in (managed bill + AI add-ons + storage + ops). Concurrent participants peak. Cost per acquired customer attributable to the call experience.
Reliability KPIs. Uptime against SLA. Reconnect rate per session. Time to recover after a regional incident.
When NOT to leave Agora
If your minute volume is under 100k a month, your engineering team is small, your users are scattered globally on emerging-market mobile, and your AI feature shopping list aligns with what Agora ships natively — stay. The math doesn’t favour migration; the engineering you’d spend rewriting signaling and replacing the SDK is better spent on product.
If any of those changes — you cross 500k minutes, you hire a media-ops engineer, your audience consolidates regionally, you need a custom AI feature — that’s when the conversation gets interesting.
Want a build-vs-buy verdict on your numbers?
We’ll model Agora, LiveKit, Daily, and custom WebRTC against your minute forecast and give you a written verdict in a 30-min call.
FAQ
Is Agora better than custom WebRTC?
It’s better at being a managed SDK with global mobile-first reach and bundled AI features. Custom WebRTC is better at giving you control, lower per-unit cost at scale, and freedom from lock-in. The right answer depends on minute volume, geography, compliance, and engineering capacity.
When does custom WebRTC pay back vs Agora?
Roughly 100k–500k minutes per month, depending on cost shape and how heavily you use Agora’s AI features. Below that band, managed wins on TCO. Above 500k with predictable shape, custom typically saves 30–40% over a 3-year horizon.
Should I pick LiveKit Cloud over Agora?
For most new builds in 2026, yes — LiveKit gives you managed SFU on open standards, an AI-agent framework, and HIPAA/SOC 2/FedRAMP at the Scale tier. Agora keeps the edge for mainland-China deployment and for teams already standardised on its SDKs.
How long does an Agora → LiveKit migration take?
Typically 4–8 weeks for a small-to-mid SaaS, with the hard parts being signaling differences (RTM → data channels), token rewrites, and any custom mobile-SDK code. With agent engineering on simpler cases we’ve closed migrations in around 3 weeks.
What replaces Twilio Programmable Video?
LiveKit and Daily are the most common drop-in replacements; Agora works if you need its AI features and global SD-RTN. Custom WebRTC is the right choice when minute volume justifies it. Either way, sit down and migrate — Twilio Video is end-of-life.
Does Agora work in mainland China without a separate setup?
Better than most Western providers. Agora has in-country deployment to comply with the Cybersecurity Law and Network Data Security Management Regulations. LiveKit and Daily don’t have native China presence; custom WebRTC requires a Chinese hosting partner and the operational overhead that comes with it.
Is Hetzner safe for production video infrastructure?
For commercial workloads, generally yes. The included-traffic policy is the cost lever. For US healthcare, government, or anything HIPAA / FedRAMP regulated, you’ll want AWS, GCP, Azure, or a sovereign cloud — not because Hetzner is unsafe, but because the compliance attestations don’t match.
How long does a Fora Soft engagement take?
A working pilot on Agora or LiveKit lands in 2–4 weeks. A custom WebRTC SFU pilot lands in 6–10 weeks via spec-driven agent engineering. A full Agora migration is typically 4–8 weeks. Bring us the scope and we’ll quote on a call.
What to Read Next
Scale
Scale Real-Time Video to 1 Million Viewers
WebRTC, LL-HLS, and MoQ at six- and seven-figure audiences.
MoQ deep dive
Building Applications with Media over QUIC
The protocol that closes the gap between WebRTC latency and CDN economics.
Hiring
Hire a WebRTC Development Company vs Build In-House
A buyer’s guide for streaming and real-time video founders.
Provider deep dive
Telnyx WebRTC Development 2026
When Telnyx wins, how to architect, what it costs.
Build vs buy
Wowza Custom Development in 2026
A build-vs-buy analysis for low-latency streaming platforms.
Ready to pick the right real-time stack?
Agora is a great managed platform. LiveKit and Daily are great managed alternatives. Custom WebRTC is great when scale, control, or compliance demand it. The honest answer for most products in 2026 is to start managed, design for portability, and migrate when the math says so — and to consider hybrid stacks where different lanes of the product use different providers.
If you’re scoping a real-time product, the technology choices are well-understood; matching them to your minute volume, geography, compliance, and AI shopping list is the work. That’s the conversation we have with prospective clients on a 30-min scoping call — bring the constraints, leave with an architecture and a delivery estimate.
Talk to a team that has shipped 600+ video products on every stack
Agora-certified, LiveKit-fluent, custom WebRTC for the long-tail. We don’t favour one over another; we pick what fits your case.
.avif)
Comments