Video chat app build versus buy decision framework with technical architecture comparison

The question isn’t whether to buy Zoom’s SDK or build it all yourself. In 2026, how to build a video chat app is really a choice between four distinct paths — white-label, SDK, open-source self-host, or fully custom — each with a clean break-point where the next one pays off. Short answer: for most teams in 2026, video chat app development means picking a managed SDK (LiveKit Cloud is our default), building your own UX and backend on top of it, and shipping an MVP in 8–12 weeks; white-label gets you live in days, self-hosting pays off only at high volume, and raw WebRTC is a niche. This guide gives you the 90-second answer, the decision framework, the build steps, and the unit economics at October 2026 list prices.

We’ve shipped more than 40 video products at Fora Soft — on Agora, Twilio, LiveKit, mediasoup, and custom WebRTC, from the iMind enterprise conferencing platform to MyOnCallDoc HIPAA telemedicine video. Every few quarters a founder asks the same question in a different framing: “Should we buy a white-label platform, pick an SDK, or build our own?” There’s no universal answer. There is a framework that makes the choice obvious in 20 minutes — and if you’d rather run it on your own numbers with us, book a 30-minute architecture call.

Key takeaways

• Buy white-label (Whereby Embedded, Daily Prebuilt) if video is adjacent to your product, you need to ship in 4–8 weeks, and a room rarely needs more than ~24 live video tiles.

• Buy an SDK (LiveKit Cloud, Agora, Zoom SDK) if video is core but the codec isn’t your moat. You own the UX; the vendor owns the media plumbing.

• Self-host open source (LiveKit, mediasoup) only past ~50M participant-minutes/year against $0.004/min SDKs (several times that against LiveKit Cloud list prices) and with 3+ engineers on hand; below that, managed is cheaper once you price the on-call tax.

• Fully custom WebRTC is almost never the answer in 2026 — the narrow cases are FDA-regulated medical devices, airgapped government, or custom codecs.

• The 2026 differentiator is AI, not latency. Real-time translation, meeting agents, and live avatars now separate winners from commoditized incumbents.

Updated October 2026: re-verified list prices for 8 video platforms (Whereby, LiveKit, Agora, Daily, 100ms, Zoom, Twilio, VideoSDK) and redrawn cost curve; new build steps and SDK-embedding guide; LiveKit’s 2026 Series C, HeyGen’s move to LiveAvatar, and the HIPAA Security Rule delay to 2027.

TL;DR — the 90-second answer

Match your situation to one of these four lines and stop reading:

  • Video is a feature, not the product. You need 1-on-1 or small-group video inside an existing app, you’ll ship in 6 weeks, and HIPAA or GDPR is already on the spec. → Buy white-label. Whereby Embedded or Daily Prebuilt. You’re done by Tuesday.
  • Video is core, but not the moat. You need custom UX, analytics, your own brand, maybe AI agents in the call, but you don’t want to run media servers. → Buy an SDK. LiveKit Cloud is the safest 2026 default; Agora if you need Asia-Pacific scale; Zoom SDK if your enterprise buyer is already Zoom-native.
  • Video is the product and you’ll outgrow managed pricing. You have 3+ engineers, you’re past ~50M participant-minutes/year on a per-minute SDK, and you need something managed platforms won’t give you (E2EE with recording, custom codecs, full data residency). → Self-host on LiveKit or mediasoup.
  • You’re in a regulated, airgapped, or hardware-integrated niche. FDA, defense, broadcast, real-time medical imaging. → Custom WebRTC. And even then, start with LiveKit and fork only what you must.

The rest of this article proves why, with 2026 pricing, a decision framework, five vertical playbooks, and the eight mistakes we see founders repeat every year.

The four paths in 2026

“Build vs buy” is a false dichotomy. The real answer to how to build a video chat app in 2026 is four distinct paths with very different ownership profiles, ops loads, and break-even points:

Path What you get What you own Time-to-ship
1. Buy white-labelFinished room UI in iframe/SDK. Whereby, Jitsi-as-a-Service, Sendbird, Daily Prebuilt.Room creation, access control, webhooks.Days to 2 weeks
2. Buy an SDKPrimitives (rooms, tracks, participants). LiveKit Cloud, Agora, Daily, 100ms, Zoom SDK.UX, signaling logic, analytics, brand, integrations.4–12 weeks
3. Self-host open sourceFull SFU code. LiveKit OSS, mediasoup, Janus, Pion.Everything above + media servers, scaling, codecs, security patches.10–20 weeks
4. Custom WebRTCRaw browser/mobile WebRTC APIs.Literally everything — signaling protocol, NAT traversal, codec ladder, congestion control.6–18 months

The four paths, ordered by control — and by ops load.

Four paths to build a video chat app, from white-label to custom WebRTC, ranked by ownership and ops load

Figure 1. The four build-vs-buy paths for a video chat app, ordered by how much of the stack you own.

Each path has a natural break-point where the next one pays off. The break-points are not about revenue — they’re about concurrent load, feature requirements, and compliance ceiling. Most products live in path 1 or 2 for their entire life; roughly a third eventually cross into path 3. Path 4 is rarer than founders think. Our video chat app development team works on all four paths, from a white-label MVP to a custom WebRTC build.

Eight questions that decide for you

Run these eight questions in order. The first “yes” that materially changes your answer is the one that decides the path. Most teams reach a clear recommendation by question four.

# Question If yes →
1Is video a feature inside an existing product, not the product itself?White-label
2Do you need a finished UI with branded skinning only, no custom layouts?White-label
3Do you need custom UX, analytics, AI agents in the call, or non-standard layouts?SDK
4Is video central to your moat (you sell quality, features, or latency as the product)?SDK or self-host
5Will you exceed ~50M participant-minutes/year with high margin sensitivity?Self-host
6Do you need E2EE + server-side recording simultaneously, or custom codecs?Self-host or custom
7Do you have 3+ senior engineers and tolerance for ~1 FTE of permanent ops?Self-host feasible
8Are you FDA-regulated, airgapped, or integrating with hardware (ultrasound, endoscope, broadcast)?Custom WebRTC

Run the eight questions in order; stop at the first material “yes”.

Decision flow for how to build a video chat app: four questions routing to white-label, SDK, self-host, or custom

Figure 2. The decision flow condensed to four branch points — stop at the first “yes” that changes your answer.

Roughly 60% of teams we advise land on path 1 or 2 after question 4. About 30% reach path 3 on question 5 or 6. Fewer than 10% genuinely need path 4.

How to build a video chat app: 7 steps

To build a video chat app, define the call types and scale, pick one of the four paths, then build signaling, rooms and UX on top of the chosen media layer, test on real networks, and launch with monitoring. On an SDK, an MVP takes 8–12 weeks; the steps are the same for 1-on-1, group and AI-assisted calls.

  1. Define the call. 1-on-1 or group, peak participants per room, web/iOS/Android, recording, and which compliance regime (HIPAA, GDPR) applies. These five answers decide 80% of the architecture.
  2. Pick the path. Run the eight questions above: white-label, SDK, open-source self-host, or custom WebRTC.
  3. Choose the media layer. For an SDK build, compare two or three vendors on your real geography and concurrency, not on the rate card alone (see the pricing table below).
  4. Build the app layer. Auth and user profiles, room creation and access tokens on your backend, waiting rooms, chat, screen share, notifications. This is where most of the engineering time goes.
  5. Add the differentiators. Recording to your own storage, transcription, AI note-taking or translation agents, integrations with your CRM, LMS or EHR.
  6. Test on bad networks. Throttle to 3G and lossy Wi-Fi, test on low-end Android, run load tests at 2× your expected peak, and check echo and noise on real headsets.
  7. Launch and watch QoE. Track join success rate, time-to-first-frame, packet loss and call drops per region from day one; these numbers tell you when to change path.

The MVP feature set rarely changes between products. What changes is who builds each piece:

MVP feature White-label Managed SDK Self-host / custom
1-on-1 and group videoIncludedSDK primitives + your UIYour SFU + your UI
Chat in the callUsually includedData channel or chat SDKYou build it
Screen shareIncludedSDK track; you build controlsYou build it
RecordingVendor storage and retentionVendor egress to your storageYour recorder service
Waiting room and access controlConfigurableYour backend issues tokensYour backend
AI notes or translationPost-call hooks at bestAgent joins the room (e.g. LiveKit Agents)Agent + your orchestration
Push notifications and call UI on mobileLimitedCallKit / ConnectionService on your sideSame

Who owns each MVP feature on each path.

What changed in video chat between 2024 and 2026

If your last build-vs-buy analysis is older than 18 months, throw it out. The market shifted on three axes:

1. Twilio Video un-sunset itself. In 2023 Twilio announced an end-of-life for Programmable Video. In October 2024 it reversed course, and as of 2026 Programmable Video is a standalone, supported product still taking new accounts — the Rooms, Recordings, and Compositions APIs remain on the roadmap. Twilio also partnered with Zoom for teams that still want to move, so the call is now “migrate or stay,” not “migrate or break.” The lesson survives the reversal: platform roadmaps whipsaw, so weight an open-source escape hatch heavily. We lay out the options in our Twilio Video migration guide.

2. LiveKit became the default “we’ll start here” platform. Two years ago it was a credible alternative. In 2026 it’s the first recommendation for most new builds because (a) the managed Cloud tier is priced competitively, (b) self-hosting is a real option you can fall back to, and (c) the Agents framework means AI voice and vision can join rooms as first-class participants.

3. White-label and SDK vendors made compliance a line item. HIPAA is now a priced add-on rather than a sales negotiation: Whereby Embedded sells it for $16.99/month on its $10.99 Build plan, while Daily and 100ms each list a $500/month HIPAA + BAA add-on (vendor pricing pages, October 2026). A 2023 decision to “build because white-label doesn’t do HIPAA” is often wrong in 2026.

4. AI is the new moat. Real-time translation, live captioning, meeting summarization, and avatar-based agents (Tavus, HeyGen’s LiveAvatar, Beyond Presence) are now the differentiator. Video quality is commoditized; AI integration isn’t. This pushes teams toward platforms with open agent frameworks (LiveKit) and away from closed stacks where AI is a premium add-on.

5. The money followed real-time AI. LiveKit raised a $100M Series C at a $1B valuation on January 22, 2026 (led by Index Ventures, per SiliconANGLE), with ChatGPT’s Advanced Voice Mode among its best-known customers. Agora moved new customers to a Free package with 10,000 monthly minutes on August 29, 2025, and Whereby’s Embedded entry plan now starts at $10.99/month. Per-minute list prices have converged around $0.004 per participant-minute for most managed SDKs, with LiveKit Cloud an order of magnitude lower on minutes but charging separately for bandwidth.

2026 market reality: the cheapest-per-minute SDK is rarely the cheapest total stack. The platform that lets you bolt AI, recording, and translation without switching vendors wins on 3-year TCO. That’s why LiveKit Cloud is our default recommendation even when Agora prints a lower per-minute line-item.

2026 pricing reference

Below are list prices we checked on each vendor’s pricing page in October 2026. Most managed SDKs now sit at about $0.004 per participant-minute for HD video; LiveKit Cloud is cheaper per minute but bills bandwidth on top. Enterprise contracts vary, and recording, transcription and dial-in add multipliers, so multiply any headline rate by 1.4–1.8 for the all-in cost.

Vendor Path List rate (Oct 2026) Notes
Whereby EmbeddedWhite-labelBuild: $10.99/mo incl. 2,000 min, then $0.0042/participant-minHIPAA add-on $16.99/mo; up to 200 participants, 24 active videos per room
Daily (Prebuilt or SDK)White-label & SDK$0.004/participant-min; audio-only $0.0009910,000 free min/mo; volume tiers down to $0.0015; HIPAA + BAA $500/mo
Sendbird CallsWhite-labelUsage-based; quoteChat + calls bundle; mobile-first; price on request
AgoraSDKHD $3.99, Full HD $8.99, audio $0.99 per 1,000 minFree package: 10,000 min/mo (new customers since Aug 29, 2025)
LiveKit CloudSDKShip $50/mo (150K min, then $0.0005); Scale $500/mo (1.5M min, then $0.0004)Bandwidth billed separately ($0.10–$0.12/GB); agent sessions $0.01/min; Build tier free
100msSDK$0.004/participant-min; audio-only 75% off10,000 free min/mo; HIPAA + BAA $500/mo on Developer plan
Zoom Video SDKSDK$0.0035/min pay-as-you-go10,000 free min/mo; enterprise-friendly
VideoSDK.liveSDKHD $0.004, Full HD $0.008, audio $0.001 per participant-min$20 one-time free credit
Twilio VideoSDK$0.004/participant-min (Group Rooms, up to 50)Standalone product since Oct 2024; recordings $0.004/min, compositions $0.01/min
LiveKit / mediasoup self-hostedOpen source~$0.0005–$0.0015/min infra (our estimate)Add ~1 FTE DevOps; break-even ~50M min/yr vs $0.004/min SDKs

Vendor pricing pages, checked October 26, 2026; enterprise and volume discounts are material. Verify before budgeting.

When to buy white-label (Whereby, Sendbird, Daily Prebuilt)

White-label means you embed a finished room UI (iframe or SDK) and configure it — you don’t design the video surface, you skin it. This is the right path when video is adjacent to your product.

Buy white-label when:

  • Video is a feature of a larger product — CRM consult rooms, LMS tutoring sessions, marketplace buyer-seller calls, HR interviews.
  • You need HIPAA or GDPR compliance out of the box without weeks of legal review. Whereby (a $16.99/month add-on) and Doxy.me offer BAAs off the shelf, Daily and 100ms list them as $500/month add-ons, and most other SDKs negotiate a BAA per contract.
  • Time-to-market is under 8 weeks, and your team has fewer than three full-stack engineers available for video work.
  • Concurrent participants per room stay under 50 and you’re not doing broadcast-style large events.
  • You accept the vendor’s UX, layout primitives, and recording retention policy without needing to extend them.

Where white-label breaks:

  • Concurrent ceiling. Whereby Embedded allows up to 200 participants but only 24 active videos per room, and Twilio Group Rooms stop at 50 (vendor pricing pages, 2026). If your roadmap includes webinars, all-hands meetings, or virtual events, plan to migrate.
  • UX rigidity. Custom grid layouts, picture-in-picture-with-whiteboard, “spotlight + reactions bar + inline polls” compositions — these are usually not possible without dropping to an SDK.
  • AI agents. Whereby/Sendbird don’t have native hooks for AI agents joining as a participant. You can bolt on post-call transcription, but live in-call AI needs an SDK (LiveKit Agents).
  • Recording retention and storage. White-label platforms usually enforce their own retention window (often 30 days). If legal needs 7-year retention or you want recordings in your own S3, pick an SDK.
  • Per-MAU pricing cliffs. Chat+calls bundles (Sendbird, Stream) are priced on active users; at 500K MAU you’ll be quoted enterprise pricing that often exceeds running an SDK stack yourself.

The typical white-label success pattern: a marketplace or EdTech product ships Whereby in six weeks, validates the feature with real users, then migrates to an SDK in year 2 once video volume and UX demands outgrow the iframe. That’s not a failed decision — it’s the correct staged rollout. Don’t over-engineer at launch.

When to buy an SDK (Agora, LiveKit Cloud, Zoom SDK)

An SDK gives you primitives (rooms, tracks, participants, publishers) and you compose the UX. The vendor owns the media servers, NAT traversal, geographic routing, codec negotiation, and uptime. You own everything above that line.

Buy an SDK when:

  • Video is core to the product — you need your own UX, your own analytics, your own brand in the room surface.
  • You need AI agents joining the call (live translation, note-taking, moderation). LiveKit Agents is the 2026 reference framework for this.
  • You need sub-200ms latency with global coverage, or specifically sub-100ms for real-time collaboration, remote control, or gaming-adjacent use cases.
  • You have 2–5 senior engineers who can own the video feature through launch and first 18 months of iteration.
  • You want to control where recordings land (your S3, your retention policy) and how transcription hooks into your product.

Which SDK in 2026:

SDK Pick when Watch out for
LiveKit CloudDefault choice. AI agents, open-source escape hatch, clean pricing.Smaller PoP footprint than Agora in APAC.
AgoraGlobal scale, APAC-heavy users, lowest latency SD-RTN.Pricing complexity, older UI kits, vendor lock-in.
DailyFastest integration for embedded conferencing, simplest API.Smaller ecosystem; voice agents run through Daily’s open-source Pipecat framework rather than in-room agents.
Zoom Video SDKEnterprise buyer is already Zoom-native; healthcare/legal verticals.Enterprise list pricing; Zoom-branded primitives bleed through.
100msIndian market, HIPAA/SOC 2 out of the box, good prebuilt UI.Smaller global footprint; community smaller than LiveKit.
VideoSDK.liveAggressive pricing, emerging markets.Smaller SLAs, community still maturing.

SDK selection matrix for 2026.

Our internal default for a new 2026 project is LiveKit Cloud, specifically because the Agents framework and open-source self-host path keep two doors open. Agora is our pick when users are concentrated in Asia-Pacific or when we need sub-100ms latency at massive scale. Zoom SDK wins when the enterprise buyer is already Zoom-standardized and insists on the familiar UX primitives.

Reach for an SDK when: video is core to the product, you want your own room UX and analytics, and at least one AI feature (translation, note-taking, an interviewer agent) is on the 12-month roadmap, but you have no appetite to run media servers at 2am.

Not sure which path matches your product?

Share your user geography, peak concurrency, and compliance constraints — we’ll map them onto Buy / SDK / OSS / Custom in 30 minutes.

Book a 30-min call → WhatsApp → Email us →

Embedding a video room SDK: iframe, UI kit, or raw SDK

An online video room SDK lets you put a ready video room inside your web or mobile app without running media servers. There are three ways to embed one, and they differ mostly in how much UI you own: a prebuilt iframe ships in a day, a UI-kit build in 1–2 weeks, and a raw-SDK build in 4–12 weeks.

Embed style Examples (2026) Time to first call You control
Prebuilt iframe / room URLWhereby Embedded, Daily Prebuilt, VideoSDK PrebuiltHours to a dayBranding, room settings, webhooks
UI kit / componentsLiveKit Components (React, Swift, Android, Flutter), 100ms Prebuilt, Zoom Video SDK UI Toolkit1–2 weeksLayout, flows, most of the UX
Raw SDKLiveKit, Agora, Daily, Twilio Video, Zoom Video SDK client SDKs4–12 weeksEverything above the media layer

How to embed a video room SDK; times are for a first production-quality call on web, from our projects.

For a mobile video calling app, the deciding factor is native call UX: CallKit on iOS and ConnectionService on Android, background audio, and push-to-wake. LiveKit, Agora, Twilio and Zoom all ship native iOS and Android SDKs; iframe-only products don’t give you that. To embed a live avatar in a web app, put the avatar in the same room as the user: LiveKit Agents has avatar plugins for providers such as Tavus and Beyond Presence, and our guide to interactive AI avatar development compares the options. If your stack started on Jitsi and you’re weighing a move, see our Jitsi alternatives shortlist.

When to build on open source (LiveKit self-hosted, mediasoup)

Open-source self-hosting gives you the full media stack, everything above the codec. It’s the right path when three things are true simultaneously: (a) your volume justifies the infrastructure investment, (b) you need a feature the managed SDKs don’t ship, and (c) you have the engineering bench to keep it running.

The three open-source options:

Platform Language Best fit Ops load
LiveKit OSSGoTeams already using Cloud; Kubernetes-friendly; has AI Agents out of box.Medium (~0.5 FTE)
mediasoupC++ / Node.jsTeams needing raw performance, custom recording/simulcast logic.High (~1 FTE)
JanusCSpecialty plugins (e.g., streaming medical imaging), research, WebRTC gateways.High (~1 FTE)

Ops load is real engineering overhead — not just server cost.

The rule of thumb we’ve calibrated across a dozen self-hosted deployments: against SDKs that charge about $0.004 per participant-minute (Agora HD, Daily, 100ms, Twilio), self-hosting gets cheaper from roughly 50M participant-minutes/year once you price in a DevOps FTE. Against LiveKit Cloud’s list price ($0.0004–$0.0005 per minute plus bandwidth, October 2026), the break-even moves out to roughly 300–400M participant-minutes/year. Below that line, managed wins on total cost. Founders routinely underestimate the FTE tax. You don’t just pay for servers, you pay for someone who can debug SRTP desync at 2am.

If you’re considering self-hosting, the right sequencing is usually: launch on LiveKit Cloud, prove volume, then migrate to LiveKit OSS when the ops load is justified. The code is the same, so the migration is mostly infrastructure provisioning rather than a platform swap. Our migration playbook article covers that path in detail.

When to build fully custom from WebRTC

Fully custom means: raw browser/mobile WebRTC APIs, your own signaling protocol, your own TURN/STUN, your own media server, your own codec ladder. It’s the right path in fewer than 10% of the “should we build?” conversations we have.

The narrow legitimate cases:

  • FDA-regulated medical imaging. Ultrasound-over-WebRTC with custom bitrate control, audit-grade logging, device-specific encoding. A Class II device needs documented control and validation of every third-party software component (SOUP under IEC 62304), which is easier when you own the media path.
  • Airgapped government or defense. No outbound internet, no third-party cloud, no unverified dependencies.
  • Broadcast-grade integration. Real-time integration with SDI hardware, vMix, or custom production pipelines where you need millisecond-accurate sync.
  • Custom codecs. Specialised audio codecs for hearing devices, or proprietary video codecs for bandwidth-constrained satellite links — the kind of work that starts with video encoding fundamentals, not an SDK.
  • Research and academic projects. Where the whole point is the WebRTC stack itself.

In every other case, starting from raw WebRTC means 12–18 months of reinventing what LiveKit or mediasoup already solved — ICE/TURN, simulcast, bandwidth estimation, SFU routing, congestion control. You’re spending engineering cycles on infrastructure your users can’t see, instead of the UX and AI features that actually differentiate your product. And if real-time voice is the core interaction rather than video, our comparison of voice chat APIs for games and social apps weighs each managed SDK against building your own.

Our standing advice: if you land on “we need to build custom,” first confirm that LiveKit self-hosted plus a narrow fork of the components you need won’t work. In 90% of cases, it will, and our LiveKit development engineers can scope that fork in a single architecture call.

Eight product features that tip the scale

Certain features force specific architecture choices. If any of these is on your roadmap, factor it in now, not after you’ve committed to a platform.

1. Recording + E2EE simultaneously. Still the hardest trade-off in 2026. End-to-end encryption means the media server sees only opaque packets, so server-side compositing and recording are off the table; recording then needs a key-holding recorder participant or client-side capture. Most enterprise clients choose transparent recording with BAA-backed compliance rather than E2EE. If both are mandatory, you’re in custom-build territory or a very narrow hybrid pattern (client-side recording).

2. Breakout rooms with moderator takeover. Native in Whereby, Zoom SDK, and Sendbird. Requires custom UX and signaling in Agora/LiveKit, though LiveKit has good room-migration primitives.

3. Real-time translation. LiveKit Agents + a low-latency STT/LLM/TTS stack now hits sub-1s translation. Agora requires third-party integration. Whereby/Sendbird have no native story. If translation is a roadmap item, bias toward LiveKit.

4. AI note-taking and meeting summaries. Table stakes for B2B in 2026. LiveKit ships this natively via Agents. Agora and Zoom SDK require your own orchestration. Whereby offers a post-call API hook.

5. Avatar-based AI participants. Tavus and Beyond Presence ship avatar plugins for LiveKit Agents, and HeyGen has moved its real-time avatars from the Interactive Avatar API to LiveAvatar. Custom implementation on other SDKs is ~6 weeks of integration work.

6. Screen share with annotation. Whereby and Zoom SDK ship this. LiveKit/Agora require you to build the annotation layer yourself.

7. Custom layouts (spotlight + grid + polls + chat). Only possible with SDK or self-hosted. White-label platforms lock you into their composition.

8. Data residency (EU, India, Middle East). LiveKit Cloud, Agora, and Zoom SDK all have regional options in 2026. Whereby has EU PoPs natively. Full data residency (on your own cloud account) requires self-hosting.

AI-native video chat: what’s new in 2026

The defining shift in video chat since 2024 is that AI agents now join calls as first-class participants. Five capabilities matter in 2026:

AI agent joins a LiveKit call: Deepgram Nova-3, Claude Haiku 4.5 and ElevenLabs Flash in a 700-900 ms speech loop

Figure 3. How an AI agent joins a call in 2026: a LiveKit Agent runs STT → LLM → TTS and speaks back into the room.

1. Real-time translation. In our builds, a LiveKit Agent running a streaming STT (Deepgram Nova-3), a fast LLM (Claude Haiku 4.5) and a low-latency TTS (ElevenLabs Flash v2.5) hits ~700–900ms end-to-end translation latency. That’s conversational. Our voice AI playbook goes deep on the pipeline.

2. Meeting summarization and action items. An agent transcribes the call, runs a summarization pass post-call (or streamed), and emits structured action items to your CRM. This is the Otter/Fireflies pattern, now embeddable in any LiveKit-powered product.

3. AI interviewers and sales agents. Full-call AI voice agent taking discovery calls, pre-screening candidates, or qualifying leads. Requires low-latency speech-to-speech and good turn detection. LiveKit Agents is the reference framework.

4. Avatar-based agents. Tavus Conversational Video Interface and HeyGen’s LiveAvatar (the successor to its Interactive Avatar API) render a talking head in real time. Useful for customer service, sales demos, and training. In our tests first-word latency is still around 800ms–1.2s — acceptable, not yet invisible.

5. Content moderation and safety. Real-time hate speech detection, toxicity flagging, underage voice detection. Critical for consumer social. Runs as a LiveKit Agent or custom webhook against Agora/Zoom.

If any of these five is on your 12-month roadmap, that’s a strong vote for LiveKit Cloud (or self-hosted LiveKit). The Agents framework makes them implementable as Python/Node services that join rooms as participants, dramatically simpler than bolting AI onto a closed SDK.

Compliance deep-dive: HIPAA, GDPR, TCPA, SOC 2

Compliance is where the “buy white-label” argument has strengthened the most between 2024 and 2026. A decade of enforcement activity and several well-publicized fines pushed vendors to ship compliance out of the box.

HIPAA (US healthcare):

Requires a signed Business Associate Agreement (BAA), encryption in transit and at rest, and access logging. BAAs are a listed option at Whereby (HIPAA add-on, $16.99/month on the Build plan), Daily and 100ms ($500/month add-ons), and Zoom, Doxy.me and VSee offer them for healthcare plans. LiveKit Cloud and Agora sign BAAs on enterprise contracts.

2026 context: HHS proposed the first major HIPAA Security Rule overhaul in January 2025 (mandatory encryption, MFA, asset inventories); OCR’s 2026 regulatory agenda pushed the final rule to July 2027. Build to the proposed controls now so the final rule isn’t a retrofit.

2026 context: the Office for Civil Rights has been enforcing normally since its COVID-era telehealth enforcement discretion ended in 2023 (with a transition period to August 9, 2023), so skipping the BAA is expensive. The risk calculus has shifted: use a platform that signs a BAA.

GDPR (EU/UK):

Requires data residency (EU data centers), a Data Processing Agreement (DPA), and user rights (access, deletion, portability). Whereby operates in EU PoPs natively. LiveKit Cloud, Agora, and Daily offer EU regions. Self-hosting on your own EU cloud gives you the strongest posture.

2026 context: the EDPB has increased coordination across member states’ DPAs, with standardized enforcement of transparency requirements. Your privacy policy must disclose exactly what data is processed, where, and for how long. Video and audio metadata counts.

TCPA (US telecommunications):

Applies when your product places calls or texts to phone numbers (including AI voice calls) without prior consent; in-app VoIP between users is not the target. Not enforced at SDK level; liability sits on your product. The February 2024 FCC ruling classified AI-generated voice in outbound calls as subject to TCPA. If your use case involves AI voice agents calling users, consent flows are non-negotiable.

SOC 2 Type II:

Third-party audit of your infrastructure, access controls, and availability. Available from all major SDK vendors (Agora, LiveKit, Daily, 100ms, Zoom) and white-label platforms (Whereby, Sendbird). Self-hosted means you run your own audit — budget $20K–$60K annually.

Compliance shortcut: if you’re in healthcare, education, or financial services and need to ship in under 8 weeks, start with Whereby or Zoom SDK. Off-the-shelf BAAs alone save 3–6 weeks of legal review versus a fresh SDK vendor. You can always migrate later when compliance is stable.

Playbooks by vertical: which stack for which product

The right path depends less on company size than on the product around the call. Here is what we recommend for the five verticals we build most, with the stack, the unit economics and a real project for each.

Telehealth & telemedicine

Recommended stack: Zoom Video SDK or Whereby Embedded for MVP; LiveKit Cloud (or self-hosted) when you need AI triage, translation, or full data residency.

Telehealth has the clearest pattern we see. The MVP must ship with HIPAA compliance already solved, a clinician UX that mirrors existing workflows (calendar invite → clickable link → waiting room → consult → notes), and EHR integration via webhook or HL7. Whereby nails the first three out of the box; Zoom SDK wins when the clinic is already Zoom-native.

Unit economics at 10K visits/month, 20-minute average (about 400,000 participant-minutes): Zoom Video SDK at $0.0035/min comes to roughly $1.4K/month, Whereby Embedded to about $1.7K including the HIPAA add-on, and LiveKit Cloud to well under $1K in minutes and bandwidth before enterprise BAA terms (October 2026 list prices). At this scale the platform bill is not the decision; compliance paperwork and clinician UX are. For the full build, see our WebRTC services for healthcare.

We’ve shipped telehealth platforms on all three paths — Whereby for a small US specialty clinic (live in 5 weeks), a HIPAA video chat for MyOnCallDoc, Zoom SDK for a hospital-network platform (existing Zoom enterprise), and LiveKit Cloud for an AI-first triage product that needed agents joining calls. The LiveKit path costs more in engineering but pays back in feature velocity for AI-native workflows.

EdTech classrooms

Recommended stack: Whereby for SMB and tutoring; LiveKit Cloud or Agora for scale (K-12 districts, MOOCs).

Teachers need three things from video: room URLs simple enough for parents to click, breakout rooms that don’t require admin permissions, and attendance that flows into the LMS. Whereby delivers all three. Agora or LiveKit deliver them once you build the UX, which is worth it past ~50K students where unit economics flip.

Key 2026 feature to plan for: AI tutors joining the room. Claude- or GPT-powered tutors answering homework questions, running oral quizzes, or flagging stuck students. This is a LiveKit Agents pattern; Whereby can’t host it natively.

We’ve shipped EdTech on Agora (a language-learning marketplace), LiveKit Cloud (an AI-tutoring platform), and Whereby (a small-group coaching startup). The right choice is almost always driven by whether AI agents are in the roadmap.

B2B SaaS conferencing

Recommended stack: Daily Prebuilt for fastest embed; LiveKit Cloud when you need AI note-taking or custom controls.

The B2B SaaS conferencing pattern — embed a video call inside your CRM, project-management, or HR tool — is almost always small groups (1–4 participants), screen-share heavy, and recording-optional. Concurrent load is low. Unit economics favor white-label or low-touch SDK.

AI summary is the table-stakes feature for 2026. If your buyer is evaluating your CRM vs competitors and one has “AI takes meeting notes, syncs to opportunities,” that’s the winner. LiveKit Agents makes this ~2 weeks of engineering work; on Daily with Pipecat it’s similar, and on Agora it’s closer to 4–6 weeks.

Here’s a build we know cold, because we’re the team behind it. iMind is a business video conferencing platform that had to hold enterprise buyers — PwC, Kazakhtelecom, Uztelecom, MTS, and divisions of the Czech Republic’s Ministry of Regional Development — so “good enough” consumer video was never going to clear the bar.

The requirements read like a stress test: up to 200 participants on video (1,000 audio-only) in one call, AES-256 encryption end to end, face recognition as a “peephole pass” for secure meeting entry, joint file editing mid-call, noise suppression, and it all had to run as a Progressive Web App so government users didn’t have to install anything. That combination rules out most white-label rooms on day one — the 200-video ceiling and the custom entry flow alone push you to an SDK-or-below path.

The result is a platform that ranks in the top 30 conferencing tools on G2 and sits at 10/10 on TrustRadius, built on a WebRTC SFU we own rather than a closed vendor we’d have to beg for a codec change. iMind is one of the rare path-3/4 cases this article describes: government buyers, face-recognition entry and 200-video rooms are exactly the requirements managed SDKs don’t cover. That’s the tell for a B2B conferencing product: the moment your buyer list includes names with a security team, you’re on the SDK-or-self-host side of the four paths, not the white-label side. Want a similar assessment for your product?

For a lighter build the math flips: we shipped a B2B SaaS conference layer for a sales-enablement platform on LiveKit Cloud plus a Claude-based note-taking agent in seven weeks, and the platform bill at 8K monthly active users stayed under $4K/month.

Consumer social video

Recommended stack: Agora for global scale, LiveKit Cloud for AI-native consumer apps.

Consumer social video — Discord-adjacent, dating apps, creator rooms, drop-in audio+video — lives or dies on latency and unit economics. Agora’s SD-RTN gives the latency edge in APAC. LiveKit wins when AI filters, translation, or avatar-based agents are the product.

At list prices, Agora HD ($3.99 per 1,000 minutes) costs about three times LiveKit Cloud at 100K MAU once you include LiveKit’s bandwidth charges (see the unit economics below); enterprise discounts narrow that gap. At 1M MAU with mixed usage, you’re approaching the self-hosting break-even — plan the migration path before you hit it.

Consumer social is also where moderation matters most. AI-based real-time moderation (hate speech detection, underage voice detection, NSFW image detection on video) is mandatory in 2026 and is trivial to run as a LiveKit Agent — harder to bolt onto closed SDKs.

Marketplace talk-to-seller

Recommended stack: Whereby Embedded or Sendbird Calls for MVP; don’t migrate unless the volume justifies it.

Marketplace video (talk-to-seller on Carousell-style platforms, buyer-seller product demos, real-estate walk-throughs) is almost exclusively 1-on-1, short-duration (<10 min), and never recorded. Whereby Embedded ($10.99/month plus $0.0042 per participant-minute after 2,000 included minutes, October 2026) is the cleanest path. Sendbird makes sense when you’re already shipping chat with the same vendor.

Don’t overthink this one. The feature set is narrow, the unit economics are thin, the brand risk of a flaky call is high. Buy white-label, ship in four weeks, revisit the platform decision only if you cross 100K monthly talk-to-seller sessions.

Unit economics at 100K MAU and 1M MAU

Actual build-vs-buy math depends on your usage pattern. Assumptions: each monthly active user spends 30 participant-minutes in calls per month, mostly 1-on-1 calls at 720p (about 0.01 GB downstream per participant-minute; 3-person rooms roughly double bandwidth and push Agora into its $8.99 Full HD tier), October 2026 list prices, no enterprise discounts, one DevOps FTE at ~$12K/month for self-hosting.

Scale Whereby Embedded LiveKit Cloud Agora SDK (HD) LiveKit self-hosted
100K MAU (3M min/mo)~$12.6K/mo~$3.8K/mo (incl. ~$2.7K bandwidth)~$12K/mo~$13–17K/mo (with FTE)
1M MAU (30M min/mo)Enterprise quote (~$126K at list)~$42K/mo at list~$120K/mo at list~$27–57K/mo (with FTE)

Directional only; computed from October 2026 list prices. Enterprise contracts and usage patterns shift the numbers materially.

Video chat platform cost by MAU, October 2026: $0.004/min SDKs vs LiveKit Cloud vs self-hosting with a DevOps FTE

Figure 4. Monthly platform cost as MAU grows (October 2026 list prices) — self-hosting beats $0.004/min SDKs early but only matches LiveKit Cloud list pricing near 1M MAU.

Three things to notice: (1) at about $0.004 per participant-minute, white-label and most SDKs cost roughly the same, so pick them on features, not price; (2) LiveKit Cloud is the cheapest managed option at every scale on list price, and bandwidth, not minutes, is most of its bill; (3) self-hosting beats the $0.004/min vendors from about 50M participant-minutes a year, but against LiveKit Cloud it only breaks even near 1M MAU. Latency, compliance, and AI features usually matter more than the rate sheet.

Want this math for your actual usage?

Send your projected MAU, group size, and session length — we’ll return a 3-year cost curve across all four paths.

Book a 30-min call → WhatsApp → Email us →

Eight mistakes founders make

These are the mistakes we see every year. Each of them costs 3–9 months of engineering or a six-figure platform bill — and every single one is avoidable.

1. Buying white-label too cheap, hitting the ceiling at 50K MAU. The team saves $2K/month at launch, then discovers the 24-active-video limit per room, the 30-day recording retention, and the enterprise-list pricing at scale. Cost of pivot: a three-month migration and customer-facing downtime risk.

2. Building too custom from WebRTC, running out of runway. Team ships MVP in three months, hits 10K users, realizes self-hosted mediasoup costs $3K/month plus a full-time DevOps engineer. At that volume an SDK would bill about $1.2K/month, so the self-hosted stack costs roughly 10x more than it should. Should have started on an SDK or committed to self-hosted ops from day one.

3. Ignoring the E2EE+recording tension early. Builds product, adds E2EE for privacy, then enterprise customer asks for recording. Realizes architecturally that E2EE breaks server-side recording. Either removes E2EE (angry users) or loses the customer.

4. Overbuying on global scale when you serve one region. Picks Agora for “global scale” but has 90% US users. Pays for APAC PoPs that never get used. Daily, LiveKit Cloud, or even Whereby would have been cheaper.

5. Betting the roadmap on one closed vendor. Twilio Video is the cautionary tale even though it survived: teams that started there in 2023 spent 2024 planning a forced migration, then watched the EOL get cancelled. Whichever way the vendor jumps, the team that had no open-source fallback ate the planning cost twice. Pick a platform (LiveKit is the obvious one) where self-hosting is a real option, so a vendor’s mood swing never forces a rewrite.

6. Assuming premium video is a revenue line. Launches “premium video quality” at a $10/mo upsell, assumes users will pay. They don’t. Video is commoditized in 2026; nobody pays for slightly better bandwidth. AI features monetize; codec quality doesn’t.

7. Confusing “low-latency” with “ultra-low-latency.” Reads that sub-200ms is the standard and picks Agora’s most expensive tier. Their users are fine with 400ms. Wastes half the platform bill on latency they don’t need. 200–500ms is acceptable for social video; sub-100ms is only needed for gaming, remote control, or truly real-time collaboration.

8. Recording storage surprises. Enables recording on Whereby/Daily for all calls. Discovers at scale that storage and egress cost $50K+/month. Should have used user-initiated recording, your own S3 passthrough, or tiered retention by plan.

Pattern behind every mistake: underestimating the three-year TCO and overweighting the launch-month cost. The right decision is the one that minimizes migration cost at your next scale, not the one with the cheapest first invoice.

FAQ

Is LiveKit really the default choice in 2026?

For most new projects, yes. LiveKit raised a $100M Series C at a $1B valuation in January 2026 and powers ChatGPT’s Advanced Voice Mode, so vendor risk is low. Three reasons: the managed Cloud tier is priced competitively against Agora and Daily; the open-source self-host option gives you an exit if pricing changes; and the Agents framework makes AI-native video chat substantially easier than on closed SDKs. Agora is still the right pick for APAC-heavy traffic and extreme global scale, and Zoom SDK wins when the enterprise buyer is already Zoom-standardized.

What does a video chat MVP actually cost to build in 2026?

White-label (Whereby Embedded) can ship in 4–6 weeks at a fixed monthly subscription plus usage — typical MVP budget is $20K–$40K engineering + ~$200/month platform. SDK-based MVP (LiveKit Cloud or Agora) typically takes 8–12 weeks at $60K–$120K engineering plus platform fees. Fora Soft’s Agent Engineering practice, which leans on AI-assisted code generation, has cut several weeks from those timelines on our recent builds.

Is Twilio Video safe to build on in 2026?

It’s supported again: Twilio cancelled the Programmable Video end-of-life in October 2024, and in 2026 it remains a standalone product still accepting new accounts. So it works, but we’d still default new builds to LiveKit Cloud — the open-source self-host path means no single vendor decision can force a rewrite. If you’re on Twilio today, you can stay; if you want out, our Twilio Video migration guide compares Vonage, Daily, LiveKit, and Chime.

What’s the cheapest HIPAA-compliant video chat option?

Whereby Embedded is the cheapest embeddable HIPAA path in October 2026: $10.99/month on the Build plan plus a $16.99/month HIPAA add-on and $0.0042 per participant-minute after 2,000 included minutes. Daily and 100ms list HIPAA + BAA at $500/month on top of usage. Zoom Video SDK is cheaper per minute ($0.0035) but its BAA comes through healthcare and enterprise plans rather than a self-serve add-on. SDK-based paths (LiveKit, Agora) can sign a BAA on enterprise contracts but add roughly 3–6 weeks of legal review.

When does self-hosting actually save money?

Against SDKs priced around $0.004 per participant-minute, from roughly 50M participant-minutes/year, once you price in the permanent DevOps FTE. Against LiveKit Cloud list pricing, not until roughly 300–400M participant-minutes/year (about 1M MAU at 30 minutes each). Above the break-even, self-hosted LiveKit or mediasoup can cut the platform bill substantially — but you own the on-call rotation, codec updates, and browser compatibility matrix. Founders consistently underestimate the ops tax by 2–3x.

Can I have end-to-end encryption AND recording?

Not cleanly at scale in 2026. E2EE means the server can’t decrypt media, so it can’t composite or record. The practical workarounds are (a) client-side recording with uploaded encrypted blobs, (b) compliance-grade recording without E2EE (most healthcare and enterprise use cases), or (c) hybrid architectures where only certain streams bypass E2EE. All three have real trade-offs; pick before architecture is locked.

How much engineering does AI-native video chat add?

On LiveKit with Agents, 1–3 weeks of engineering to get a first voice agent (translation, note-taker, AI interviewer) into rooms. On Daily with its open-source Pipecat framework the effort is comparable; on Agora or Zoom Video SDK it’s closer to 4–8 weeks because you’re building more of the orchestration layer yourself. Our voice AI playbook covers the full stack.

Does Fora Soft help with both the decision and the build?

Yes. A typical engagement starts with a 72-hour discovery: we map your usage pattern, compliance needs, and roadmap, and return a recommendation with a 3-year cost curve. If you move forward with us, the Agent Engineering practice ships the MVP in 6–12 weeks on the chosen stack. We’ve delivered on every major path — portfolio.

What approaches exist for building a video chat app: SDKs, WebRTC, or managed services?

Four: a white-label managed service you embed (Whereby, Daily Prebuilt), a managed SDK you build your own UI on (LiveKit Cloud, Agora, Zoom Video SDK, Twilio Video), a self-hosted open-source media server (LiveKit, mediasoup, Janus), or raw WebRTC with your own signaling and SFU. Most products in 2026 should start on a managed SDK.

What is an online video room SDK?

It is a library plus hosted media infrastructure that gives your app ready video rooms: join, publish camera and mic, subscribe to others, screen share and recording. You embed it as a prebuilt iframe, a UI kit, or raw client SDK calls, and pay per participant-minute (about $0.004 in October 2026 for most vendors).

What is the best SDK for embedding a live avatar in a web app?

In 2026 we use LiveKit with an avatar provider such as Tavus or Beyond Presence, because the avatar joins the same WebRTC room as the user through LiveKit Agents plugins. HeyGen’s real-time avatars now run through its LiveAvatar product.

Which SDK is best for a mobile video calling app?

Pick one with native iOS and Android SDKs and CallKit / ConnectionService support: LiveKit, Agora, Twilio Video or Zoom Video SDK. Iframe-only embeds work on mobile web but can’t ring a locked phone like a native call.

Framework

Build vs Buy a Video SDK in 2026

The SDK-layer companion to this guide — when to license versus wrap your own.

Migration

Build vs Buy: Switching From a Video SDK to a Custom Platform

Already on an SDK? Here’s when the migration economics start to favor self-hosted.

Voice AI

LiveKit Voice AI: The Engineer’s Playbook

How to ship human-sounding voice agents with sub-700ms latency on LiveKit Agents 1.x.

Topologies

P2P vs MCU vs SFU for Video Conferencing

When mesh topology actually works and why SFU is the default above six participants.

Vendor

Agora Alternatives in 2026

A shortlist for teams looking beyond Agora — when LiveKit, Daily, or 100ms are the better fit.

Ready to pick the right path for your video chat platform?

The four-path framework collapses the build-vs-buy decision into a 30-minute exercise: define your usage, compliance, AI roadmap, and runway, then map them onto Buy a white-label, License an SDK, Self-host an OSS engine, or Build a custom stack. Most founders pick the wrong path because they over-weight launch speed and under-weight unit economics 18 months out — managed-SDK platforms scale to seven-figure annual bills faster than expected, while custom builds push break-even past most seed runways.

Fora Soft has shipped on every path — Agora, Daily, LiveKit Cloud, self-hosted LiveKit, mediasoup, and fully custom WebRTC SFUs. The right answer depends on your numbers, not your taste in stacks. If you’re early, an SDK gets you to product-market fit in weeks. If you’re well past ~50M participant-minutes/year on a $0.004/min SDK and AI features are core, self-hosted LiveKit typically pays back within 12–18 months; a fully custom build rarely does. Either way, the cheapest decision is the one you make before you ship the wrong architecture — and if you’d rather not make it alone, that’s exactly what our video streaming app development team does every week.

Book cover: Video Streaming, Volume 3 of the Fora Soft Video Engineering Handbook

The book · Volume 3 of 7

Video Streaming: A Complete Guide to Video Delivery: From ABR and CDNs to Players, WebRTC, DRM, and the Economics of Streaming

Whichever of the four paths you pick, the call quality your users feel comes from the same delivery layer. Volume 3 of our handbook covers it: WebRTC and real-time, ABR, CDNs, players, QoE monitoring and the economics of delivery. Read it before you compare SDK invoices with a self-hosted SFU bill. Written by Nikolay Sapunov, CEO at Fora Soft.

Read it on Amazon →Kindle & paperback · Fora Soft Video Engineering Handbook

Ready to decide between buy, SDK, OSS, or custom?

Book a 30-minute architecture call. We’ll walk your use case through the four-path framework and return a recommendation with a migration timeline.

Book a 30-min call → WhatsApp → Email us →

  • Technologies