Custom intercom software architecture with video streaming, authentication, and visitor management

Key takeaways

• The video intercom devices market is about $43.8B in 2026, heading toward $74.4B by 2030 (The Business Research Company, 2026, 14.1% CAGR). Video intercom software is the layer taking share as IP and cloud-native systems eat proprietary hardware, and interoperability is now the buying criterion.

• Cloud TCO runs 25–35% under on-prem across five years. But cloud-only breaks when the internet does, so a hybrid that keeps local SIP fallback is the production default.

• Face recognition is a compliance decision, not just a feature. GDPR Art. 9 and Art. 22 apply now. Under the EU AI Act, 1:N face identification is high-risk with obligations from 2 Dec 2027 (AI Omnibus, Regulation (EU) 2026/1744), while 1:1 verification sits outside Annex III.

• Mobile push latency is the top UX complaint. Sub-2-second answer on residents’ phones is the bar; miss it with plain notifications and people stop answering the door.

• Fora Soft ships cloud video products for a living. WebRTC work since 2012, 250+ projects since 2005, and shipped intercom systems like our Doorbell App (a SIP-to-WebRTC bridge) and Netcam Studio.

Updated November 2026: new audio pipeline section (AEC, Opus 1.6), EU AI Omnibus 1:1 vs 1:N rules, CRA reporting since 11 Sep 2026, Matter controller status and 2026 market data.

Why Fora Soft wrote this playbook

Fora Soft has shipped video-first software since 2005, across 250+ projects. We built Netcam Studio, a video surveillance product with roots in WebcamXP, one of the earliest IP-camera apps. We ship WebRTC and SIP systems for clients every week (see our WebRTC services and realtime voice integration guide). One of those was a Doorbell App: a LiveKit SIP-to-WebRTC bridge that rings any SIP intercom straight through to a resident’s phone and releases the door over DTMF, with no vendor lock-in. We also built Doma.ai, a browser-based video intercom for a property-management platform used by 4,305+ management companies across 40+ cities, where a resident sees the door feed and opens it from a phone browser with no app to install.

This is the playbook we use when a PropTech founder, a building integrator, or a security-platform CTO asks: “What does a modern video intercom product look like, what does it cost to build, and where do teams get it wrong?” You’ll get the reference architecture, the protocol picks, honest AI-feature sanity, the 2026 compliance timeline, a real cost model, and a five-question decision framework. If you only need the buy-vs-build answer, jump to the decision framework.

Related reading: IoT + video surveillance architecture, our Video Streaming learning track.

Planning a video intercom product or a smart-building rollout?

Tell us your deployment (residential, commercial, multi-tenant) and we’ll sketch architecture and cost in a 30-minute call.

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

What changed: from buzzer to cloud

Three shifts drove the last five years. First, PoE and IP cameras turned the door station into a general-purpose computer. Second, WebRTC and SIP trunking let a call reach any phone anywhere, not just the wired monitor inside the apartment. Third, edge AI (face recognition, package detection) dropped in price to the point where residents now expect it by default.

So a modern video intercom is four products in one front panel: a video product, an access-control product, an IoT device, and a compliance artifact. Building it as “a doorbell with a camera” is how these projects die.

If you need the definition of a cloud intercom, the vocabulary (virtual, apartment, cloud platform) and per-door pricing of SaaS versus custom, our cloud-based intercom system guide covers that. This playbook is about what sits under the panel: how video and audio get captured, cleaned, encoded and delivered to a phone.

How do you integrate video and audio in an intercom? The short answer

Video intercom software integrates video and audio by running two pipelines that share one call session. Audio goes through echo cancellation, noise suppression and gain control on the door station, is encoded with Opus (or G.711 for legacy SIP), and travels over RTP. Video is encoded as H.264 (H.265 on newer panels) and travels on the same session. A SIP or WebRTC backend delivers both streams to monitors, phones and browsers, and the player keeps lip sync and the door-open command in step.

StageWhat runs thereWhat goes wrong
CaptureMic array, speaker, 2K camera with IRThe speaker sits a few centimetres from the mic, so the panel hears itself
Clean-up (audio)AEC, noise suppression, AGC (WebRTC APM / AEC3 or the SoC’s DSP)Without AEC, full-duplex talk turns into echo; half-duplex push-to-talk is the fallback
EncodeOpus 1.6 or G.711 for audio; H.264 / H.265 for videoOpus 1.6 (Dec 2025) adds ML bandwidth extension and better Deep Redundancy (DRED) for lossy uplinks
TransportSIP/RTP to monitors and PBX; WebRTC (SRTP, ICE, TURN) to appsBridge the two in the backend, never on the phone
PlayoutJitter buffer, lip sync, CallKit / Android full-screen call UILip sync drifts first when the uplink is 4G or Wi-Fi
ControlDTMF, SIP INFO or REST command for the lock relayThe door release must ride the same authenticated session as the call

Most intercom complaints are audio problems, even in video products: echo on hands-free panels, wind and traffic noise at street doors, and clipped first words when the phone takes a second to open the media path. We tune echo cancellation on the actual panel enclosure, because a lab test misses the reflections of a metal housing. We also start audio before video on answer, so the visitor hears the resident at once while the first video keyframe is still on its way.

The 2026 market snapshot

The video intercom devices market is worth about $43.8B in 2026 and is forecast to reach $74.4B by 2030, a 14.1% CAGR (The Business Research Company, 2026). North America holds the largest share and Asia-Pacific grows fastest (Mordor Intelligence, 2025). The number that matters for a build decision is the mix shift underneath that headline: IP and cloud-native systems are taking share from analog and proprietary hardware, and interoperability (SIP, ONVIF, Matter) is now what buyers screen for.

Metric20262030 (forecast)Source / note
Video intercom devices$43.8B$74.4BThe Business Research Company 2026, 14.1% CAGR
Regional leadersNorth America #1APAC fastestMordor Intelligence 2025
Architecture shiftIP + cloud risingcloud-native mainstreaminteroperability is the buying criterion
Face-recognition entrypremium tierbroader mid-tiergated by 2026–2027 EU AI Act posture
Call-answer latency target< 2 s< 1.5 sWebRTC glass-to-glass, mobile app

Analyst scopes differ wildly: Mordor Intelligence put “devices & equipment” at $50.8B for 2025 under a wider definition; narrower “video intercom software” or “IP video door phone” slices come in far smaller. Use the trend (double-digit growth, IP and cloud taking share), not any single headline dollar figure, to justify a build.

Reference architecture for a modern video intercom

Five logical tiers, with clean seams between them. Blur two tiers together and you pay for it at integration time.

Five-tier video intercom software architecture: door station, in-unit monitor, cloud backend, apps, integration adapters

Figure 1. The five-tier reference architecture: door station, in-unit monitor, cloud backend, apps, and integration adapters.

TierResponsibilityTypical tech
1. Door station / front panelCapture video + audio, authenticate, drive lockIP cam (2K), mic array, speaker, NFC/PIN, PoE++ power, SIP client, edge AI
2. In-unit monitor (optional)Wired answer point, guaranteed response7–10” Android/Linux panel, PoE, local SIP fallback
3. Cloud backendSignaling, directory, recording, analyticsSIP proxy, WebRTC SFU, REST/GraphQL API, object storage, RBAC
4. Mobile & web appsResidents answer, guests, admin dashboardiOS / Android / React web, WebRTC, PushKit / CallKit, FCM
5. Integration adaptersAccess control, PMS, smart home, surveillanceOSDP, Matter 1.5 intercom, HomeKit, Alexa, Yardi/AppFolio/Entrata, ONVIF

Reach for a cloud-native architecture when: you’re a SaaS, an ISV, or a property-manager-facing product. Reach for a hybrid with a local NVR when you sell into buildings with strict uptime or privacy needs (luxury residential, government, regulated healthcare).

Transports: SIP, WebRTC, and when to use each

Two call-delivery paths are in common use. Most serious products run both, for different endpoints, and bridge them in the backend.

SIP and WebRTC video intercom call path: wired door station over SIP, cloud bridge, WebRTC to resident phones

Figure 2. SIP keeps wired door stations reliable; WebRTC gives phones sub-second answer; the cloud backend bridges the two.

DimensionSIPWebRTC
Best forDoor station, in-unit monitor, landline phonesMobile app, web dashboard, property-manager console
Latency300–800 ms150–500 ms
NAT traversalSTUN / TURN + SIP ALG headachesBuilt-in ICE + TURN
CodecsG.711 / G.729 / Opus; H.264Opus + VP8/VP9/H.264
When to bridgeLandlines, legacy PBX, some door stationsAnything browser- or mobile-native

The common pattern: door stations and in-unit monitors talk SIP to your backend; mobile and web clients talk WebRTC; the backend bridges. That buys you the reliability of SIP on the door and the latency of WebRTC on the phone. Our Doorbell App did exactly this with a LiveKit SIP-to-WebRTC bridge, so any SIP door station works without swapping hardware.

Reach for SIP-first when: your endpoints are wired door stations, in-unit panels, or landlines that must survive an internet outage. Reach for WebRTC-first when: the answer point is a phone or a browser and every 100 ms of glass-to-glass latency shows.

AI features that residents actually use

Not every “AI” feature earns its panel space. These four land in real deployments, and each maps to work our AI video recognition team does on edge hardware: face matching, object detection and tracking.

1. Face recognition with liveness / anti-spoof

Hands-free entry for authorized people. On curated photos, the top algorithms in NIST’s ongoing FRTE 1:1 evaluation hit vanishingly low false-non-match rates. The honest caveat: those are visa- and border-quality images, not a door camera shooting a backlit face at night. Budget a fallback path (PIN, fob, app entry) and mandatory anti-spoof, because a printed photo defeats a detection-only model. Depth or IR liveness is the floor, not a nice-to-have.

2. Package / delivery detection

Solves the top pain point in multi-tenant buildings. A classifier that tells people from packages, vehicles, and pets cuts nuisance push notifications sharply and opens a porch-piracy alert path. It runs on-edge on the door station, so it works even when the uplink is flaky.

3. QR / one-time PIN entry

Best for delivery drivers and short-term guests. Generate a time-boxed code in the resident’s app, deliver it by SMS or email, and let the camera scan the QR at the door. No shared PINs, no forgotten codes, and a full audit trail.

4. Video voicemail + missed-call clip

Nobody answers every time. When no resident picks up, record 30 seconds around the button press and notify. This one feature is often the difference between “we love it” and “we turned it off.”

Reach for on-device face recognition when: latency and privacy both matter (embeddings stay on the panel, no cloud round-trip). Push it to the cloud only for heavier reasoning like visitor-intent classification, and never auto-deny entry on model output alone.

The protocol stack — what you’ll touch

Protocol / standardRoleWatch-out
SIP (RFC 3261)Call signalingDisable SIP ALG on customer routers or expect one-way audio
WebRTCBrowser / mobile mediaRun your own TURN server; don’t rely on public relays
ONVIF Profile S / TSurveillance integrationEssential for multi-vendor, optional for single-vendor
PoE+ (802.3at) / PoE++ (802.3bt)Power the door station (25–90 W)Budget IR LEDs + heater + PTZ; long runs derate
Matter 1.5 / 1.5.1 / 1.6Smart-home interoperabilityIntercom is a first-class device type since Matter 1.5 (Nov 2025); Matter 1.6 (Jun 2026) is current; the gap is controller support, not the spec
OSDP / WiegandAccess-control readersOSDP Secure Channel encrypts; Wiegand is plaintext and spoofable
APNs VoIP Push / FCMWake mobile on incoming calliOS CallKit + PushKit only; never a regular push for ringing

The one that trips teams up most: Matter. As of Matter 1.5 (finalized Nov 2025), “intercom” is a defined device type alongside cameras and video doorbells, and Matter 1.5.1 (Mar 2026) refined its signaling and added integrated chimes. Matter 1.6 followed in June 2026. So “wait for the spec” is the wrong mental model. The real wait is on controllers. Apple Home still did not support Matter cameras in the iOS 27 beta (June 2026). Google’s Nest cameras and doorbell were certified for Matter 1.5.1 by August 2026, but the regular Google Home app was still on Matter 1.4.1 at that point. Plan for native SIP and your own app first, and treat Matter as an extra answer point.

The 2026 vendor map

A short map of the players we see in live deployments and where each sits in the value chain.

Businesses usually source video intercom systems from authorized integrators of these brands, or directly from the SaaS players for multi-tenant residential. In US federal and many enterprise tenders, check the FCC Covered List and NDAA Section 889 first: Hikvision and Dahua equipment is excluded there.

VendorBest forWatch-outs
Aiphone (IX / IXG)Commercial + residential, well-integrated to major VMSHardware cost, slower cloud roadmap
AkuvoxFace-recognition-heavy deployments, mid priceFirmware quality varies by SKU; vet CVE history
2NStrong SIP compatibility, PBX-friendlyPremium price
DoorBirdSmall-building and single-familyCloud-tied features, API depth varies
Comelit / SiedleEuropean builds, 2-wire retrofit pathsSlower cloud roadmap, higher integration cost
ButterflyMX / Swiftlane / DOOR (formerly Latch)Multi-tenant residential SaaSCoupled hardware + SaaS; margin pressure on ISVs building alternatives

If you’re a building integrator, this is a shortlist. If you’re an ISV building your own product, it’s a competitive map: the SaaS players (ButterflyMX, Swiftlane) own the resident app and the recurring revenue, which is exactly the layer you’d be fighting for.

Reach for a SaaS incumbent when: you manage a handful of buildings and want a resident app tomorrow. Reach for a custom build when: intercom is core to your product, you need to own the app, AI, and data, and you have a credible path to hundreds of buildings.

Need a second opinion on build vs buy for your intercom stack?

Send us your target deployment and the vendors you’re weighing; we’ll give you a straight read in a 30-minute call.

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

Integrations that actually close deals

For B2B multi-tenant products, five integrations consistently separate a demo from a signed contract.

1. Property management (PMS). Yardi, Entrata, AppFolio, Buildium. Resident directories, move-in/move-out automation, lease-status sync. Skip this and expect to lose to a vendor that has it.

2. Access control. Salto, Brivo, Kisi, LenelS2, HID. OSDP Secure Channel, not Wiegand. One credential set for door, elevator, and amenities.

3. Smart home / voice. HomeKit, Alexa, Google Home, Matter. Residents want the door to ring on their kitchen speaker, not only in a single app.

4. Surveillance / VMS. ONVIF Profile S/T into Milestone, Genetec, Eagle Eye. Share the door feed with the existing security platform instead of adding another operator tab.

5. Delivery platforms. Amazon Key, UPS, FedEx, Instacart. One-time QR codes or short-lived tokens for couriers, never a shared building code.

Voice-AI concierge at the door. OpenAI Realtime and similar models (see our realtime voice guide) are already shipping in pilots as “AI receptionists.” Visitors state their purpose; the model handles delivery, guest, and service-worker cases without waking a resident.

Matter intercom maturity moves to the controllers. The spec already defines intercom (Matter 1.5, Nov 2025; refined in 1.5.1, Mar 2026). The next 18 months are about Apple Home, Google Home, and Alexa shipping consistent support, at which point answer-the-door-from-a-smart-speaker stops being a project and becomes a checkbox.

EU AI Act obligations land in two waves. Article 50 transparency duties (disclose AI interaction, label synthetic media) apply from 2 Aug 2026; systems already on the market before that date have until 2 Dec 2026 for machine-readable labels. High-risk obligations for remote biometric identification (a 1:N match against enrolled residents) were moved to 2 Dec 2027 by the AI Omnibus (Regulation (EU) 2026/1744, in force 27 Jul 2026). Face verification whose only purpose is to confirm that a person is who they claim to be (1:1) is outside Annex III, which makes it the cheaper design for door entry. Plan for the dossier now; you have runway, not a reprieve.

Cyber Resilience Act reporting is live. Since 11 Sep 2026, manufacturers of connected products sold in the EU, door stations included, must report actively exploited vulnerabilities within 24 hours and send a fuller notification within 72 hours. The full CRA obligations (secure-by-default, SBOM, security updates for the support period) apply from 11 Dec 2027.

5G fixed-wireless as primary uplink. For buildings without fiber, 5G FWA plus an LTE failover is becoming the default WAN. Design the backend for dual-path uplinks.

Mini case: a cloud-first video intercom without ripping out the panels

Situation. A property operator ran a decade-old 2-wire audio intercom across a 200-apartment portfolio. Residents ignored it, deliveries piled up, and complaint tickets came in weekly. They wanted one system (web admin, resident mobile app, cloud directory, delivery access) without pulling the existing audio panels on day one.

The plan. Weeks 1–2: retire-and-replace audit, a PoE lobby door station, and a 2-wire bridge for the in-unit panels. Weeks 3–5: cloud backend, SIP proxy, and a resident mobile app with VoIP push. Weeks 6–7: opt-in face-recognition enrollment, one-time QR for deliveries, and PMS directory sync. Weeks 8–10: pilot, UX tuning, and a phased rollout across two buildings. The SIP-to-WebRTC bridge pattern from our Doorbell App carried straight over.

Outcome. Delivery-related tickets fell sharply, median answer time on residents’ phones landed at 1.6 seconds against a sub-2-second goal, and the five-year TCO came in well under the incumbent on-prem bid. Want a similar assessment of your stack? Book a 30-minute call and we’ll walk your options.

Cost model: hardware + software + integration

Line itemResidential 50 unitsCommercial 200 doorsNotes
Door stations (hw)$1,200–4,000$40k–120k$400–2,500/door by feature tier
Cabling + labor$3k–8k$30k–80kNew PoE pulls at $50–200/drop
Cloud / SaaS$400–1,200/mo$2k–5k/mo$8–25 per door/month at scale
Integration (PMS, access)$2k–6k one-time$6k–20k one-timeYardi / Entrata / AppFolio / custom
Custom SaaS build (MVP)n/a$120k–280k16–24 weeks; Agent Engineering trims the scaffolding
5-year TCO bars: cloud-native video intercom vs hybrid vs on-prem for a 200-door build, cloud 25-35% lower

Figure 3. Five-year TCO for a 200-door build: cloud-native runs roughly 25–35% under on-prem, mostly on avoided on-site servers.

For an ISV building the SaaS itself (not just deploying one), the meaningful row is the custom build. Our video and audio streaming software development team uses Agent Engineering, which typically trims the timeline on heavily pattern-based work: the SIP proxy, push gateway, device provisioning, and directory-sync scaffolding. We keep estimates conservative; if a number isn’t defensible, we don’t quote it.

Want a realistic MVP scope and cost for your intercom software?

Send us your target deployment size and compliance region and we’ll come back with a phased plan and a rough number in a 30-minute call.

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

A decision framework in five questions

Video intercom decision tree: buy off-the-shelf, hybrid with SIP fallback, or build a cloud-native platform

Figure 4. Build, buy, or bridge: walk the trunk top to bottom; a Yes branches to the call you should make.

Q1. Is this a product or a deployment? Product (your SaaS, many buildings): cloud-native, multi-tenant, ISV-grade. Deployment (one building, internal): off-the-shelf Aiphone / Akuvox / DoorBird plus integration.

Q2. What’s your uptime requirement? If tenants tolerate “delivery-only” behavior during internet outages, cloud-only is fine. If building access must always work, hybrid with local SIP fallback is mandatory.

Q3. What’s the compliance region? EU/UK: DPIA, GDPR Art. 22, and a high-risk EU AI Act posture (dossier due Dec 2027). US: state-level (IL BIPA, TX CUBI, CA CCPA/CPRA). Your legal posture decides whether face recognition ships at all.

Q4. What’s the existing wiring? PoE Cat6 in every riser: greenfield is easy. 2-wire legacy: budget a bridge or a phased rip-and-replace. Luxury residential accepts the disruption; budget rental usually doesn’t.

Q5. Where does the call go? In-unit monitors only: SIP-only is enough. Mobile apps: WebRTC + VoIP push. Landlines: a SIP trunk to PSTN. Most modern deployments answer all three at once.

Five pitfalls we see sink video intercom projects

1. SIP NAT-traversal black holes. Symmetric NAT plus SIP ALG eats a chunk of calls on residential ISPs. Run a TURN server, disable ALG at provisioning, and test from live routers before declaring victory.

2. Mobile push delivery delays. Android’s Doze and iOS throttling can turn a 200 ms round-trip into eight seconds. Use VoIP-class push (iOS PushKit + CallKit, Android high-priority FCM), not regular notifications. On Android 14 and later, the full-screen call screen needs the USE_FULL_SCREEN_INTENT permission, which Google Play keeps only for apps with calling or alarm features, so declare the intercom app as a calling app.

3. PoE budget exhaustion. 25 W PoE+ covers a basic station; add IR LEDs, a heated touchscreen, and a PTZ motor and you’re into 60–90 W PoE++ (802.3bt). Spec it at the switch or pick wall-powered variants.

4. Skipping the offline mode. The cloud goes down; the door still has to work. Local SIP peer-to-peer between door and in-unit monitor, plus events queued for sync on recovery, are non-negotiable in production.

5. Shipping face recognition without a DPIA. In the EU, GDPR Art. 22 means a denied-entry decision can’t be solely automated, and the AI Act adds high-risk documentation duties for 1:N identification by Dec 2027. Ship with a DPIA, an opt-out, and human review, or don’t ship.

KPIs to track from day one

Quality KPIs. Call-answer latency P95 on mobile (target < 2 s), face-recognition true-positive rate (> 99%), false-reject rate (< 0.5%), package-detection precision (> 92%).

Business KPIs. Monthly answer-rate per resident (> 75%), delivery-ticket volume trend (month-over-month down), face-recognition opt-in rate (benchmark 40–60%), and time-to-provision a new unit (< 10 minutes self-serve).

Reliability KPIs. Door-station uptime (> 99.9%), push-delivery success (> 98%), offline-mode fallback success (> 99.5%), and PMS integration-sync lag (< 5 minutes).

Security, compliance, and accessibility

GDPR + EU AI Act. Face data is special-category under GDPR Art. 9, and Art. 22 blocks a solely-automated denied-entry decision. On top of that, the EU AI Act treats remote biometric identification (1:N) as high-risk: Article 50 transparency duties apply since 2 Aug 2026, and the high-risk obligations were moved to 2 Dec 2027 by the AI Omnibus. 1:1 verification is outside Annex III, but GDPR still applies to it. Keep a DPIA on file, require consent, and put a human in the loop for denials.

EU Cyber Resilience Act. Door stations and their apps are products with digital elements. Exploited-vulnerability reporting (24 h early warning, 72 h notification) applies since 11 Sep 2026; the rest of the CRA applies from 11 Dec 2027. If you ship into US federal sites, FIPS 140-2 certificates moved to the historical list on 21 Sep 2026, so ask for FIPS 140-3 modules.

US biometric privacy. Illinois BIPA carries a private right of action (statutory damages per violation; SB 2979 in 2024 limited them to a single recovery per person). Texas CUBI, Washington HB 1493, and California CCPA/CPRA all impose duties on biometric capture. Written consent, disclosure, data minimization, and audit logs are non-negotiable.

Accessibility (ADA + EN 301 549). Audio and visual feedback on the door station, tactile buttons, hearing-loop support, and color-contrast compliance in the app. In new commercial builds the inspector may actively check.

Fire code + egress. Door hardware must fail-safe open on power loss where fire code requires it. Your software cannot block egress, not even for unpaid rent.

When NOT to build your own video intercom product

You’re deploying to one building. Pick off-the-shelf: Aiphone, Akuvox, DoorBird, 2N. Integrating to your PMS is a project; the hardware and firmware are not.

You need a product in three months and have never built real-time video. Partner or OEM. Building the SIP + WebRTC + mobile-push stack from scratch on that timeline is a false economy.

Your target market is tiny. Build-to-own pays back only if you’re serving hundreds to thousands of buildings. Below that, margins don’t cover custom firmware and compliance.

FAQ

How do video and audio work together in a video intercom?

Both streams share one call session. Audio is cleaned on the door station (echo cancellation, noise suppression, gain control) and encoded with Opus or G.711, video is encoded as H.264 or H.265, and a SIP or WebRTC backend delivers both to monitors and phones with lip sync and the door-release command on the same session.

We have 40 doors on old panels and a $30k budget. Can we get to cloud without rip and replace?

Often yes, if you keep the existing wiring and phase it. Use 2-wire-to-IP adapters or SIP-capable entry panels at the doors that matter most, and bridge them to a cloud backend and resident app. At $400–600 per entry-level IP station, 40 doors is $16k–24k in hardware before labor, so $30k fits a phased retrofit with entry-level panels, not premium face-recognition stations on every door.

Can we replace a proprietary intercom with an API-first platform?

Yes, if the door stations speak SIP. Put a SIP-to-WebRTC bridge and your own API in front of them, as in our Doorbell App, and the hardware stays while the app, directory and integrations move to your platform. Proprietary 2-wire panels need an IP adapter or replacement first.

Can I reuse my building’s 2-wire intercom wiring?

Not directly for IP video — the bandwidth isn’t there. Two options: a 2-wire-to-IP bridge (Comelit and Akuvox sell retrofit kits), which works but adds latency and cost; or rip-and-replace with Cat6 + PoE in a phased upgrade. Bridge first if budget or disruption tolerance is low, then replace.

How long does installation take?

Single-door commercial: 2–4 hours. A 50-unit residential building: 2–4 weeks including PMS integration and resident onboarding. A 200+ door multi-building rollout: 6–10 weeks with phased go-live. Most schedule risk is in the PMS and access-control integration, not the door hardware.

What happens when the internet goes down?

If you designed for offline, local SIP between door and in-unit monitor still works, queued entry events fire when connectivity returns, and cached resident directories stay valid. Remote mobile answering stops. Cloud-only products go dark entirely, which is why they don’t belong in a production building.

Is a biometric intercom “high-risk” under the EU AI Act, and when do the rules bite?

It depends on the matching mode. 1:N identification against a gallery of residents is high-risk, and those obligations (documentation, human oversight, conformity) apply from 2 Dec 2027 under the AI Omnibus. 1:1 verification that only confirms a claimed identity is outside Annex III. Article 50 transparency duties apply since 2 Aug 2026, and GDPR Art. 9 and Art. 22 apply now regardless.

Does Matter support intercoms yet?

Yes. Matter 1.5 (finalized Nov 2025) defines intercom as a device type alongside cameras and video doorbells, and Matter 1.5.1 (Mar 2026) refined its signaling and added integrated chimes. Matter 1.6 (June 2026) is the current release. The remaining gap is controller support: Apple Home had no Matter camera support in the iOS 27 beta (June 2026), and Google Home was still on Matter 1.4.1 in August 2026.

What’s a realistic hardware budget per door?

About $400–600 for an entry-level IP station, and $1,500–2,500 for a premium multi-tenant panel with touchscreen and face recognition. Cabling and labor add $50–200 per drop. Cloud or SaaS runs $8–25 per door per month at scale.

Why are mobile notifications slow or sometimes missed?

Two reasons: OS background throttling and non-VoIP push categorization. On iOS use PushKit + CallKit for immediate wake-up; on Android use high-priority FCM. Regular notifications can be delayed 5–30 seconds by the OS, which kills the door-answering experience.

How long does it take to build the SaaS platform itself?

An MVP in 16–24 weeks if you lean on mature door-station hardware and focus your build on the cloud backend, mobile apps, directory, and integrations. A full production platform with face recognition, PMS integrations, and compliance posture: 30–40 weeks. Agent Engineering trims the scaffolding and adapter work.

Surveillance

Integrating IoT with Video Surveillance Software

The neighboring domain: reference architecture for IoT video surveillance.

Intercom

Enterprise Intercom System: 2026 Software Playbook

Multi-site intercom platforms: SIP to PBX and Teams, VMS, SSO and audit.

Voice AI

OpenAI Realtime + WebRTC / SIP / WebSockets

The voice-AI layer that makes an intercom conversational in 2026.

Testing

How to test WebRTC stream quality

Measuring glass-to-glass latency, MOS, and freeze rate — the metrics that matter.

Services

Fora Soft video surveillance development

What we build and how — Netcam, VALT, and related deployments.

Ready to ship a video intercom residents actually use?

Three moves cover most of the risk. Pick hybrid cloud with a local SIP fallback so the door still works when the internet doesn’t. Spend your latency budget where residents feel it: mobile push and WebRTC answer. Ship face recognition only with the full compliance scaffolding, because the EU AI Act now classifies it as high-risk and residents will ask.

If you’re building a product, the architecture above is the starting blueprint. If you’re deploying one, use it as the shortlist of questions to hand your vendor. Either way, real-time video products live or die on the details under the hood.

Book cover: Audio for Video, Volume 4 of the Fora Soft Video Engineering HandbookBook cover: Video Streaming, Volume 3 of the Fora Soft Video Engineering HandbookBook cover: AI for Video, Volume 7 of the Fora Soft Video Engineering Handbook

The handbook · 7 volumes

Fora Soft Video Engineering Handbook

A hybrid intercom that keeps local SIP fallback still has to sound clear and connect fast when a resident answers from a phone. We wrote a seven-volume handbook on the whole video stack: encoding, delivery, sound and AI. Start with volume 3 on WebRTC and real-time delivery, add volume 4 for echo cancellation and noise suppression, and volume 7 for biometric rules under the EU AI Act. Written by Nikolay Sapunov, CEO at Fora Soft.

Browse the handbook on Amazon →7 volumes · 9,000+ pages · Kindle & paperback

Talk to the team behind 250+ video and real-time products

Thirty minutes, your architecture on a whiteboard, and a walk-away plan for your video intercom product or integration — whether we build it together or not.

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

  • Technologies