Learn · track
Voice and video
Calls and video inside your product: how real-time media works, what latency feels like, and when video is worth it.
- 1WebRTC in plain termsWhat the browser gives you for free, what it does not, the three acronyms you cannot avoid, and where the real costs and support load sit.
- WebRTC handles capture, connection and encrypted media. Signalling, relays, group calls and recording are what you build or buy.
- Budget for TURN relays and a media server for group calls; those are the components with running costs.
- Expect most support tickets from camera permissions, device choice and network changes. Show users clear, specific messages for each.
- 2Peer-to-peer video: what it removes, and what it does notFor small calls, a media server is optional. When it stops being optional — and what a hosted video service gives you in exchange for its price — is the whole decision.
- Peer-to-peer suits one-to-one calls. Plan for a media server once you need more participants, recording, transcription or streaming.
- Measure what share of your calls need a relay server. Corporate and hotel networks push it up, and relays carry the bandwidth bill.
- Compare self-run and hosted options on total effort — operations, security, compliance evidence, mobile support — not just the per-minute price.
- 3Latency you can actually feelFor messages, the network is rarely the slow part: queues, carriers and handsets dominate, and most of it is chosen at purchase. For live calls, a few hundred milliseconds decides whether a conversation feels natural.
- Measure from the moment you decide to send until the delivery receipt, per country, and watch the slowest five per cent.
- Give code screens a visible timer, an early alternative route, and codes that stay valid longer than a slow delivery.
- For live voice and video, keep one-way delay under about 150 milliseconds; beyond that, people start talking over each other.
- 4When video belongs in a support conversation, and when it is theatreVideo resolves a narrow class of problems dramatically faster and every other class slightly slower. Knowing which is which — and deciding recording and data flows up front — is the entire design.
- Offer video only when seeing the problem beats describing it, or to verify identity — never for questions text can answer.
- Start in messaging, offer video as a one-tap browser link with the camera optional, and write the outcome back to the thread.
- Decide recording, consent and where media travels before launch; even one-to-one calls often pass through relay servers.