Choosing a video API: embed, record, and co-browse — the buyer checklist

A video API evaluation fails the same way most of the time: the committee scores a hundred-line feature table and the four features that actually decide the program — the browser embed, the recording, the co-browse, and the moderation surface — get one row each among ninety filler rows. This checklist runs those four in order, plus the two that decide regulated and AI programs, and scores them against what Devotel Orbit's Video API ships today. The companion video and RTC provider comparison frames the vendor field; this post is the criteria battery to run on it.

Checklist item 1 — embedded rooms, not a redirect

The embed question is whether the room lives inside your product or beside it. A video API that answers with a hosted app your customers redirect to is a meeting vendor, not a video API.

• Ask: is the embed a declarative element or SDK class in my page, with the tenant API key server-side?
• Orbit's answer: yes. The embed is a three-step wire — a server-minted join token, the <orbit-video-room> element or the OrbitVideoRoom SDK class, and a moderated room in the browser. The walkthrough in Embedding Video for Customer Support runs it end to end; the browser carries only the short-lived token.
• Verify on the vendor page: the embed demo should join a room from your own page, not link out.

Checklist item 2 — session recording you can operate

Recording is where video APIs diverge hardest, because "we record" hides the operating questions: who can record, when does it pause, who plays the file back, and who revokes access.

• Ask: can the host start, pause, and resume mid-session; exclude a participant from capture; and share the result via a link that revokes?
• Orbit's answer: yes on all three. Any session can be recorded, paused and resumed mid-call, with individual participants excluded from capture, and output shared through revocable links. Recordings land back on the account — and because the video channel sits on the same account as voice and messaging, the file attaches to the same contact profile the session belongs to, so playback and retention are one dashboard question rather than a second vendor's console.
• Red flag on the vendor page: recording that requires a separate egress pipeline purchase, or a global share URL with no revocation.

Checklist item 3 — co-browse during the call

Support-driven video programs do not end at camera and microphone. The moment text chat stops resolving an issue, the agent needs to see and guide the customer's actual screen.

• Ask: does the platform ship agent-to-customer screen guidance as a named capability, or is "screenshare" the whole answer?
• Orbit's answer: co-browse ships as a first-class capability — screen guidance inside the live session, part of the RTC pillar alongside rooms, broadcast, and recording. The co-browse and digital support post covers the support-center pattern.
• Red flag: "screenshare" listed as a room SDK feature is a presenter tool, not an agent-side guidance surface. Named co-browse or nothing.

Checklist item 4 — moderation that is enforced, not cosmetic

A lobby that only delays the join is a UI flourish. Moderation is enforced when the gate is minted into the join token itself.

• Ask: can a host mute, spotlight, lock the room, and remove a participant — and does the waiting room hold joiners as receive-only viewers until admitted?
• Orbit's answer: the host controls cover mute, spotlight, lock, and removal; waiting-room admission is stamped into the join token, so a joiner held in the lobby cannot publish until a host admits them. The per-session audit trail covers admission, mute, lock, and removal — the property a regulated program (telehealth, financial advice) has to reconstruct later.
• Red flag: moderation implemented client-side only, so a join with a crafted token bypasses the lobby.

Checklist item 5 — AI participants, if the program goes there now

An increasing share of video programs add a digital participant — onboarding walkthroughs, intake, concierge recaps. That is a different capability from raw media transport and it decides the vendor list early if your roadmap includes it.

• Ask: can an AI agent join the room as a rendered, talking participant, dispatched per call?
• Orbit's answer: yes. AI video avatar agents bind a conversational agent to a rendered talking-head persona that joins a live room as a digital host; renders and joins are the same moderated room the human participants use. The avatar agents explainer walks the persona registry and join plan.
• Scoring note: a dedicated video API that moves raw media can never answer yes here — if AI participants are on the roadmap, weight this item before the price column.

How to score the checklist against the vendor field

Run the five items as numbered gates: a vendor that cannot answer item 1 (real embed) is out regardless of everything else; items 2–4 separate real operations tooling from demo features; item 5 only gates when the program asks for it. The four RTC specialists on the shortlist share the embed-and-room core, so scoring usually comes down to items 2–4:

Orbit's entry in that table is the one-account posture: the video room, its recordings, and its moderation sit on the same pay-as-you-go account as voice, SMS, WhatsApp, RCS, email, and the AI agents — one vendor bill rather than a video vendor stacked next to the message and email vendors.

Frequently asked questions

Does a video API need to be WebRTC-based to take the checklist?

No. The checklist gates on capability, not protocol. WebRTC is the transport most in-browser video APIs converge on, but a vendor that embeds rooms, operates recordings, enforces moderation in the token, and supports co-browse passes regardless of what carries the media. Gate on the four items, then ask about transport.

Should the checklist rank vendors or just gate them?

Gate first, rank second. Any vendor failing the embedded-room item leaves the shortlist; ranking among the passers then happens on the operating rows (recording, co-browse, moderation) and finally on price — which the RTC pricing-models post works through. A ranked table built before the gates pass invites a priced vendor that cannot embed.

How does this checklist relate to the evaluation guide?

The RTC/video evaluation guide runs the full criteria battery — room model, recording, moderation, SFU ownership, pricing posture, channel adjacency — in longer form. This checklist is the short gate version: five items, numeric ranking among passers, per-vendor head-to-heads linked at the end.

Published 21 September 2026.