Stop reading about real-time. Open it.

2026-08-28 · General

Five finished real-time products you can open in a browser tab with no account, an assistant that answers with citations from the page you are already on, a studio that turns a description into a runnable React project, and documentation that finally shows the pipeline instead of describing it.

Stop reading about real-time. Open it.

Most infrastructure sites ask you to believe a feature list. We would rather you pressed something.

Each of the rooms below is a real MediaSFU session. It starts when you press it, it costs nothing, it needs no account, and it closes when your tab does.

Live Auction Salesroom

Server-ordered bidding, so two people bidding at once still produce one winner. A late bid extends the clock, and the lot settles only when reserve is met.

Broadcast Control Room

Preview and program monitors, a four-source matrix, and a take that actually cuts to air. Arm Camera 2, take it, and the rundown moves with you.

Watch Party

Everyone on the same frame within about 0.2 seconds, a floor you can hand to someone else, and reactions on the stream. WHIP ingest, WebRTC or HLS out.

WebRTC Stage + Synced HLS Audience

Speakers on WebRTC, the audience on HLS, one clock across both. Promote someone from the audience and they switch from watching a stream to publishing media, mid-session.

Familiar Chat Call

A private two-person call from a single link. No install, no account, no room password to leak.

Interactive Classroom — in progress

The tools are already live: shared whiteboard, polls, breakout rooms, hand-raise and moderation all ship in every MediaSFU meeting today. Only the guided showcase is still being built.

Open one and see

No signup, nothing installed. Every room closes when you close the tab.

Guided tours that point at the control

Every showcase now opens with a tour that names the verb — press this, then this — rather than describing the region you are looking at. We rewrote them this cycle and roughly doubled their length.

The old tours stopped before the part that matters: inviting a second person. A room with one person in it does not demonstrate anything. Bidding against yourself proves the clock works; bidding against someone else proves the ordering does.

Ask MediaSFU, on the page you are already reading

Documentation answers the question you already knew how to ask. Ask MediaSFU is for the other kind — the one where you are not yet sure which product, which SDK, or which delivery path you are supposed to be choosing between.

It sits docked on product, documentation, pricing and dashboard pages, and it reads the page you are on. Open it on the telephony guide and it offers SIP trunk and handoff questions. Open it on the live pages and it offers HLS versus WebRTC, WHIP and WHEP. Open it on widgets and it starts from the six production widgets. There is a full workspace at mediasfu.com/copilot when you would rather sit with it than have it float over the page.

Cited, or it does not answer

Evidence answers require canonical MediaSFU sources, and those sources stay visible. Where something is guide-only rather than a live path, the answer says so instead of smoothing over it.

It knows which page you are on

Meetings, telephony, agents, human handoff, live, recording, translation, AI notes, widgets and pricing each bring their own starting questions.

Knowledge, not your account

Account lookups and account actions are disabled by design. It explains the product; it does not reach into your data to do it.

When an answer carries enough to plan from, it offers Plan this in Vibe Studio. Only your own words and the canonical operation IDs travel across — not the assistant’s prose, not your account state, not any borrowed authority. Vibe Studio then re-validates that input through its own clarification and review gates rather than trusting the handoff. Ask, plan, export, open.

Vibe Studio: describe the room, leave with a project

Vibe Studio is live. Describe what you want in a sentence, or start from one of seven realtime patterns. Review the plan in plain language. Then export a React project — a frontend plus a room backend, with your credentials kept server-side rather than shipped inside browser code.

Say it in a sentence, or pick a pattern. Your words stay in your browser.

Session length, roles, recording and delivery become a plan you can read.

Check it over before anything is generated. Nothing is created until you ask.

An installable React project, no keys inside it.

Every pattern is labelled with what it can actually do for you:

• tested with real media — taken through staging with genuine audio and video
• builds and runs — compiles and works, not yet media-tested
• plan only, no export — the plan exists, the generator does not yet

We would rather tell you which rung a pattern sits on than let you discover it after you have built on it.

Plan one now

Planning and export need no account. Opening a live session stays behind sign-in, on purpose.

Documentation that shows the thing

🧠 AI Pipeline Guide

A pipeline spine now sits above every section — enter, listen, think, speak, hand off — so you always know which stage of an agent turn you are reading about. The credentials precedence rule, previously three paragraphs, is a three-row table answering the only question anyone had: when a value exists on the dashboard and in llmParams, which one runs?

mediasfu.com/agents

☎️ Telephony Guide

A call-path diagram that makes plainly the point prose kept fumbling: inbound and outbound are not separate integrations. The same trunk and the same room carry both, which is why configuring one configures most of the other.

mediasfu.com/telephony

Starter repos, per showcase

Every showcase now links to the code behind it, and the starters are refreshed.

• mediasfu-live-auction — the salesroom above
• mediasfu-watch-together — the watch party
• MediaSFU-WhatsApp-Calls-MultiSDK — the familiar call pattern
• Agents — voice, vision and multimodal agent starters
• MediaSFU-QuickStart-Apps, VOIP, SpacesTekFinal, mcp-sdk-tools

Ten open-source SDKs, including Apple, Kotlin, Unity and embeddable widgets.

What it costs to try

50,000 free build minutes — 20,000 production and 30,000 sandbox.

Separately, 100,000 demo minutes for rooms you create from the dashboard. Those do not apply to SDK or API usage, so we list them apart rather than folding them into the headline.

Published rates after that: $0.002 per agent minute and $0.0001 per voice transport minute. Bandwidth, storage, egress and AI providers are billed separately — by us and by every other vendor — so they are excluded from every comparison we publish.

If you want a room nobody has built before, you build it. Take the SDKs, go headless, wire your own surface. That path is slower — and it is still cheap.

Start where you like

Open a finished room, plan one in the studio, or clone a starter and run it locally.