Studio-quality voice rooms with a camera for each person, running on Cloudflare’s global network. Pay only for what you use, see the usage of every room, and moderate everything from your server.
All in the same SDK, with the same account and console. Beta = tested in the lab and on emulators, not yet used in production. A platform next to a tag (web, Android) = where it works today.
Rooms with a stage and an audience, with the voice processed on the device itself.
A camera for each person, up to 9 on screen. Viewers only receive what is on screen.
Video effects on the device, before encoding. A subscription per app, with a license bound to the app ID.
In-app chat. A subscription by monthly active users (MAU): plans by MAU — pricing coming soon.
Voice and video between two people or in a group, with the quality of the rooms.
Rooms that see each other (PK) and, next, a single stream for large audiences.
Animated gifts, entrance vehicles and effects ready to play in your live app, with real transparency. Try each item in a room before you buy.
Live commerce: the host shows the product and viewers buy without leaving the live. Bring your Shopify or Nuvemshop store, or open a store with us (catalogue, cart and checkout).
See what happens on each device, without asking users for logs.
Streamers go live into your rooms from their computer — a game, a capture card, a desktop studio. To your viewers they are one more person on stage: the app you already have shows them, with no new player.
Create a stream key in the console or from your server and paste two fields into OBS. OBS 30 and newer sends straight into the room over WHIP.
Our streaming app for creators: build the scene and go live straight into a room through the SDK — no OBS, no RTMP.
Beta builds are unsigned: macOS and Windows will warn before opening them, and the Windows build has not been run on a Windows PC yet. Ask for a build and we send it with the steps:
Load tests through the full path — our edge Worker, the room coordinator, Cloudflare’s media servers and TURN — with a generator that speaks the SDK’s protocol. We publish the weak spots too.
99.7% of the audio delivered; joining took 0.7 s (median).
97.7% of the audio delivered with people arriving at 4 per second.
For networks that block UDP. No rate limit reached.
From joining to hearing the room (median, from São Paulo). The room coordinator is in North America; this is the number we are working on.
Our own measurement, November 2026, on our development environment with one test site in São Paulo: a single office link tops out around 25 Mbps of received audio, so larger tests need more test sites. During every live the console shows each person’s echo, voice, device, video and effects numbers, on Android, iOS and the browser.
What we already have, what is in progress and what we don’t have yet, area by area. Updated with every release.
ILORB Signal was born inside a voice-room app in production. Every feature below solved a real problem.
Echo, noise and level handled on the device, in the right order: device echo canceller, AI noise filter and gain that only rises when someone speaks.
A muted mic and a closed camera send nothing. Whoever isn’t listening or watching receives nothing. You don’t pay for silence.
Everyone joins through the closest point and the voice crosses the world over Cloudflare’s backbone.
See who is in the room, remove someone or close the room. Kicks are enforced on the server: the user stops hearing immediately.
Live rooms, minutes per room, peak participants, mic and camera minutes and estimated cost, day by day.
Network dropped, a phone call, screen locked or the user switched apps? The room stays: a watchdog checks the audio all the time and voice is back within 1–2 s.
Every device gets the right settings without you touching anything: we calibrate from the server and the SDK applies them when the app opens. No new build, no store update.
About 188 SDK settings, plus the native audio ones, change from the server. In layers: default → device class → brand → model → user.
The SDK reads the hardware (memory, fast cores, frequency, chip) and the server decides whether the phone is low, mid or high end.
The SDK measures echo on each phone with a detector that only measures and never touches the audio. From those measurements the server decides echo cancellation per model. Today it only observes: the automatic decision is not switched on yet.
DTX and the voice gate are switched on from the server. In the lab, with DTX, an open but silent mic dropped from ~58 kbps to ~4 kbps.
Where real-time voice and video make a difference.
A TypeScript SDK for web and React Native, and a token generated on your server — the secret never reaches the phone.
Voice and video flow through Cloudflare Realtime: a distributed SFU, with no central server and no region to pick.
Each device connects to the nearest data center, chosen by the network itself (anycast).
The app only gets a short-lived token. SFU and app keys never leave the server.
You buy credit in participant-minutes: one person connected to a room for one minute. Balance and per-room usage are live in the console.
We are opening to our first customers with trial credit. Tell us the size of your app and we’ll put together a proposal.
Send us a message and we’ll open console access with trial credit.