Cloudflare Calls Turns the Network Into a WebRTC Super Peer
Real-time video and audio apps have moved from a niche capability to a baseline expectation. Users now assume that group calls, interactive streams, and remote collaborations will work with near-zero latency, no matter where participants are located. For developers, however, building the underlying infrastructure for these experiences remains a heavy lift, consuming time that could otherwise go into product differentiators.
Cloudflare Calls is a new offering that addresses this gap. It exposes a set of APIs for building real-time audio and video applications on top of the WebRTC protocol, letting developers create a wide range of products: video conferencing with custom UI, moderated panel discussions, privacy-first fitness classes where only the instructor sees all participants, or fireside chats broadcast to audiences of 10,000+ with sub-100ms delay.
The core idea behind Calls is to position Cloudflare's network as a super peer in the WebRTC topology. This abstracts away much of the complexity traditionally associated with scaling real-time communication.
Moving Beyond Traditional SFUs
WebRTC is a peer-to-peer protocol that allows devices to communicate directly through the browser. For a simple call between two people, this works well. Adding more participants, however, rapidly introduces reliability problems. The coordination overhead between devices increases, and the data consumption of each participant grows as they must send media to every other person on the call. This often results in video freezes, desynchronization, and an overall poor experience.
To solve this, many applications turn to a selective forwarding unit (SFU). An SFU routes video and audio between participants, reducing the data load on each individual device. SFUs also become necessary when an application needs to determine who is actively speaking or when using WebRTC simulcast to send appropriate resolution video.
But a centralized SFU comes with its own inherent weaknesses. It lives in a specific region, making it fast for users nearby and slow for everyone else. Public cloud SFUs are also bandwidth-intensive and present significant devops challenges, requiring teams to manually manage scaling across regions.
Recognizing that SFU-as-a-service alone would not be enough, Cloudflare built Calls on a different architectural premise: treat the entire network as the server.
Region: Earth
In a typical five-person WebRTC call, each participant is known as a peer, and the call's reliability is limited by the participant with the weakest connection. Calls flips this model. Each person connects to their nearest Cloudflare data center instead of to each other. When one participant's media is requested by another, the data is forwarded between data centers automatically. Participants in close proximity will share the same data center, avoiding the latency of a round trip to a centralized region.
This eliminates many of the operational questions that plague teams running their own WebRTC servers, such as which regions to support or how to scale for unplanned spikes in usage. With Calls, every connection is anycast-based, routing to the closest Cloudflare location by default. The service scales with demand without requiring custom auto-scaling code. Cloudflare claims its network is within 10 milliseconds of 90% of the global population, so adding Calls does not introduce noticeable latency.
Observability from the Server Side
Troubleshooting WebRTC issues is a pain point for many developers. A failed web page can be reloaded; a disrupted video call often ends the conversation. Calls includes a server-side view of the WebRTC Statistics API, a standard interface developers already use in browsers to monitor connection performance. This allows individual RTCPeerConnection objects and media flow to be inspected without relying solely on client-reported data, giving developers end-to-end visibility across the network stack.
The service is a viable alternative for teams currently operating their own WebRTC servers like Janus or MediaSoup, particularly those with clients spread across the globe connecting to a centralized deployment.
Privacy and Security by Default
Traditional WebRTC calls expose participants to each other, including sensitive data like IP addresses. This can leave users vulnerable to denial-of-service attacks. Calls provides a layer of abstraction: all participants connect to the Cloudflare network rather than to each other. From a user's perspective, the experience remains peer-to-peer, but with added privacy and security benefits.
All video and audio passing through Calls is encrypted by default, leveraging Cloudflare's existing infrastructure products like Argo for secure and efficient routing. The API, which implements the WebRTC RTCPeerConnection interface across the network, also exposes granular controls not possible with vanilla WebRTC, allowing developers to push media tracks and request others within a shared Peer Connection context.
Availability and Next Steps
Cloudflare Calls is launching in closed beta today. Those interested can request an invitation, and the service will be free during the beta period. Cloudflare says it is seeking early customers to help bring the product from beta to general availability, particularly those building real-time video applications or facing scaling challenges with traditional WebRTC infrastructure.
