Infrastructure for multimodal AI
People and agents trade video, audio and data, each over one QUIC connection. Anything else that needs the stream subscribes to it.
WebRTC was built for people calling people. This is built for everything listening in.
switch tracks, or a whole side, on the caller and agent boxes; lines run in the same order as the tracks in each box
What's included
full control over streams and clients| Easy connection setup | No rooms. Clients connect once and subscribe to what they need; nothing to create, join or tear down. |
|---|---|
| Tunable for every listener and every track | Each subscriber gets its own trade-offs on the same track. See below. |
| Authentication and scoped access | Included. Each listener's token covers only its tracks and expires on its own. |
| Subscribe-based streaming | Nothing is sent until something subscribes, so the transport carries only what is used. |
You decide how every track travels
Set encoding and packaging for each track. Rank tracks, so a tool result goes before a video frame. Decide what is dropped first and what happens during congestion. Send generated speech faster than real time and let the player buffer ahead. Every subscriber can get the freshest frames or every frame, from the same track.
Under the hood
Our relays, or yours
Caprid runs routing, access and the keys. If media can't leave your network, bring your own relays; the control plane stays with Caprid.
Extend your capabilities with operators
Attach operators to stream events to do whatever you need: transcribe, moderate, record, or package the stream as low-latency HLS for screens that can't run a real-time client.
Built by the team behind Mux, on open standards
From the founders of Mux, who built Video.js. On QUIC and Media over QUIC, which is being standardized at the IETF.