Matrix 2.0 — How we're making Matrix go voom!

The Matrix team spent 15 years building SIP stacks before we gave up and created Matrix, so I can answer this with some confidence: Superficially, Matrix looks a bit like SIP: for 1:1 calls, you send an m.call.invite (like a SIP INVITE) to someone; they answer with an m.call.answer (like a SIP 200 OK); eventually someone hangs up with an m.call.hangup (like a SIP BYE). However, the differences are: * As a transport, everything goes over normal Matrix signalling (by default HTTPS+JSON) rather than SIP's mix of UDP and TLS sockets. As a result, no need for SIP's three-way handshakes inherited from its UDP transport * As a result, you inherit Matrix's end-to-end-encryption and decentralisation for free (so no special Routes, Vias, Record-Routes, branch parameters etc from SIP - it uses the Matrix client-server and server-server APIs over HTTPS instead) * Everything is trickle ICE by default for rapid call setup, no need to wait until you have all the ICE candidates to proceed with the call * No offerful/offerless invites: everything is offerful. * Matrix piggybacks on WebRTC for its media protocol, so you don't have the fragmentation of different media transports that SIP has inherited * Matrix (as of [link] ) now supports multiparty native VoIP calls in the same conversation: effectively letting you signal full-mesh, SFU and MCU style multiway video/voip using the same mechanism as you'd use for a 1:1 call. This is probably the biggest difference in the end: with Matrix's VoIP

Element Call ( [link] ) is (very) late stage beta - it doesn’t currently use SFUs although we demonstrated one (waterfall) at FOSDEM: [this video] The thing keeping it in beta has been a set of really annoying intermittent bugs, which we’ve ended up tracking down by adding full opentelementry support; should expect to exit beta in the next few weeks (currently gathering data to check we really did fix them). Meanwhile we’re experimenting with other SFU implementations (eg LiveKit) to see if we can avoid reinventing the wheel there. Element Call style VoIP is already proposed for the Matrix spec - MSC3401 and MSC3898.

P2P Matrix is a dialect of Matrix where the server runs inside the client itself - see arewep2pyet.com for details. In other words, there are no servers (other than optional store-and-forward relays). It's insanely cool - you can see a demo at [this video] . It's also unfunded and on hiatus until someone provides some $ so we can work on it again. Right now, normal Matrix is a client-server model: you can't send messages from a client without it talking to a server (and then to another server, and then to another client). MatrixRTC VoIP and Video calls in Matrix 2.0 also go via server (for now), in order to support multiple participants and firewall traversal. Obviously, both messages and calls are end-to-end-encrypted (other than in public chatrooms), which makes it less important that they go via a server today.