I remember seeing on the Elixir subreddit a long time ago that a user was building a Discord alternative (Probably because Discord mines all data for profit and is closed source).
I doubt that project kept going (it’s quite ambitious after all).
But I wanted to open this thread to discuss ideas about how such a thing could be built.
Building a group chat application in Elixir should be fairly straightforward. However, the biggest hurdle would likely be the group audio.
I know Discord uses WebRTC (with their own enhancements), but I’ve also seen someone suggest using Mumble as the audio backend.
I know there is an Elixir GUI system in the works, but for now, the obvious choice would put Elixir only as the backend and something like React Native / Proton Native as the frontend.
Plus, there’s already Mongoose from Erlang Solutions. Would be interesting to just build on top of what’s already there.
That’s a great find!
Mongoose for the chat. And something like Jitsi for the voice. That would be a much quicker bootstrap.
For the voice channels, you could instantiate an always-on jitsi room that people could join and leave as they wish.
Looks even more lightweight than something like Jitsi. Thanks!
Audio room example is what it would need https://janus.conf.meetecho.com/audiobridgetest.html
Here’s an example of how to interact with janus from elixir via unix sockets. The SDP packets were sent to the elixir server via websockets, which were then relayed to janus. Making basic chats is actually already easy in erlang/elixir, I’d probably wouldn’t use mongoose …
Looks like Evasyst uses Janus for their backend and is trying to be a discord alternative.
I think the best bet would be implementing a Matrix server. It’s not necessarily the smallest project (see the spec) but in terms of functionality I think it’s great.
You might want to look into the Elixir Membrane Framework if you need to do any server-side audio processing: https://www.membraneframework.org/
Chat rooms are relatively simple (and yet matrix seems to make this extremely complex).
I think the biggest hurdle is rooms that are multi-media.
In practice, most discord “servers” have text and voice channels with the same name. So when you’re in this voice channel, you type in the text channel with the same name.
I think there’s room for improvement by simply making generic multi-media rooms that are persistent only in text.
So for example, it would be like having a Jitsi meeting that is always running and users can jump in and out of seamlessly.
I’m continuing to work on elixir integration with janus (now via websockets instead of unix sockets), and have a very preliminary example: https://github.com/janus-ex/janus_ws_example.