Comcent CE - an open-source voice/contact-center platform on Elixir/OTP, with call queues modeled as processes

I’ve been building Comcent CE, an open-source, self-hostable voice/contact-center platform, and Elixir/OTP is the core of it. Sharing the design here since the process model shaped the whole architecture, and the “why Elixir” reasoning might be useful to others working on similar systems.

Why Elixir, concretely:

1. The actor model let the design mirror the mental model. Just like a real person managing a queue, a process is spawned whose sole job is managing that call queue. You reason about the system the same way you’d reason about people doing the work — the domain model and the code line up almost one-to-one, which made the hard parts (transfers, requeues, routing) far easier to implement.

2. Built-in distribution made scaling out the easy part. Each queue needs exactly one manager process — and that “exactly one” guarantee has to hold across the entire cluster, even as the number of queues grows into the thousands or millions. Elixir’s clustering plus a distributed registry gives us a cluster-wide singleton per queue: only one Queue Manager runs for a given queue anywhere in the cluster, and if the node hosting it goes down, it’s restarted elsewhere while staying unique. We get horizontal scale without changing the mental model at all — the same process abstraction that made single-node reasoning simple is what makes the distributed version work.

The rest of the stack:

• Phoenix — APIs, the real-time layer, and the per-call timeline

• FreeSWITCH — the media/SIP engine; Elixir drives call control over FreeSWITCH’s HTTP API (mod_httapi), with the Event Socket (ESL) in some cases

• Go — small services powering the SIP-aware voice bots

• RabbitMQ — event pipeline between telephony and processing

• Postgres — calls, transcripts, vCon records · Redis — caching/ephemeral state

• SvelteKit frontend with an embeddable browser dialer

On top of this it produces full call timelines, dual-channel recording, diarized transcripts, AI summaries, and implements the IETF vCon standard.

GitHub:

Walkthrough (bare VM → first call):

Website: