We built this stack last summer while working on AI voice agents for healthcare practices.
Twilio worked well functionally, but its HIPAA option started at $2k per month before a single call, and that fixed cost dominated almost immediately, long before usage-based costs like LLM tokens mattered. That imbalance pushed us to build and operate our own telephony layer.
This repo is the result. It's a self-hosted, HIPAA-eligible telephony stack built on Asterisk + AWS Chime that handles phone numbers, SIP/TLS, RTP media, and real-time audio streaming to AI voice agents via a Twilio-like WebSocket interface. A shim server bridges audio to modern WSS-based AI services. You bring the voice AI. This just handles the phone infrastructure. (Example server using OpenAI Realtime API for voice-to-voice included in repo.)
We've since moved on to other projects, but as AI voice agents keep proliferating, we noticed we couldn't find a comprehensive, open-source, self-hostable option in this space. Rather than let it sit unused, we cleaned it up and open-sourced it.
This isn't for everyone. If you don't want to run infrastructure, Twilio is still the easiest option.
Happy to answer questions about architecture, trade-offs, scaling, or design decisions!
---
Blog post: https://vectorly.app/blog/open-telephony-stack
