Hey FreePBX community,
I’m Anil, and I’ve been working on a voice AI gateway that bridges FreePBX/Asterisk to modern AI services (STT, LLM, TTS) without ripping out your existing telephony infrastructure.
What is VEXYL?
VEXYL is a self-hosted middleware that sits between your Asterisk/FreePBX system and AI providers. It connects via AudioSocket protocol and handles the entire voice AI pipeline - speech recognition, LLM processing, and text-to-speech - while your PBX continues managing call routing, billing, and all the telephony logic you’ve already built.
Why I Built This
After deploying voice AI for healthcare systems in India, I realized there’s a massive gap between “expensive cloud per-minute platforms” and “build everything yourself.” Organizations with existing PBX investments shouldn’t have to choose between $0.15-0.40/min cloud costs or spending months building custom integrations.
Key Technical Details:
-
Protocol: Asterisk AudioSocket (also planning SIP support)
-
Architecture: Redis for session/cache
-
Response Times: 2.2-3.3 seconds end-to-end (STT→LLM→TTS)
-
TTS Caching: 90% hit rate, reduces repeated responses to 2ms
-
Providers: 17+ AI providers supported (bring your own keys)
-
Languages: 100 + 10+ Indian languages (Malayalam, Hindi, Tamil, etc.)
-
Deployment: Self-hosted binary or Docker, complete data sovereignty
Real-World Usage:
Currently processing 1000+ patient interactions monthly for healthcare systems - appointment reminders, callback confirmations, all in local India language. 95% satisfaction rates.
Two Operating Modes:
-
Standard Mode: Full pipeline (STT→LLM→TTS) with custom data integration
-
Gateway Mode: Direct audio streaming to OpenAI Realtime API, ElevenLabs
Current Capabilities:
-
Inbound call handling
-
Outbound calling with pre-warmed greetings (eliminates 3-5 sec initial delay)
-
Context-aware conversations
-
Multi-provider failover
-
Circuit breakers and retry logic
-
TTS response caching
-
Session management across transfers
On the Roadmap:
-
FreeSWITCH compatibility
-
FCC SIP 603+ compliance (March 2026)
Pricing Philosophy:
Tiered licensing based on concurrent calls (10/20/50 channels), with BYOK model. Goal is 87-91% cost savings vs building in-house, 95% vs custom dev.
Why Post Here?
The FreePBX community understands telephony infrastructure better than anyone. I’d love feedback on:
-
Integration patterns you’d want to see
-
Use cases you’re encountering
-
Technical concerns about AudioSocket vs other approaches
-
Provider preferences for STT/TTS/LLM
Not Here to Spam
I’m genuinely looking for feedback from folks who’ve “been there, done that” with PBX systems. If this doesn’t fit your needs, no worries - but if you’re exploring voice AI integration, happy to discuss technical approaches.
Website: https://vexyl.ai
Would love to hear your thoughts, especially from anyone who’s tackled similar integration challenges.