AVA is a self-hosted, open-source AI voice agent that plugs directly into your Asterisk or FreePBX system. No cloud telephony migration required. Choose from 6 production-ready configurations: OpenAI Realtime, Deepgram, Google Gemini Live, ElevenLabs, Local Hybrid, or 100% fully Local. Modular STT/LLM/TTS pipeline — swap providers without changing code. Try it live: dial (925) 736-6718.
Hey everyone 👋
I built AVA because I was frustrated that most “AI voice” solutions required replacing your entire telephony stack or relying on SaaS platforms.
I wanted something that works directly with Asterisk—something you can run on your own infrastructure, with full control over models, latency, and cost.
AVA lets you plug in any STT, TTS, and LLM (cloud or local) and run real-time, two-way conversations directly on your PBX. No external provider required.
A big focus has been:
• Low-latency conversations (not IVR-style)
• Modular architecture (swap providers anytime)
• Local AI support for privacy + cost control
Would love feedback from the community—especially:
• Voice quality and latency
• Local model performance
• Real-world use cases you’d want this for
If you run Asterisk or care about voice AI, I’d really appreciate your thoughts 🙏
User of AVA for ISP Triage, works great! Strongly encourage use of this system. I use it with integrations into Ollama vs the known AI systems, and interfacing to help identify whether or not a connection is at its quality limit, create tickets, and check billing status. This is done by knowing the phone number, and such aligning with the billing system.
how's the jitter buffer handling the llm stream? if you've solved real-time on asterisk, that's impressive. great release. @hkjarral
