I usually test performance. First during the build via cli. Then twilio/telephony.

Looks like I opened up a can of worms sort of speak. First let me go over my tech stack and hopefully I don’t leave anything out: Hetzner sever Ultravox Ai Twilio LLM- could be open source like ollama or not open source like OpenAI,etc.im currently using open Ai model. Side note: I’m currently using Wispr flow speech to text dictation for this post. It’s amazing time saver. I forgot to mention Claude code is my partner in crime!haha. I build out my voice agents in the cli. I find it to be fast, robust. My take on ultravox ai : I believe the voice quality is outstanding. The turn detection also stands out to me.. Im currently testing out mem0 for memory. I’m still in testing phases. I’m not discounting the other platforms; livekit, pipecat. I do like them both as well. I love the idea owning the code and building the infrastructure rather than paying for platforms like Vapi, retell ai, bland, etc. cause they control cost and features. So I’m grateful for this community and look forward to learning as much as possible.

@Dan Quixote Have you tried using sub-agents in your projects for both to reduce cost and limits?

Thank you for this. I will check ii out. Thanks again

I’m interested in hearing what is working really well for others in this space.