Hey everyone! I’m Fraser, the creator of RoohAI. I built RoohAI because I was tired of 'soulless' voice agents. You know the ones: they treat a thoughtful silence as an invitation to interrupt, or they bulldoze through your sentences because they can't handle a simple barge-in. Most voice stacks are just a mess of glue code that treats conversation like a file transfer rather than a human interaction. Rooh (which means 'soul') is a Python-native framework designed to solve the three pillars of voice: latency, interruption handling, and modularity. I built this specifically from my experience interacting with a Skill Interview Agent. I realized that if an agent doesn't understand 'thinking time' vs. 'finished speaking,' the illusion of intelligence breaks immediately. RoohAI handles that heavy orchestration so you can focus on the actual conversation logic. It’s 100% open-source and modular. Want to swap Deepgram for Whisper or ElevenLabs for? It’s one line of code. I’d love your feedback on: The 'Silence' Problem: How do you handle the gap between a user 'pondering' and a user being 'finished'? Provider Support: Which STT/TTS models are non-negotiable for your stack? The Pain: What’s the biggest 'gotcha' you’ve hit while building voice apps? I’ll be here all day to chat. Check out the repo, break things, and let’s build voice agents that actually listen.

RoohAI (Soul) for your Voice Agents

Empathetic, patient, and context-aware Voice Agents/Robots

Empathetic, patient, and context-aware Voice Agents/Robots