Felona Voice: sub-10ms voice AI agents for TypeScript

Felona Voice is an open-source TypeScript framework that replaces generative LLMs with Joint Embedding Vectors and a stateful VoiceGraph for intent matching. It resolves user intent in about 5ms, removing LLM latency, hallucinations and token costs.

• Intent resolution in ~5ms versus 500–1200ms with LLM pipelines
• Uses embedding similarity and VoiceGraph instead of text generation
• Deterministic responses with no hallucinations or token costs
• Open-source TypeScript framework with a fluent builder API