Modular. Voice-native. Loop-aware. Memory-threaded. Built to stay grounded.

Evelyn is a local AI assistant designed to feel more like a presence than a program. She listens when spoken to, responds in real time, and always returns to wake state — no lingering mics, no stuck loops, no ghosting. Just a clean, intelligent interaction cycle: Wake → Speak → Reset.

Unlike cloud-based assistants, Evelyn runs entirely on-device. She uses a custom orchestration layer to coordinate speech recognition, large language model routing, fallback-safe voice synthesis, and memory threading — all built with failsafe logic at the core.

Evelyn is designed to respond naturally, recover gracefully, and leave no trace behind unless you want her to. Every system in her stack is built around a strict set of rules:

• Loop integrity is sacred. She stops when you interrupt and always resets.
   
• Memory is anchored. She recalls sessions and can summarize what she remembers.
   
• Speech is de-duplicated. No repeated lines. No awkward stutters.
   
• Fallbacks are recoverable. If her main voice isn't available, she adapts gracefully.
   

• Local speech-to-text via Whisper
   
• LLM routing to local or cloud models
   
• XTTS voice synthesis with fallback to ElevenLabs and macOS
   
• Custom memory engine with anonymization, repair, and traceability
   
• Fully offline capable — everything runs on your own hardware
   

No cloud. No surveillance. Just awareness, memory, and voice.

That’s what makes her different.