HWUI Pro launches with memory, web search

The interesting part here is not “local chat” by itself, it’s the attempt to fix two of the category’s weakest points: persistence and grounding. If it works as described, it moves beyond a thin UI for a local model and into actual assistant territory.

• –Per-character memory that the model can update mid-chat is the right abstraction for roleplay and long-running personas
• –Web search with honest fallback behavior is a practical answer to local models hallucinating on unknown questions
• –Project folders and document injection make it feel closer to a workspace than a single chat window
• –Multi-engine TTS and local Whisper input widen the use case beyond text-only power users
• –The remaining risk is adoption: this is still a fairly specialized local-AI stack, so the live demo will matter a lot for trust and usability