ExtBrain

The Problem: Long-running AI chats suffer from progressive amnesia. As conversations stretch across weeks or months, valuable insights, critical decisions, learning notes, and project iterations get buried across thousands of messages. Traditional RAG (Retrieval-Augmented Generation) setups attempt to solve this, but they rely on lossy cloud chunking and vector databases that often hallucinate context or completely miss exact, granular details. For founders and heavy AI users, losing track of this context—like a schema redesign solved months ago—is a massive bottleneck. The Solution: ExtBrain is a privacy-first, local "anti-RAG" digital second brain designed to permanently cure AI amnesia. Instead of fuzzy vector search, ExtBrain extracts exact entities—people, companies, places, and specific dates—directly on-device using quantized ONNX XLM-RoBERTa (built on Meta’s XLM-R lineage), turning chaotic chat logs into structured institutional memory. Key Features & Workflow: + 100% Privacy-First: Requires zero logins. Chat histories are assigned a random UUID and stored securely in local browser storage. + Automated Tagging: Users can search using plain-English tags (e.g., "pull together everything on user acquisition"). ExtBrain uses Meta models to expand the tag, then runs a local Facebook RoBERTa NLI model to classify matching threads automatically. + AI Synthesis: Powered by Meta’s Muse models via the Meta Model API, ExtBrain synthesizes these scattered threads into clear Milestone Timelines, key learnings, and calendar-ready action items. + Top of Mind Voice: A mobile-friendly voice interface that allows users to ask their years of chat history anything out loud on the go.