NovaCue: LLM Preference Learning

BETA TESTING — NovaCue: Automatic preference learning across ChatGPT, Claude, Gemini, and Grok.

AI remembers you — but it can't forget. When you evolve, your AI is stuck in the past. NovaCue's preferences evolve WITH you — old habits decay, new patterns reinforce, contradictions resolve automatically. No manual memory management. Your AI grows as you grow. # NovaCue - AI Preference Learning Extension > **Stop repeating yourself.** Let AI remember what you want. NovaCue is a Chrome extension that automatically learns and applies your AI preferences across ChatGPT, Claude, Gemini, and other LLM platforms. No more copy-pasting custom instructions. No more repeating your coding style every session. Just intelligent, context-aware AI interactions. --- ## 🎯 The Problem The core concept (auto-learning preferences and injecting them contextually) is unique. No competitor does this. ChatGPT's "Memory" is manual, Claude's "Projects" are static, Gemini has nothing. NovaCue's approach — automatic preference learning with semantic embeddings — is genuinely novel. AI remembers you — but it can't forget. When you evolve, your AI is stuck in the past. NovaCue's preferences evolve WITH you — old habits decay, new patterns reinforce, contradictions resolve automatically. No manual memory management. Your AI grows as you grow **AI assistants have amnesia.** Every conversation starts from scratch. Thousands of users express this frustration daily on Reddit and Twitter: > *"I'm so tired of telling ChatGPT 'use TypeScript with ESLint rules' EVERY. SINGLE. TIME."* > *"Why can't Claude remember I prefer concise answers? I've told it 50 times."* > *"ChatGPT's custom instructions are too broad. I want different preferences for coding vs. writing."* **The data backs this up:** - 16,000+ monthly searches for "ChatGPT remember preferences" - 500+ social media complaints per month - 15M+ power users experiencing this daily **The result?** Wasted time, inconsistent outputs, and constant frustration. --- ## ✨ The Solution NovaCue learns your preferences automatically and injects them contextually into your prompts. ### **How It Works** 1. **🧠 Learns Automatically** - No manual setup required. Use AI naturally, NovaCue learns what you like. 2. **🎯 Context-Aware** - Different preferences for different tasks. Python preferences for coding, formal tone for writing. 3. **⚡ Adaptive** - Current session takes priority. If you say "use blue" today, it overrides your usual "red" preference. 4. **🔄 Cross-Platform** - Works with ChatGPT, Claude, Gemini, and more. One preference system for all your AI tools. --- ## 🚀 Key Features ### **Smart Preference Learning** - **Automatic Detection** - Recognizes patterns like coding languages, color choices, formatting styles - **Type-Based Organization** - Preferences grouped by context (TikZ Diagrams, API Design, Python Code) - **Weight-Based Ranking** - Frequently used preferences have higher priority ### **Session-Aware Intelligence** - **Momentum Scoring** - Recent preferences get 5x weight boost - **Context Switching** - Automatically detects when you switch between tasks - **Conflict Resolution** - Current intent overrides historical patterns ### **Enterprise-Ready** - **Team Preferences** - Share and enforce preferences across your organization - **White-Label Support** - Custom branding for enterprise deployments - **Audit Logging** - Track all preference applications and changes - **Privacy-First** - All data stored locally, optional cloud sync ### **Developer-Friendly** - **Multi-Type Support** - Handle complex preferences (Python 3.11 + type hints + pytest) - **Token Budgeting** - Smart injection that respects context window limits - **Episodic Memory** - Vector-based search through conversation history - **Orama Integration** - Fast semantic preference matching --- ## 📊 Who Is This For? ### **Power Coders** *"Stop repeating your coding preferences"* - Consistent code style across sessions - Language-specific defaults (Python → type hints, JS → ES6) - Framework preferences (React → functional components) ### **Content Creators** *"Keep your brand voice consistent"* - Maintain tone across all AI-generated content - Platform-specific formatting (Twitter threads, LinkedIn posts) - Brand guideline enforcement ### **Academics & Researchers** *"Maintain citation consistency effortlessly"* - Automatic citation format (APA, MLA, Chicago) - Academic tone preservation - Subject-specific terminology ### **Design Teams** *"Keep your design system consistent"* - Design system adherence (colors, typography) - Style guide preferences - Tool-specific defaults (Figma vs. Adobe) --- ## 🎨 Use Cases **Scenario 1: Developer Workflow** ``` Without NovaCue: You: "Create a React component" AI: [Generates class component] You: "Use functional components with hooks" AI: [Regenerates] You: [Next day, same thing repeats...] With NovaCue: You: "Create a React component" AI: [Automatically uses functional components with hooks] ✅ Saved 2 minutes, got it right the first time ``` **Scenario 2: Content Creation** ``` Without NovaCue: You: "Write a LinkedIn post about AI" AI: [Uses casual tone] You: "Make it more professional and add hashtags" AI: [Rewrites] You: [Next post, same corrections needed...] With NovaCue: You: "Write a LinkedIn post about AI" AI: [Professional tone with hashtags automatically] ✅ Consistent brand voice, zero manual corrections ``` **Scenario 3: Multi-Context Work** ``` Without NovaCue: Morning: "Create TikZ diagram" → Manually specify: red boxes Afternoon: "Create TikZ diagram" → Manually specify: red boxes again Evening: "Create API design" → Manually specify: REST + JSON With NovaCue: Morning: "Create TikZ diagram" → Auto: red boxes Afternoon: "Create TikZ diagram" → Auto: red boxes Evening: "Create API design" → Auto: REST + JSON (different context!) ✅ Adapts to task context automatically ``` --- ## 📖 How to Use ### **Basic Usage** 1. **Use AI as normal** - No setup required 2. **NovaCue learns automatically** - After a few interactions, preferences are detected 3. **Review in Preference Manager** - Click the extension icon to see learned preferences 4. **Lock important ones** - Click 🔒 to make a preference permanent 5. **Delete unwanted ones** - Click 🗑️ to remove incorrect preferences ### **Advanced Features** - **Manual Preferences** - Add custom preferences via Preference Manager - **Type-Based Organization** - Group preferences by task type - **Session Momentum** - Recent preferences automatically prioritized - **Cross-Platform Sync** - Sync preferences across devices (optional) --- ## 🏗️ Architecture ### **Core Components** ``` ┌══════════════════════════════════════════════════════════════════════════════════════┐ │ 🔥 NOVACUE FEATURE MATRIX │ │ LLM Preference Learning System │ ╠══════════════════════════════════════════════════════════════════════════════════════╣ │ │ │ ┌─────────────────────────────────────────────────────────────────────────────────┐ │ │ │ 🆓 FREE TIER (All Users) │ │ │ ├─────────────────────────────────────────────────────────────────────────────────┤ │ │ │ │ │ │ │ 📊 CORE PREFERENCE ENGINE │ │ │ │ ├── Automatic preference detection from conversations │ │ │ │ ├── Token extraction & weighting │ │ │ │ ├── Preference proposals (pending → accepted) │ │ │ │ ├── Preference decay (0.99 rate, 0.5 threshold) │ │ │ │ ├── Session momentum tracking │ │ │ │ └── 🔒 LIMIT: 100 preferences max (prunes to 80 when hit) │ │ │ │ │ │ │ │ 🧠 VECTOR EMBEDDINGS │ │ │ │ ├── MiniLM-L6-v2 model (384 dimensions) │ │ │ │ ├── ONNX runtime with WebAssembly SIMD │ │ │ │ ├── Semantic similarity search │ │ │ │ └── Cosine similarity scoring │ │ │ │ │ │ │ │ 🏷️ DYNAMIC SEMANTIC TYPING │ │ │ │ ├── Auto-detect conversation type (TikZ, Python, Writing, etc.) │ │ │ │ ├── New type discovery via concept extraction │ │ │ │ ├── Type centroids with embeddings │ │ │ │ └── 0.65 similarity threshold for matching │ │ │ │ │ │ │ │ 🔗 MARKOV CHAIN │ │ │ │ ├── Transition matrix for type sequences │ │ │ │ ├── 70% semantic + 30% Markov blending │ │ │ │ └── Context inference for vague prompts │ │ │ │ │ │ │ │ 📝 EPISODIC MEMORY GRAPH (EMG) │ │ │ │ ├── Episode creation on each interaction │ │ │ │ ├── Sparse token extraction (top 50) │ │ │ │ ├── Salience scoring │ │ │ │ ├── Graph edges (temporal + semantic + overlap) │ │ │ │ ├── 1-hop graph walk for context retrieval │ │ │ │ └── 30-day retention (manual pruning) │ │ │ │ │ │ │ │ 🗄️ ORAMA DATABASE │ │ │ │ ├── In-memory vector search │ │ │ │ ├── Document types: episode, preference, type │ │ │ │ ├── Hybrid search (vector + keyword) │ │ │ │ └── Auto-rebuild from storage │ │ │ │ │ │ │ │ 🔄 SESSION CONTINUITY │ │ │ │ ├── Cross-LLM context detection │ │ │ │ ├── "Continue Session?" popup │ │ │ │ ├── Context curation (goal + top 3 + recent 2) │ │ │ │ ├── Auto-inject into new LLM │ │ │ │ └── 30-minute freshness window │ │ │ │ │ │ │ │ ✨ PROMPT AUGMENTATION │ │ │ │ ├── Auto-inject relevant preferences │ │ │ │ ├── Type-based preference filtering │ │ │ │ ├── Weight-sorted injection │ │ │ │ └── Non-intrusive format │ │ │ │ │ │ │ │ 👍 RLHF (Reinforcement Learning from Human Feedback) │ │ │ │ ├── Accept/Reject proposals │ │ │ │ ├── Vector alignment (+0.1 / -0.05) │ │ │ │ ├── Type cluster reinforcement │ │ │ │ ├── Rejection history (7 days) │ │ │ │ └── 0.85 similarity check for duplicates │ │ │ │ │ │ │ │ 🛡️ PII DETECTION │ │ │ │ ├── Email addresses │ │ │ │ ├── SSN (XXX-XX-XXXX) │ │ │ │ ├── Phone numbers (US/International) │ │ │ │ ├── IP addresses │ │ │ │ ├── Credit cards (with Luhn validation) │ │ │ │ ├── AWS access keys (AKIA...) │ │ │ │ ├── Bearer tokens │ │ │ │ └── Password/secret patterns │ │ │ │ │ │ │ │ 💾 STORAGE │ │ │ │ ├── chrome.storage.local (10 MB limit) │ │ │ │ ├── Cross-browser wrapper (Chrome/Edge/Firefox) │ │ │ │ ├── Cache hydration on startup │ │ │ │ └── navigator.storage.persist() request │ │ │ │ │ │ │ │ 🌐 SUPPORTED PLATFORMS │ │ │ │ ├── ChatGPT (chat.openai.com, chatgpt.com) │ │ │ │ ├── Claude (claude.ai) │ │ │ │ ├── Gemini (gemini.google.com) │ │ │ │ └── Grok (grok.x.ai, grok.com) │ │ │ │ │ │ │ │ 🎨 UI COMPONENTS │ │ │ │ ├── Popup (stats, quick actions) │ │ │ │ ├── Options page (preference manager) │ │ │ │ ├── Context switch popup │ │ │ │ └── AI Arena (new tab page) │ │ │ │ │ │ │ └─────────────────────────────────────────────────────────────────────────────────┘ │ ```