ContextPrompt AI

About

ContextOS turns your daily browsing into a structured, searchable knowledge base — and makes it available to any AI conversation.

  🔹 CAPTURE
  • One-click page capture: title, URL, body text, metadata, Open Graph, JSON-LD
  • Three capture depths: Lightweight (save tokens) · Standard · Deep (full content)
  • Batch capture all open tabs at once
  • AI chat extraction from 12 platforms
  • Text selection toolbar: select text → floating bar → save / AI summarize / translate / copy as Markdown
  • Auto-capture by URL pattern matching with notification confirmation
  • Keyboard shortcuts: Ctrl+Shift+C to capture, Ctrl+Shift+P to generate prompt
  • Right-click context menu: capture page / selected text / link

  🔹 KNOWLEDGE GRAPH
  • NLP keyword extraction with automatic tagging (TF-IDF scoring)
  • Vector embeddings: all-MiniLM-L6-v2 (384-dim), computed in-browser via Transformers.js
  • Relationship discovery: semantic similarity, URL domain, tag overlap (Jaccard), temporal proximity
  • Interactive D3.js force-directed graph visualization
  • Hybrid search: vector similarity (60%) + BM25 keyword (40%), Reciprocal Rank Fusion

  🔹 AI INTEGRATION
  Three-tier routing with automatic fallback:
    1. Chrome Gemini Nano (on-device, when available)
    2. Cloud APIs
    3. Rule-based NLP (always available, no API needed) AI capabilities: summarization, translation, quality analysis, context fusion, prompt optimization.

  🔹 CONTEXT ORCHESTRATION
  Click "Craft Prompt" on any supported AI platform to:
    1. Search your knowledge graph for relevant nodes
    2. Include current page context
    3. Allocate token budget by model limits
    4. Compress and format into a structured context package
    5. Preview and insert into the AI chat input

  🔹 MCP PROTOCOL
  Expose your knowledge base to external AI tools via Native Messaging Host:
  • search_knowledge — semantic search your knowledge graph
  • get_context — assemble context packages for target models
  • capture_page — capture current browser tab
  • list_knowledge — list recent knowledge nodes
  • get_stats — knowledge graph statistics

  🔹 WORKFLOW ENGINE
  Pre-built multi-step workflows: Academic Research, Competitive Analysis, Code Review, Content Curation.

  🔹 PRIVACY-FIRST
  • All data stored locally in your browser (IndexedDB + Chrome Storage)
  • Vector embeddings computed in-browser (model downloaded once from CDN, then cached)
  • No telemetry, no data collection
  • Cloud AI is optional — only activated when you trigger an AI action with your own API key
  • MCP Server listens on 127.0.0.1 only (localhost)

  Supports English and Chinese with runtime language switching.