🤖 05. AI Services

Configure models and understand how your data is protected.
FlashNote integrates multi-model AI capabilities designed as a creative copilot, while strictly preserving local-first principles and stringent data privacy.

📑 In This Guide

1. AI Capabilities Overview
2. Configuring AI Providers
3. Base URLs, Models & API Keys
4. Common Connection Errors
5. Privacy & Data Boundaries

1. AI Capabilities Overview

FlashNote integrates AI precisely at high-value authoring touchpoints:

• ✨ Single-Round-Trip FrontMatter Auto-Extraction:
  • Click “AI Extract” in the toolbar;
  • A single API call extracts a concise 150–280 word summary and automatically identifies core entities as tags (e.g., #TypeScript, #Architecture);
  • Cuts API token costs and latency by 50% compared to split requests;
• 🌐 Automatic Language Matching: Generated summaries and tags automatically match the note’s primary language;
• 📝 Smart Continuation & Polish: Select any paragraph for grammar correction, academic tone polish, translation, or expansion;
• 🔄 429 Rate-Limit Failover: Automatically switches to a configured healthy fallback model if your primary provider hits quota limits.

2. Configuring AI Providers

Open “Settings” -> “AI Service Configuration” in the bottom left corner to configure your preferred provider:

1. Cloud Foundation Models:
   • Google Gemini: Gemini 1.5 Pro / Flash (fast response and generous free tier);
   • DeepSeek: DeepSeek-V3 / R1 for high price-performance;
   • OpenAI: GPT-4o / GPT-4o-mini and compatible proxies;
   • Anthropic Claude: Claude 3.5 Sonnet for articulate prose;
2. Local Offline Private Models:
   • Ollama: Run Llama 3, Qwen 2.5, or Mistral locally on your machine with 100% data isolation;
   • Chrome Built-in AI (Prompt API): On-device Gemini Nano inside Chromium with zero latency.

3. Base URLs, Models & API Keys

Configuration fields explained:

• Provider: Choose your platform or select “Custom OpenAI-Compatible”;
• Base URL:
  • Official providers use default endpoints;
  • For reverse proxies, enter e.g. https://api.your-proxy.com/v1;
  • For local Ollama, use http://localhost:11434/v1;
• API Key:
  • Paste your secret key;
  • Storage Security: Stored exclusively within Chrome’s local protected sandbox with encryption. Never transmitted to Aicrosoft servers;
• Model:
  • Enter the model identifier (e.g., gemini-1.5-flash, deepseek-chat, or gpt-4o-mini);
• Fallback Model (Optional):
  • Second-priority model for automatic seamless handover during 429 errors.

Click “Test Connection” to verify latency and credential validity.

4. Common Connection Errors

5. Privacy & Data Boundaries

FlashNote operates under strict minimal context and explicit trigger principles:

1. Zero Background Uploads: FlashNote never scans, uploads, or indexes your local notes in the background;
2. On-Demand Single Transfers: Only when you explicitly click an AI action is the selected text fragment packaged and sent directly to your configured endpoint;
3. Prompt Sanitization: Automatically strips unrelated environment variables before dispatch;
4. Maximum Privacy with Ollama: Pair with local Ollama so all computations run 100% on your local GPU/CPU even while disconnected from the Internet.