Tags

Chrome ExtensionTypeScriptGemini APIGroq APIGemini Nanoaichromeprivacyllm

PromptSmith is a Chrome Extension that intelligently polishes your prompts before you send them to AI agents like ChatGPT, Claude, or Gemini. It features a Smart AI Mode that automatically chooses the best privacy/performance balance—using cloud APIs for speed or Gemini Nano for 100% on-device processing.

Key Features

  • Smart AI Mode: Automatically selects Cloud API for speed or Local Nano for privacy based on what's available on your system.
  • Multi-Model Support: Gemini 2.0/3.0 (Google) and Llama 3.1/3.3 (Groq) are supported out of the box.
  • Four Personas: Choose between Polisher, Developer, Thinker, and Custom prompt transformation modes.
  • Native Integration: Injects a polished button directly into ChatGPT, Claude, and Gemini interfaces.
  • Fully Customizable: Edit system prompts per-persona or create your own from the settings page.
  • Privacy-First: Zero data collection. API keys are stored locally in Chrome Sync.
  • Prompt History: Track your last 10 polished prompts with timestamps.

Privacy Architecture

PromptSmith is designed with a privacy-by-default philosophy. There is no middleman—we do not run a backend server. Your data goes directly from your browser to the AI provider, or stays entirely on your device when using Local mode.

Data Flow Options

1. Google Gemini API (Cloud): Your prompt is sent via HTTPS directly to Google's servers.

2. Groq API (Cloud): Your prompt is sent via HTTPS directly to Groq's servers.

3. Gemini Nano (Local): Your prompt never leaves your device. Processing happens entirely in Chrome.

API keys are stored in `chrome.storage.sync`, which is encrypted by Google when Sync Passphrase is enabled. The extension is fully open source, so you can inspect the code to verify no analytics or telemetry are sent.

Smart AI Mode Logic

The extension uses a tiered system to determine how to process your prompt:

1. Cloud Mode (Priority 1): If an API key is configured (Gemini or Groq), the extension uses cloud processing for the fastest results (~1 second) with the smartest models.

2. Local Mode (Priority 2): If no API key is set but Gemini Nano is available in your Chrome browser, the extension processes prompts entirely on-device. This is 100% private and completely free, though slower (3-8 seconds).

3. Fallback (Priority 3): If neither option is available, the extension prompts you to configure settings.

Supported Cloud Models

Google Gemini:

  • Gemini 3.0 Flash (Newest, Experimental)
  • Gemini 2.5 Flash (Balanced)
  • Gemini 2.0 Flash (Reliable)

Groq (Llama):

  • Llama 3.3 70B Versatile (Smartest Open Source)
  • Llama 3.1 8B Instant (Fastest)
  • GPT-OSS 120B (High Reasoning)
  • GPT-OSS 20B (Efficient Mid-Range)

Persona System

PromptSmith offers four specialized personas for different prompt transformation needs:

  • Polisher (✨): Focuses on grammar and clarity. Produces professional tone, concise and readable output.
  • Developer (💻): Optimized for code and data tasks. Encourages structured output (JSON, XML) and edge case handling.
  • Thinker (🧠): Designed for reasoning tasks. Uses Chain of Thought prompting and step-by-step verification.
  • Custom (⚙️): Fully customizable from the settings page. Define your own transformation logic.

You can edit the system instructions for any persona in the Options page, and changes are saved per-persona.

Requirements

For Local Mode (Gemini Nano):

  • Chrome version 128 or higher
  • Windows, Mac, or Linux
  • Approximately 2GB of disk space for the model download
  • Works offline after initial setup
  • Completely free

For Cloud Mode (API):

  • Any modern Chrome version
  • Internet connection required
  • Free tier available, or pay-as-you-go pricing