PostPolice 🛡️

PostPolice is a Chrome extension that detects and verifies claims on web pages in real-time. It uses a local Node.js proxy to securely handle AI requests through Groq and caches results using Valkey for high performance and low latency.

Prerequisites

• Node.js (v18 or higher)
• Homebrew (for installing Valkey)
• Valkey (A high-performance Redis-compatible data store)
• Groq API Key (Get one at console.groq.com)

Getting Started

1. Install & Start Valkey

# Install Valkey
brew install valkey

# Start the service
brew services start valkey

2. Project Setup

Clone the repository and install server dependencies:

cd PostPolice/server
npm install

3. Configure Environment

Create a .env file in the root of the project:

# In PostPolice/.env
GROQ_API_KEY=your_api_key_here

4. Start the Cache Server

Run the local bridge server:

cd PostPolice/server
npm start

The server will be running at http://localhost:3000.

5. Install the Chrome Extension

1. Open Chrome and navigate to chrome://extensions.
2. Enable Developer mode (top right).
3. Click Load unpacked.
4. Select the PostPolice directory (the root folder containing manifest.json).

Features

• Real-time Summarization: Automatically extracts verifiable claims from pages.
• Fact Verification: Checks claims against whitelisted news sources using Groq LLM logic.
• Semantic Verification Cache: Skips redundant AI calls by matching claims semantically (>95% similarity) using local vectors.
• High Performance Caching: Uses Valkey to store summaries and verification results for high efficiency.
• Secure Handling: API keys are stored server-side and never exposed to the client.

Semantic Verification Cache (Local AI)

PostPolice uses a dual-layer caching strategy to minimize AI costs and latency:

1. Exact Matching: Checked via SHA-256 hashes of the page content.
2. Semantic Matching: Uses the HuggingFace Transformers.js library on the server to generate 384-dimensional embeddings for every claim.
   • Model: Xenova/all-MiniLM-L6-v2 (Runs entirely locally).
   • Logic: If a claim being verified is 95% semantically similar to a previously verified claim, the server reuses the cached verdict instantly.

Monitoring & Management

Visit the built-in dashboard to monitor cache performance: 👉 http://localhost:3000/metrics

From the dashboard, you can:

• View Hit/Miss rates, Semantic Hits, and memory usage.
• Clear Cache: Instantly purge all cached data.
• Reset Stats: Zero out the performance counters.

License

MIT