Skip to content

Trust OS

A “nutrition label” for the internet. Doda’s Trust OS (Web Provenance) analyzes every page you visit for AI-generated content, misinformation, and manipulation — all processed locally.

Overview

Trust OS is a native C++ component that automatically analyzes web pages for trustworthiness. It detects AI-generated text, extracts factual claims, compares sources, and tracks how information spreads across the web.

Key Features

  • AI Content Detection — ngram repetition, entropy analysis, sentence variance, readability scoring
  • Trust Score — composite 0-100% metric for every page
  • Trust Badge — URL bar shows green/yellow/red trust level at a glance
  • Claim Extraction — splits page text into factual claims, scores each for confidence
  • Source Comparison — matches claims across different URLs/domains, computes reliability scores
  • Spread Analysis — tracks how claims propagate over time with temporal visualization
  • Image Provenance — detects likely AI-generated or stock images
  • Content Fingerprinting — DJB2-hash based fingerprints for cross-site source matching
  • Ollama Deep Analysis — optional local LLM classification (human vs AI, with reasoning)

How It Works

Layer Role
C++ Service Text analysis, ngram repetition, Shannon entropy, claim scoring
Rust FFI 6 accelerated functions: normalize, hash, trigrams, entropy, claim score, entities
SQLite provenance.sqlite — claims, sources, claim_occurrences tables
JSWindowActor Auto-analysis on every page load
About Page about:provenance — “Trust OS” with 8 tabs

Trust Badge

The URL bar shows a small shield icon reflecting the trust assessment:

  • Green (score >= 70%) — trustworthy content
  • Yellow (score 40-70%) — some trust concerns
  • Red (score < 40%) — likely unreliable or AI-generated

Hover for tooltip. Click to open full Trust OS analysis.

Access

  • Navigate to about:provenance in Doda
  • Click the Trust Badge in the URL bar
  • Command Palette (Ctrl+Shift+,)

Privacy

All analysis is performed locally. No content is ever sent to external servers. When Ollama deep analysis is enabled, it connects to your local Ollama instance only.