Glassbox

About

Every AI chatbot hedges. Glassbox shows you exactly where and why.
Glassbox is a browser extension that analyzes AI-generated responses in real time. It detects hedging language, extracts factual claims, scores how direct the response actually is, and highlights the parts you should look twice at.
WHAT IT DOES
When an AI responds to your question, Glassbox adds a small analysis banner showing:
Trust level: Is this response well-supported, advisory, or vague?
Directness score: Did the AI actually answer your question, or give you a framework to think about it yourself?
Response type: Direct answer, framework, conditional, or redirect
Flagged language: Hedging, filler, deflection, vague sourcing, and unsupported claims — each with an explanation of why it matters
Factual claims: Every checkable claim is extracted with a confidence rating and a one-click verification link
Inline highlights: The flagged text is highlighted directly in the response so you can see it in context. Hover for details.
Rephrase suggestions: When the response is evasive, Glassbox suggests how to rephrase your question to get a more direct answer
TWO MODES
Free (no API key needed): Local analysis detects hedge phrases, weasel words, structural patterns, confidence language, and readability metrics. Works instantly, no data leaves your browser.
Full analysis (API key required): Deep analysis powered by OpenAI or Anthropic that evaluates trust, extracts claims, scores directness, and generates rephrase suggestions. Bring your own API key — Glassbox never stores or transmits your conversations.
Works on most Modern LLM Chatbots
PRIVACY
Glassbox runs entirely in your browser. Your conversations are never stored, collected, or transmitted to any server. When using the full analysis mode, text is sent directly to OpenAI or Anthropic's API using your own key — Glassbox has no backend and no analytics.
OPEN SOURCE
Glassbox is open source. View the code, report issues, or contribute at:
https://github.com/owenkleinmaier/glassbox