I read an article about Byzantine fault tolerance last week. Three paragraphs in, I hit 'quorum-based consensus' — a phrase I could gesture at but not actually explain. I opened a new tab, read half of a Wikipedia article, came back, and had lost the thread of the original piece entirely.
rabbitholes fixes this at the browser level. Highlight any text on any page, and a shadow-DOM tooltip renders an explanation from Claude Haiku 4.5 next to your cursor — it doesn't inject styles into the host page or reflow anything. The explanation ends with two suggested rabbit-hole topics, the most interesting threads to pull from where you landed. Click any word in the explanation to go deeper on that term. Drag across words to pick a phrase.
The thing I use most: the globe icon re-answers your query enriched with Brave Search results, with source chips you can click. So for something time-sensitive or contested, you're not just getting a language model's cached understanding — you're getting it grounded in current sources.
There's a rabbit-hole counter that tracks how many hops deep you've gone. It's partly a curiosity feature, partly a practical anchor — you can retrace your path without reconstructing it from memory.
Privacy setup: zero analytics, zero telemetry, no intermediary server. Your Anthropic and Brave API keys live in chrome.storage.sync, encrypted by Chrome, and requests go directly from your browser to api.anthropic.com and api.search.brave.com. Nothing passes through anything I run.
Manifest V3. Works on any page Chrome renders.
The repo, including the full extension source: github.com/robertnowell/rabbitholes
Top comments (0)