The Wikipedia philosophy game — where every article eventually traces back to 'Philosophy' through a chain of first-linked concepts — works because knowledge is a graph, not a list. Most reading tools treat it as a list.
I built rabbitholes because I wanted to follow those threads without leaving the page I was on. Highlight any text and a shadow-DOM tooltip renders an explanation from Claude Haiku 4.5 next to your cursor. The tooltip doesn't inject into the host page's DOM, so it can't break layouts or pollute styles. Click any word in the explanation to explore it. Drag across a phrase to do the same.
The part I use most: every answer ends with two suggested rabbit-hole topics — the most interesting threads from that point. One click continues the chain. A counter tracks how many hops deep you've gone, and when you finally surface from six links of particle physics into the history of the Standard Model, you get a shareable trail of what you actually read.
For cases where the inline model answer isn't enough, a globe icon re-runs the query enriched with Brave Search results, with source chips you can click. The pencil icon opens a free-form follow-up that inherits the current context as background — so you can ask 'how does this connect to what I read three hops ago' without re-explaining.
No analytics, no telemetry, no intermediary server. Requests go directly from your browser to api.anthropic.com and api.search.brave.com. API key stored in chrome.storage.sync, never transmitted elsewhere.
github.com/robertnowell/rabbitholes
Top comments (0)