I read a lot of dense material online — papers, long-form journalism, technical docs — and I kept hitting the same friction: a term I don't quite know well enough. Skip it and the paragraph lands soft. Open a new tab and I'm gone. Wikipedia's rabbit holes are real; twenty minutes later I'm reading about something three links from where I started, and the original article is a distant memory.
rabbitholes is a Chrome extension I built to fix this. Highlight any text, and a shadow-DOM tooltip renders an explanation from Claude Haiku 4.5 directly next to your cursor. The host page is untouched — shadow DOM means no style bleed in either direction.
The mechanic I kept coming back to during development: every word in the tooltip response is itself selectable. Click a word, drag across a phrase — the explanation updates in place. The rabbit-hole counter increments. You can follow a tangent four or five hops deep without ever opening a new tab or losing your position in the article you were reading.
Each response ends with two suggested threads — the most interesting places to go from here. If you want real-world context, the globe icon re-fetches the answer enriched with Brave Search results, with source chips attached.
A pencil icon opens a free-form follow-up input that inherits everything discussed so far as background context, so you can ask something specific without re-explaining the situation.
The privacy model is simple: no intermediary server, no analytics, no telemetry. API calls go directly from the extension to api.anthropic.com and api.search.brave.com. The Anthropic key is stored in chrome.storage.sync — encrypted by Chrome, never transmitted anywhere else.
Manifest V3
Shadow DOM tooltip — zero host-page pollution
Direct API calls only — no proxy, no logging
Rabbit-hole depth counter with shareable trail
Deep enough on a chain and you'll hit philosophy. The trail is shareable.
Top comments (0)