Mid-read, I hit a term I half-knew — "shadow DOM" in someone's architecture post. Old habit: cmd+T, Wikipedia, read, come back, scroll to find where I was. By then the argument I was following had gone cold.
rabbitholes fixes the specific thing that was breaking my reading: you highlight the term, a tooltip appears next to your cursor (shadow DOM, so nothing in the host page breaks), read the explanation, and keep going. If a word in the explanation itself needs unpacking, click it. You can chain that as many levels deep as you want — the extension tracks your hop count and keeps the full trail.
The detail I ended up caring most about: every explanation surfaces two suggested rabbit-hole topics at the bottom. Not a list of links — just two threads the answer naturally opens up, one click away. That's where the name comes from.
When I want to go further than the model knows, the globe icon re-runs the query against Brave Search and returns source chips alongside the answer. The pencil opens a free-form follow-up that inherits everything in the current context as background, so you're not starting from scratch.
No telemetry, no intermediary server. Your API key lives in chrome.storage.sync and requests go directly to api.anthropic.com. Manifest V3.
// The extension never touches a background server.
// Calls go: browser → api.anthropic.com (or api.search.brave.com)
// Your key stays in chrome.storage.sync — encrypted, local.
Top comments (0)