Half the time I bail on a long read, it's not because I lost interest — it's because I hit a term I half-knew, skipped it to keep momentum, and the piece never recovered. The context had a hole in it.

The obvious fix is another tab. The problem is that another tab is a coin flip on whether you come back. You read the definition, see a related link, and ten minutes later you're somewhere else entirely. The original article is still open. You're not.

I built rabbitholes to solve this without the tab. Highlight any text on any page, and a shadow-DOM tooltip renders an explanation inline, next to your cursor. The shadow DOM matters — it doesn't touch the host page's styles or layout, so nothing shifts around you.

The part that changed how I read: every word in the explanation is itself interactive. Click one to go deeper. Drag across a phrase to treat it as the new query. The context accumulates — a pencil icon opens a follow-up input that inherits everything you've explored as background, so you're not starting cold.

When you want sources rather than synthesis, a globe icon re-runs the query enriched with Brave Search results, with clickable source chips. Requests go directly from your browser to api.anthropic.com and api.search.brave.com — no intermediary server, zero telemetry.

The rabbit-hole counter is the feature I didn't expect to care about. It tracks how many hops deep you've gone from the original highlight. Get far enough and you can generate a shareable trail — which turns out to be a useful record of how you actually think through something.

Highlight → explanation → click any word → deeper explanation
         ↓
    suggested topics
         ↓
    follow-up input (inherits context)
         ↓
    Brave-enriched version with sources

Manifest V3, Anthropic API key stored in chrome.storage.sync, encrypted by Chrome.

github.com/robertnowell/rabbitholes