Hey HN, just starting out with this project and would love feedback. I kept switching between my browser and Claude constantly—summarizing articles, asking questions about docs, extracting data from pages. The tab-switching and copy-paste friction added up. So I built a browser where Claude is just… there. How it works: There’s a slide-out panel (Studio) with three modes: ∙ Chat – Talk to Claude with automatic page context. No need to paste anything, it sees what you’re viewing. ∙ Actions – One-click buttons for common tasks: summarize, extract key points, pull all links, translate, explain code, review code. ∙ Terminal – Command-line style interface for when you want more control. Browser features (it still needs to browse well): Vertical tabs, tab stacks for grouping, split view, workspaces with color coding, session save/restore, reading mode, picture-in-picture, dark/light themes. Took cues from Arc and Vivaldi but kept things minimal. Technical stuff: Electron with main/renderer process separation. Direct Anthropic API integration with streaming—no middleware or CLI wrappers. You bring your own API key and choose your model (Opus 4, Sonnet 4, or Haiku 3.5). Config stored locally at ~/.claude-browser/config.json. What I’m unsure about: Right now it’s API-key only. Keeps things simple and private, but wondering if that limits who’d actually use it. Open to thoughts. Still rough around the edges but usable as a daily driver. Happy to answer questions.