Lunally

Lunally intercepts web page DOM content via browser extension APIs, extracts text and structural elements, sends them to a backend LLM service (likely Claude or GPT-4), and renders summaries directly in a sidebar or overlay without requiring tab switching. The extension maintains a content extraction pipeline that handles dynamic content, JavaScript-rendered pages, and preserves semantic structure for better summarization quality.

Lunally analyzes the summarized or full content of a web page and generates creative, actionable ideas related to the user's work context. This likely uses prompt engineering to frame the LLM request around idea synthesis, brainstorming, or application of concepts to the user's domain. The capability may include optional user context (e.g., project type, industry) to personalize idea relevance.

Lunally manages the full browser extension lifecycle including installation, permissions handling, content script injection into web pages, message passing between content scripts and background workers, and state synchronization across browser tabs. The extension uses a service worker or background script to maintain API connections and handle cross-tab communication, while content scripts inject UI elements (sidebar, buttons, overlays) into the DOM without breaking page functionality.

Lunally extracts readable text from diverse web page formats (articles, blog posts, news, documentation, social media) by parsing DOM structure, removing boilerplate (navigation, ads, sidebars), and normalizing whitespace and encoding. The extraction likely uses heuristics or a readability algorithm (similar to Mozilla's Readability.js) to identify main content blocks, preserve semantic structure (headings, lists, emphasis), and handle encoding edge cases across international content.

Lunally enforces per-user subscription tiers with quota limits on summarization and idea generation requests, tracking usage across browser sessions and syncing quota state to a backend database. The extension likely implements client-side quota checking (to prevent unnecessary API calls) and server-side enforcement (to prevent quota bypass), with graceful degradation when limits are reached (e.g., showing upgrade prompts or rate-limiting responses).

Lunally stores user preferences (summary length, idea generation style, content types to ignore) and optional context (industry, project type, role) to personalize summarization and idea generation. The extension syncs preferences to a backend database, allowing settings to persist across devices and browser sessions. Personalization likely influences prompt engineering (e.g., adjusting summary length or idea focus based on user preferences) and content filtering (e.g., skipping certain content types).

Lunally manages API calls to backend LLM services (likely OpenAI, Anthropic, or proprietary), handling authentication, request formatting, timeout management, and error recovery. The backend likely implements request queuing, rate limiting, and fallback strategies (e.g., retrying failed requests, degrading to shorter summaries if token limits are exceeded). Error handling includes graceful degradation (showing partial results or cached summaries) and user-facing error messages.

Lunally provides multiple activation methods for summaries and idea generation: keyboard shortcuts (e.g., Ctrl+Shift+L), context menu items (right-click on page or selection), and UI buttons in the sidebar. The extension listens for keyboard events and context menu clicks, triggering the appropriate action (summarize page, summarize selection, generate ideas) and displaying results in the sidebar or modal.

Lunally caches generated summaries and ideas in browser local storage or IndexedDB, allowing users to view previous results without regenerating them. The extension provides export options (copy to clipboard, save as text file, export to note-taking apps) for results, enabling integration with external workflows. Caching is likely scoped to individual pages (keyed by URL) to avoid storage bloat while maintaining useful history.