Beamcast

Embeds a persistent AI chat sidebar within the browser that automatically captures and injects the current webpage's DOM content, text, and metadata into the LLM context window without requiring manual copy-paste. Uses a content script to extract page state and pass it to a sidebar iframe that maintains conversation history across navigation, enabling the assistant to reference page content in real-time without losing context.

Analyzes the current webpage's structure and content to provide context-aware suggestions, explanations, or edits that reference specific page elements. The assistant understands the semantic meaning of the page (forms, tables, navigation, content blocks) and can generate responses that directly relate to what the user is viewing, such as form-filling suggestions, table analysis, or content editing recommendations.

Implements a freemium model that abstracts underlying LLM API costs by routing free-tier users through a shared or rate-limited API gateway, while premium users either get higher rate limits, faster response times, or access to more capable models. The backend likely uses token counting and request throttling to manage costs, with a paywall that gates access to premium model variants or removes rate limits for paid subscribers.

Maintains chat conversation history and context across browser restarts, tab closures, and navigation events by storing messages in browser local storage or IndexedDB, with optional cloud sync to a backend database. Allows users to resume previous conversations and reference earlier messages without losing context, though storage is typically limited by browser quota (50MB-1GB depending on browser).

Uses a content script manifest to inject the sidebar and page-context extraction logic into any website the user visits, with a dynamic allowlist/blocklist to prevent injection on sensitive sites (banking, password managers, etc.). The extension detects page load events and injects the necessary JavaScript to enable sidebar functionality, handling both static and dynamically-loaded content through MutationObserver or similar DOM monitoring.

Abstracts the underlying LLM provider (OpenAI, Anthropic, or other APIs) behind a unified interface, allowing users to select which model to use (e.g., GPT-4, Claude 3, etc.) without changing the UI or workflow. The backend likely implements a provider adapter pattern that translates requests to the appropriate API format, handles authentication, and manages rate limits per provider.

Implements a sidebar UI as an iframe or shadow DOM component that loads asynchronously and does not block page rendering or interaction. Uses lazy loading and code splitting to minimize initial extension size and startup time, with the sidebar only initializing when explicitly opened by the user. The sidebar communicates with the background service worker via message passing to avoid blocking the main thread.

Manages user accounts, authentication (likely OAuth or email/password), and tier tracking (free vs. premium) to enforce rate limits and feature gates. Stores user preferences, API key associations (if applicable), and usage metrics in a backend database, with session management via browser cookies or local tokens. Syncs tier status and rate limit quotas to the browser extension for client-side enforcement.