@AGENTS.md
- Pure frontend: All LLM calls happen directly in the browser via OpenAI SDK (
dangerouslyAllowBrowser: true) - No API routes: No server-side logic. The Next.js app only provides client-side routing and static hosting.
- Configuration: API keys stored in browser localStorage via Settings modal (
src/lib/config.ts) - Caching: All page data cached in localStorage (
src/lib/client-store.ts)
src/lib/config.ts— Frontend config management (API keys, model settings)src/lib/openai.ts— Browser-side OpenAI streaming + completionsrc/lib/tools.ts— Browser-side search/images/news/data fetching (Serper API + LLM fallback)src/lib/prompt.ts— System prompt + user prompt buildersrc/lib/client-store.ts— localStorage page cache + tree structuresrc/components/SettingsModal.tsx— API key configuration UI