Summary

Added a client-side RAG (Retrieval-Augmented Generation) pipeline that gives country intelligence briefs historical context from previously seen headlines. Here's the flow:

• When RSS feeds are parsed, headlines are embedded in the background via the ML worker and stored as vectors in IndexedDB
• When a country brief is requested, current headlines get embedded and searched against the local vector store for similar past events
• The top-3 matching historical events are appended to the context snapshot sent to the Groq-powered intel brief endpoint

The new vector-store.ts module handles all the IndexedDB operations — cosine similarity search, TTL-based pruning, and storage quota error handling.

On the client side, country-intel.ts now enriches the existing buildBriefContextSnapshot output with RAG-retrieved historical events before passing it to fetchCountryIntelBrief. The RAG context flows through the existing context query param, so no proto or server changes were needed.

Also caught and fixed two issues during review:

• The e2e test was deleting the wrong IndexedDB name in beforeEach (wm-vector-store instead of worldmonitor_vector_store)
• rss.ts was importing hashString from server/_shared/hash, which was the only cross-boundary import in the entire src/ tree — moved it to src/utils/hash.ts

Type of change

• New feature

Affected areas

• News panels / RSS feeds
• AI Insights / World Brief

Checklist

• Tested on worldmonitor.app variant
• Tested on tech.worldmonitor.app variant (if applicable)
• New RSS feed domains added to api/rss-proxy.js allowlist (if adding feeds)
• No API keys or secrets committed
• TypeScript compiles without errors (npm run typecheck)

Screenshots

No UI changes — this is a data pipeline addition.