Daily AI + VC Digest Agent

This repo contains a small "news agent" that fetches public RSS feeds daily, ranks items for AI/ML progress and startup/VC investing news, generates a <= 10 minute on-site digest, and renders it in a clean, organized UI.

What it does

• Fetches configured RSS/Atom feeds (polite timeouts + lightweight parsing)
• Classifies items into ai vs vc using keyword signals
• De-duplicates and ranks by relevance + recency + source weight
• Produces one daily digest per category in data/digests/<categorySlug>/ (and .../latest.json)
• Renders category digests in the Next.js app (/ and /category/<slug>)

Setup

npm install
npm run digest   # generate today's digests (one per category)
npm run smoke    # validate latest digest JSON parses correctly
npm run dev      # view UI

Postgres (recommended for Vercel)

If DATABASE_URL is set, digests are written to Postgres instead of data/digests/.

# 1) set DATABASE_URL in .env.local
npm run db:push
npm run db:seed
npm run digest

Sources

Edit lib/agent/sources.ts to add/remove feeds. Prefer RSS/Atom endpoints from public sites rather than scraping article HTML; it's more stable and less likely to violate site policies.

Daily automation (server)

Option A (simplest): run the script via cron on your server:

0 7 * * * cd /path/to/repo && npm run digest

Option B: call the built-in API route from your scheduler:

• POST /api/cron/digest with Authorization: Bearer $CRON_SECRET
• Set CRON_SECRET in your environment
  • Copy .env.example to .env.local for local dev
  • Vercel Cron Jobs call your endpoints with GET (supported by these routes)

Send emails (requires Postgres + Resend):

• POST /api/cron/send-digests with Authorization: Bearer $CRON_SECRET
• Set RESEND_API_KEY, EMAIL_FROM, and SITE_URL

Vercel Cron (vercel.json)

This repo includes vercel.json with a daily cron pointing at GET /api/cron, which runs:

• digest generation (all categories)
• email send (only if Resend env vars are configured)

Note: cron schedules are interpreted in UTC.

Note: writing to the filesystem works great on a traditional server. On serverless platforms with ephemeral disks, store digests in a DB/object store instead of data/.

Optional: better ranking/summaries via OpenAI

If OPENAI_API_KEY is set, the agent will ask a model to refine the final top items and produce tighter bullet summaries (still capped to the 10-minute daily read budget).