Autonomous X (Twitter) posting system. Runs a daily pipeline that researches trending AI topics, writes developer-focused posts, and schedules them throughout the day.

researcher → writer → scheduler → neon DB → pg_cron → X API

1. Researcher — searches X and the web for AI news, model releases, infra patterns, and developer tooling from the last 24 hours. Produces a research brief with 5–8 content angles. Uses grok-4-latest via the Responses API with webSearch and xSearch tools.

2. Writer — turns the brief into 4–6 X posts (single tweets or threads). Practical, builder-focused voice: no hype, no emojis, no em dashes. Uses grok-4-latest with structured output.

3. Scheduler — assigns optimal posting times targeting developer-active windows (8–10 AM, 12–2 PM, 5–7 PM, 9–11 PM EST), minimum 90 minutes apart. Uses grok-4-1-fast-non-reasoning with structured output.

Posts are persisted to Neon and published via pg_cron calling /cron/execute-post.

• Vercel AI SDK — generateText / generateObject for all LLM calls
• xAI Grok — model provider for all three agents
• Hono — HTTP server
• Neon + Drizzle ORM — Postgres storage
• X API (XDK) — tweet publishing via OAuth1

npm install
cp .env.example .env
# Fill in API keys

Required environment variables:

# Start server (port 3010, hot reload)
npm run dev

# Test individual agents against real APIs
npm run test:researcher
npm run test:writer
npm run test:scheduler

# Run the full pipeline end-to-end
npm run test:agents

# Trigger cron routes manually
npm run test:cron:daily
npm run test:cron:execute-post

All cron routes require x-cron-secret header (or ?secret= query param) matching CRON_SECRET.

Create a second job on cron-job.org:

• URL: POST https://<your-vercel-url>/cron/execute-post
• Headers: x-cron-secret: <CRON_SECRET>, Content-Type: application/json
• Body: empty
• Schedule: every 30 minutes

No body means scan mode: queries all pending posts with scheduled_at <= NOW() and publishes each.