Your AI intern that actually writes the WeChat post, not just the summary
An automation pipeline that treats AI-generated newsletters as a production workflow, not a chat completion.

What it does
TrendPublish is a TypeScript/Deno pipeline that scrapes sources (RSS, Hacker News, arXiv, Twitter/X, FireCrawl, etc.), uses LLMs to cluster topics, fill evidence gaps, write headlines and body copy, and renders WeChat-compatible HTML. It can dry-run to local files or publish drafts to WeChat Official Accounts. A dashboard lets you edit sources, article plans, and scheduling without touching config files.
The interesting bit
The author explicitly rejects the “RSS summary script” trap. The system keeps an audit trail: every run snapshots steps, errors, quality reviews, HTML, images, and config state. You can see why an article was chosen and where the AI hallucinated, which is rarer than it should be in this space. Multi-account “matrix” publishing and Cloudflare-native deployment (Workers/Workflows/D1/KV/R2) are also built in, not bolted on.
Key highlights
- Provider/adapter architecture for LLMs (OpenAI, DeepSeek, Qwen), image gen (DashScope, MiniMax), fetchers, and publishers
- Vector deduplication with SQLite (local) or D1 (Cloudflare)
- WeChat-specific HTML sanitization and multiple templates including dynamic layout
dry-runmode outputs inspectable artifacts before any real publish- Dashboard with run timeline, error explanations, and quality reports
- Deno v2.0+ runtime; local dev spins up backend + Vite frontend concurrently
Caveats
- README is thorough but the truncated sections suggest some docs may still be incomplete
- Requires Deno, which is less common than Node; the “Node.js” badge in the repo description appears aspirational or outdated
- WeChat publishing requires fixed-IP relay or Cloudflare relay setup; not zero-config
Verdict
Worth a look if you run a daily tech newsletter on WeChat and are tired of hand-gluing RSS feeds. Skip it if you just need a personal read-later digest — this is overbuilt for one reader and underbuilt for non-WeChat platforms (though the author hints at future expansion).