A link aggregator with threaded comments, built by Claude Code on Elements

Upvotehall is a full-stack app built with Claude Code on Elements.

Links and text posts ranked by votes and age, newest and top pages, threaded comments, karma and reply emails, all live.

Run this app on your machine. Use this demo as inspiration, or as the starting point for your own app. Your agent creates a new app with this one as the scaffold, and you prompt it to add whatever features you want.

How it's built

Upvotehall needed votes that move scores on every open page, a front page ranked by votes and age, live threaded comments, and an email when someone replies. Each of those is a part of Elements, so the agent spent its 16 minutes on the community itself.

• Live posts and scores. Posts are a LiveTable opened by rank for the front page, by time for newest and by author for profiles, 25 at a time. Database triggers keep each post's score and comment count, and a notify trigger carries every change to the open lists.

• Votes through a view. Votes are LiveTables opened for the signed-in reader, so a vote is an insert or delete through the view, with the voter set from the session.

• Live threads. Comments are a LiveTable opened per post. A new comment appears in every open copy of the thread and schedules its reply email in the same transaction.

• Ranking on a schedule. A SQL function scores each post by votes and age, and one cron line runs a job every five minutes that re-ranks the last two weeks, so the front page keeps moving.

• Reply emails and karma. A background job emails the author of the parent comment or post. A profile's karma counts the votes other people gave that user's posts and comments.

• Data from SQL files. Two migrations define the schema and its triggers, then seed eight users, thirty posts over the last four days, votes, and comment threads several levels deep.

Why Elements

Elements is an integrated full-stack app framework and a project server. The project server checks the agent's work as it builds: it type-checks the server and browser code, analyzes the program, applies migrations and runs the tests, then reports the results to the agent. A security mistake, such as calling a server-only function from the browser, is a compile error. The agent reads the report and fixes its own mistakes before it tells you a feature is done. The same checks run on every feature you ask for next.