AIHOT is a self-hosted news-site framework in which a language model finds, scores and ranks its own daily AI headlines, with no human editor in the loop. It is for developers who want an automated news feed running on their own domain, and for readers who would rather have an AI-industry digest delivered into a coding agent than sit in a newsletter inbox. The project comes from KKKKhazix, a solo developer in China also known online as easychen. One backend feeds four front ends: a free website at aihot.news, an RSS feed, a public API, and an installable Agent Skill.
What it does
The repository describes itself as a website framework that finds its own trending topics and writes its own daily report. In practice the site has no writers: software picks the stories, rates them, groups them and publishes on a schedule.
The live site is free and needs no login. Every story carries a heat score, there is a leaderboard of models, and the digest can be read by day, by week or by month.
The part that matters most to developers is the Skill, named aihot: a small installable capability for AI coding agents (Claude Code, Codex, Cursor and others that follow the open Agent Skills standard) that pulls the same digest into a chat with one sentence and no API key. The video notes that the sentence is written in Chinese. The Skill does not ship alone; it is one of six skills in the author's MIT-licensed khazix-skills collection.
How it works
This is the project's own account of its pipeline, not an outside audit:
- It collects from six kinds of sources, including RSS, scrapers, JSON APIs, X/Twitter accounts and WeChat public accounts.
- A language model scores and filters every item twice, each pass independent of the other.
- Stories about the same event from different outlets are clustered into one ranked item, which is what appears on the page.
The README as saved does not name which model does the scoring, or what the final ranking formula is.
Getting started
The code lives at github.com/KKKKhazix/AIHOT and is MIT licensed; the AIHOT name and logo are explicitly not covered by that licence. The README as saved records 7.8k stars and 1.9k forks, which are GitHub's own counters, not a measure of quality.
The material names no operating system, chip, memory or disk budget, so treat hardware as an open question until you read the repository's setup notes. What the project plainly needs is a language model, since the scoring and the writing are the model's job.
Reading the hosted site, the RSS feed, the public API and the Skill costs nothing and asks for no key, because the author pays for that instance. Running your own copy is different: you supply the model, which means a paid cloud API key or a local model on hardware you own.
The Skill is not a plain clone either: it installs as part of khazix-skills through the Agent Skills standard your agent already supports, and the materials here do not print the install command, so copy it from that repository rather than guessing.
When to use it / when not
The catch the video leads with is licensing: the code is open, the brand is not, so anything you deploy carries your own name and logo. The second catch matters more for an American reader: this is a Chinese-language service about the AI industry, and the Skill is driven by a Chinese sentence, so what comes back is not an English newsletter.
Use it instead of a hand-edited AI newsletter when you want the selection rules to be yours, inspectable and on your own schedule. Don't use it if you only want something to read in English, or if you need the feed as an MCP server, because an Agent Skill is a different mechanism.
Alternatives
For the site, the honest comparison is the Hacker News front page or a TLDR-style AI newsletter with the editor swapped for a model. For the agent side, the nearest thing a reader already knows is an MCP server or a ChatGPT plugin. The cheapest alternative is just asking Claude what happened in AI today, which takes no setup but gives you no heat score, no source list and no clustering of duplicate coverage.
Take AIHOT seriously if you build content pipelines, run a publication, or want a readable example of a newsroom with the editor removed: the source list, the two scoring passes and the duplicate clustering are the interesting part, and they are code rather than a pitch. Developers who live inside a coding agent get a second reason to look, since the aihot Skill is a compact demonstration of handing an agent a live data feed with no key in the way. Readers who just want English AI news should bookmark the site and leave the code alone: the language barrier, not the technology, is what will stop them.