AIHOT’s public GitHub repository packages the engine behind an AI-news site as a reusable template: developers can replace its sources, selection criteria and branding to run a sector-specific news site of their own. The practical consequence is that the project offers more than feed aggregation; it supplies an editorial pipeline whose rules are intended to be changed by the operator. The AIHOT repository describes the project as a public template and framework for this purpose.
The repository says the system ingests material from six source types: RSS, web lists, JSON interfaces, X accounts, WeChat public accounts and externally pushed content. It then deduplicates and pre-filters items, scores potentially important material twice independently, generates Chinese titles and summaries, groups coverage into events, and produces daily reports. The AIHOT repository also states that sources are assigned three tiers with different admission thresholds.
Editorial rules are configuration, not a black box
AIHOT exposes its prompts and selection thresholds, with the project directing operators to industry/prompts/ for changes to editorial standards. Its customization guidance identifies industry/sources.json for initial sources, industry/taxonomy.ts and industry/topics.json for categories and topics, and industry/selection.ts for admission thresholds. The AIHOT repository recommends calibrating selection rules against one or two hundred operator-labelled items using its evaluation script.
For clustering, the project says it first finds candidates from the previous two weeks using title-and-summary vectors, or text overlap when no vector service is configured, and then asks a model whether the material is the same event, a follow-up, or a separate event. Hotness is calculated at the event level: within 48 hours, each independent source counts once, with a 24-hour half-life. The AIHOT repository says that repeated retrievals and multiple articles from one outlet therefore do not add repeated weight.
The stack listed by the project is Node.js 24, TypeScript, React Router server-side rendering, Fastify, PostgreSQL, pg-boss, Tailwind CSS and Docker Compose. The AIHOT repository requires Docker, Node.js 24 for its configuration initializer, and an OpenAI-compatible model API key for the documented quick-start path.
The unresolved issue is editorial accountability
Analysis: AIHOT’s useful distinction is that it treats an industry-news site as a configurable editorial system rather than as a generic chatbot wrapper. Exposed prompts, thresholds and evaluation hooks give an operator places to express domain judgement. But that also leaves the central quality problem with the operator: selecting authoritative sources, defining what is important, and checking whether generated titles, summaries and event groupings preserve the underlying reporting. The repository itself emphasizes that its included 18 public overseas AI sources are examples and that real sources should be replaced for another industry. The AIHOT repository
The licensing boundary is also concrete. The code uses the MIT license, while the repository says the AIHOT name and logo are not included; a deployment intended as an independent publication needs its own identity as well as its own source policy. The AIHOT repository
Top comments (0)