AIHOT Opens Its Full AI News Curation Stack to Developers
An open-source TypeScript framework that scrapes six source types, double-scores every item with an LLM, clusters duplicate stories, and publishes a daily briefing you can rebrand for any industry.
- AIHOT open-sources the full framework behind aihot.news, including every prompt and threshold
- Pulls from six source types: RSS, HTML lists, JSON APIs, X accounts, WeChat, and custom scripts
- Items are pre-filtered then independently double-scored by an LLM before entering the curated set
- Clustering groups multi-source reports into single events, with heat ranked by independent source count
- Ships daily, weekly, and monthly briefings plus RSS, public API, MCP, and
llms.txtfor agents - MIT licensed on GitHub, runs via Docker Compose with any OpenAI-compatible model
AIHOT publishes its full AI news-curation stack
The developer behind aihot.news has released the complete code for the Chinese AI news aggregator, including the website, backend, curation pipeline, prompts, scoring thresholds, clustering logic, and ranking system. The MIT-licensed GitHub repository provides an end-to-end template for building similar briefings for fields such as law, finance, HR, and healthcare.
Developers can adapt the system by replacing its sources, rewriting the selection rubric, adjusting thresholds, and changing the output language. The inclusion of prompts and an evaluation harness makes the release particularly useful for teams studying how language models can filter and organize high-volume news feeds.
The whole stack is included
| Component | Technology or default |
|---|---|
| License | MIT |
| Runtime | TypeScript on Node.js 24 |
| Database | PostgreSQL |
| Deployment | Docker Compose |
| Default output | Chinese headlines, summaries, and briefings |
| Included systems | Frontend, backend, ingestion, curation, clustering, ranking, search, API, and administration |
Six gates from crawl to briefing
AIHOT moves each candidate through six stages, filtering weak or duplicate material before spending more model calls on writing and analysis:
- Ingest and filter: Fetch an item, remove duplicates, and apply an initial relevance check.
- Score: Run two independent evaluations against the selection rubric, then apply a threshold based on the source tier.
- Rewrite: Generate a Chinese headline and summary for selected items.
- Cluster: Group reports that cover the same event or a direct follow-up.
- Calculate heat: Rank events using source diversity and time decay.
- Publish: Assemble selected events into the scheduled daily briefing.
This story is for Pro members
You've reached the end of the free preview. Upgrade to AlphaSignal Pro to read the full article - and everything else behind the paywall.
Comments
