AIHOT: The Open-Source Newsroom That Learns What Matters

A self-hosted framework for turning noisy industry feeds into scored, clustered, agent-readable daily briefings.

10 min read • View on GitHub • More from KKKKhazix

A lone editorial desk is buried in clippings, RSS symbols, and scattered notes on one side, then resolves into a clean newsroom tray on the other with clustered topic cards and a small agent terminal. The scene explains how AIHOT turns raw noise into a curated daily briefing for both people and machines.
AIHOT does not just collect links. It turns them into an editorial workflow with judgment built in.
Key Takeaways

Most RSS readers stop at collection. AIHOT is interested in what happens next: which items deserve attention, which stories belong together, and how the final result should be delivered to a person or an agent. That makes the project feel closer to a private editorial desk than a feed reader.

The repository is explicit about that ambition. In the README, Zhang Renjie, better known as KKKKhazix, describes it as "一个自己找热点、自己写日报的网站框架。把信源和精选标准换成你的,它就是你的行业热点站。" The phrase matters because it names the real product: a framework for encoding taste.

The private newsroom problem

Every niche industry has the same problem. There is too much signal, too much repetition, and too much noise masquerading as importance. A good analyst does not just collect links. They decide what matters, what repeats, and what can wait until tomorrow.

AIHOT automates that editorial judgment. The codebase is built around a flow from ingestion to selection to clustering to digest generation, with the worker process doing the heavy lifting in the background. The result is not a timeline. It is a briefing.

AI 圈一天发太多东西,等我反应过来已经过气了——干脆让Agent 帮我每天扫一遍。

Digital Life Kazik, Author · AIHOT Skill Description

What AIHOT actually produces

The outputs are easy to miss if you only think in RSS terms. AIHOT generates daily reports, topic clusters, translated summaries, and agent-readable pages that can be consumed by Claude Desktop or similar tools. The whole system is built to serve the same material in multiple forms without redoing the editorial work.

LayerTraditional RSS readerAIHOT
CollectionPulls items into a feedPulls items, ingests pushed content, and protects the pipeline with rate limiting
SelectionManual starring or simple rulesMulti-stage scoring and filtering
Story groupingMostly source-basedClusters related coverage into events
DeliveryHuman UI firstHuman digest and agent-readable Markdown first

That last point is the tell. AIHOT is not only a presentation layer. It is also a publishing layer, exposing content in forms that language models can actually use. In practice, that means a newsroom that can brief a person in the morning and an agent during a coding session.

The scoring stack: from noise to judgment

The system’s real magic is not in fetching content. It is in the sequence that turns a pile of items into something legible. The pipeline starts with raw sources, then filters duplicates and low-quality inputs, then scores relevance, then clusters related items into events, and finally calculates heat so the most important topics rise to the top.

AIHOT does not rank stories once. It repeatedly refines them until a noisy stream becomes a few readable events.

A close-up pipeline shows many small story tokens entering a filter gate, collapsing into one clustered event, then splitting into two outputs for a human digest and an agent terminal. The image explains how AIHOT deduplicates coverage and routes the same editorial judgment to different readers.
One noisy item is not the unit of value. The event is.

The architecture matters because it mimics an editorial desk. One pass is not enough. Stories are reconsidered, merged, reweighted, and only then published. That is how the project avoids the usual failure mode of feed tools, which is to make volume feel like value.

Why agents are first-class readers

AIHOT stands out because it does not treat agents as an afterthought. The backend exposes Markdown-oriented agent routes and supports the Model Context Protocol, which means tools like Claude Desktop can browse the same briefings a person sees, but in a format optimized for machine consumption.

一个自己找热点、自己写日报的网站框架。把信源和精选标准换成你的,它就是你的行业热点站。

Zhang Renjie (KKKKhazix), Author/Developer · AIHOT GitHub Repository

That design choice changes the product category. A conventional reader helps you stay informed. AIHOT helps you build a reusable intelligence layer that can serve dashboards, daily reports, and agent workflows from the same curated source.

# Daily briefing

## Hot topics
- Clustered event one
- Clustered event two

## Agent summary
- Key entities
- Why it matters
- Suggested follow-up

The worker is the real product

If the frontend is the newspaper, the worker is the newsroom floor. The repository’s background jobs handle translation, heat ranking, digest composition, and recovery for stale work. It uses pg-boss with singleton scheduling so expensive jobs do not overlap and waste resources.

That is a practical choice, not just an architectural flourish. The expensive parts are the ones you do not want to rerun blindly. Recovery logic protects against interrupted runs, and scheduled jobs keep the daily briefing on rhythm without manual babysitting.

The other important detail is that AIHOT is opinionated about its inputs. It is built to be pointed at a specific industry or domain, then reshaped by source lists and prompts. That makes it useful for a solo operator, but also for teams that want a vertical intelligence layer without hiring a newsroom.

Why this feels different from Feedly, Huginn, or RSSHub

The comparison is simple once you stop asking whether AIHOT is a better reader. It is not trying to win that game. It is trying to define a different one.

ProjectWhat it is best atWhat it is not
Feedly or InoreaderHuman-friendly RSS aggregationA self-hosted editorial pipeline
HuginnGeneral-purpose automationA tuned daily report engine
RSSHubSource generation and feed normalizationJudgment, clustering, and digest composition
AIHOTEditorial scoring, event clustering, and agent-ready outputA generic catch-all automation tool

The real distinction is taste. AIHOT encodes a point of view about what deserves attention, then makes that point of view repeatable. That is why it feels closer to a newsroom than a utility.

Who this is for

AIHOT is for domain experts who know what matters but do not want to hand-curate every morning. It is for internal research teams, solo operators, and builders of vertical products who need a private way to turn raw sources into a usable brief.

It is also for people who want their systems to speak both human and machine. If your workflow already includes agents, markdown, and scheduled digests, AIHOT gives you a scaffold for all three.