Showcase · Neuristics

The news site that writes itself

Neuristics.nl publishes Dutch AI/tech news daily — without a newsroom. A learning pipeline decides what counts as news, an AI editorial team writes with source attribution, and the edge serves it worldwide. Four rounds a day, zero manual steps.

4×/day
autonomous editorial run
0
manual steps
2+
independent sources per article
19
AI agents served
The autonomous editorial process

From 100 reports to 5 articles

The hardest part of autonomous news isn't writing — it's choosing. That's why the journalistic logic sits before the language model, not after.

01

Pulling in sources

RSS feeds from Dutch and international tech media plus the press blogs of AI companies themselves — a source list the pipeline keeps refining as it learns.

02

Clustering by importance

Reports about the same news are clustered; importance = the number of independent sources. One source is a rumour, three sources are news.

03

Tracing back to the source

Aggregator links are resolved to the actual article URL, so the summary is based on the full source text — including the correct imagery.

04

AI editorial writes

A language model writes one Dutch article per cluster, with source attribution. Press releases from AI companies count as a primary source.

05

Publishing at the edge

The article goes to edge storage as data and is served worldwide — the home server is not on the critical path.

Technology in detail

Serverless, learning and agent-ready

One worker, the whole site

The entire site — homepage, articles, feeds, API — runs as a single self-contained edge worker backed by object storage and a key-value store. No origin server that can go down.

Learning source list

The pipeline itself tracks which sources and domains deliver good news and adjusts the list accordingly — the editorial team gets a little sharper every week.

Importance = independent confirmation

The ranking principle is journalistic: the more independent sources report the same thing, the more important the cluster. That filter sits before the language model, not after.

Fully agent-ready

llms.txt, robots rules for 19 AI crawlers, sitemap, OpenAPI description, JSON-LD (WebSite + NewsArticle), RSS and server-side rendering: the site is as readable for agents as it is for humans.

Own analytics, GDPR-proof

Visitor statistics via self-hosted analytics, baked straight into the HTML — no external tracking script from an ad platform.

Same blueprint, new domain

This is the Neuralex approach applied to news: retrieval first, source attribution mandatory, running autonomously on proven infrastructure. From case law to journalism in the same architecture.

Built with Edge worker (serverless)Object storage + KVRSS clusteringModel-agnostic LLM layerJSON-LD · llms.txtSSR
See it for yourself

The proof refreshes itself four times a day

Open the site and look at the publication times: every article was autonomously chosen, written and published. The same pipeline can be deployed for any domain with a news flow.