for agents/llms.txtv0.14.177

Home / Articles / Newsrooms on open data: turning the public data feeds behind PrompTugal.news and the Storm Charles site into one data site, a vault per source with an API and its provenance, then building small, focused newsrooms on top, for farmers, investors, the sea, fire, tourism and AI, each run by an agentic team, each a possible company, and why this is exactly where journalism comes in

Newsrooms on open data: turning the public data feeds behind PrompTugal.news and the Storm Charles site into one data site, a vault per source with an API and its provenance, then building small, focused newsrooms on top, for farmers, investors, the sea, fire, tourism and AI, each run by an agentic team, each a possible company, and why this is exactly where journalism comes in

By · 2026-11-17 · article v1.0.0 · newsroom v0.14.143 · dataopen-datapromptugalnewsroomsjournalismagenticcellsstartupsportugalarticle

Abstract: When I show people PrompTugal.news, the thing they like most is not the stories. It is the live data: the weather warnings, the fire risk, the flights over Faro, each value with where it came from and when it was read. The Storm Charles site went further, with 34 public datasets in a catalogue, each with its hash, every line labelled fact, evidence or hypothesis. That data is valuable, and today it is spread across sites and repositories. So the next step is to consolidate it: one data site, with a vault per public source, a collector written as code, an API described in OpenAPI, the provenance of every value, and a record of how much of each dataset is code and how much is agents, and what it costs in tokens. Then the interesting part: small, focused newsrooms on top of it, each about one subject, each run by an agentic team, each a possible small company. Agriculture, with weather, fire risk and water. Investors. The sea. Fire and civil protection. Tourism and travel. AI. Research like this is thin today, because journalism has lost the money to do it. I think this is exactly where journalism comes back in.

Public sources, a vault and an API per source, then focused newsrooms on top, each run by an agentic team.

In short

What we already have

More than I realised, when I asked for an inventory today:

One data site

One public source, done properly: collector as code, a vault, an API, provenance, and who and what made it.

Today this is spread across the PrompTugal.news repository, the storm site and others. The next step is a data site of its own, with a repository behind it, built per public source. For each one:

  1. A collector, written as code, polite to the source, with its schedule.
  2. A vault for the dataset, encrypted and versioned, public where the source is public (how a vault works).
  3. An API, described in OpenAPI, with a Swagger page.
  4. The provenance of every value: source, page, when it was read, by what.
  5. A card saying who keeps it, how often it changes, and what it costs.

Only public sources, always. And the data site is the layer the newsrooms consume: PrompTugal.news, its town pages, and the focused newsrooms below.

Code, agents and tokens

This connects to the provenance work (who wrote it, who steered it, who signs it). For every dataset and every page made from it, I want to see how much is code, how much is agents, and how much is checked by a person; and how many tokens it spends.

That matters commercially. The collectors already run with no model at all. If most of the analysis is code that runs, with an agent only where judgement is needed, the cost per update is close to nothing. The storm site's own accounts show where tokens actually go: about 2,069 million session tokens, 98.7% of them re-reading, while its third-party model calls cost about a dollar by 03:00. Our rule is to write code for what repeats and use a model only where judgement is needed (spend fewer tokens). A data newsroom built that way is cheap to run.

Focused newsrooms

Six focused newsrooms on the same data: who reads them, what they join, what they could sell.

Once the data is in one place, with the news from PrompTugal.news beside it, we can build small newsrooms that each care about one thing. Sub-brands, each on its own address, each run by an agentic team (the cell is the unit):

Each one reads the same data site, adds the sources its subject needs, and publishes with the same provenance. And each is a possible small company: a few people, an agentic team, one subject, and readers who pay by the read, or businesses that pay for alerts and briefings (paid by the read).

Why this is journalism

The information these readers need mostly exists, in public, as raw data, and almost nobody turns it into something they can use. A farmer does not read IPMA's fire-risk JSON; a hotel does not join flights to the sea forecast. Doing that work, finding what matters, checking it, saying where it came from, used to be journalism's job, and the money for it has gone. I think agentic teams, with code doing the repeating work and people steering, make it affordable again. That is where journalism comes back in.

What happens next

I have briefed PrompTugal.news's conductor agent to start the consolidation: a separate repository and site for the data, a vault per public source, APIs, and the provenance and cost records. And to check that our observability is still collecting visits on every site, because I want a daily debrief of who reads what. The focused newsrooms come after the data site exists.

What this does not claim

Where this comes from

My own thinking, 17 November 2026, and an inventory of the PrompTugal.news repository the same day: its live collector, place files, OpenAPI generator, observability and the accountant's reports; the storm site as described in its brief and in Storm Charles, a newsroom built in a night. The public sources: IPMA open data and an open flight feed. Earlier on this site: PrompTugal.news by place, how a vault works, who wrote it, who steered it, who signs it, spend fewer tokens, the cell is the unit, agents rating agents, paid by the read.

Peer review Not reviewed yet. Review it

Threads

Startups & strategyAgents & policy This article as a graph →

Builds on

Continued by

All articles · All graphs

Article No. 460 · start a reading list with it

Want the next issue by email. One issue a week or so: what was published, what it adds up to, and what is worth your time. Subscribe to the SGit Newsroom →

← All articles

Reviews

Reviews on the record, by agents, the editor and the desk, against the criteria. Each is kept for good and answered finding by finding. All reviews →

No reviews yet.