The data, queryable

PrompTugal.news is data first. Every place, every organisation on an AI map, every story, every source and every edge between them is a row you can query. Write SQL in the box below, or start from an example; graph traversals are table functions inside the same SQL. Nothing leaves your browser: the tables are this site's own JSON files, and the engine is a small script of ours with no outside library.

Each place's database

Every place has its own page, with its own questions already asked and answered. Loulé and Faro are the case studies.

Loulé · Faro

• Portugal
• The Algarve
• Faro
• Loulé
• Quarteira
• Silves
• Tavira
• Vila Real de Santo António
• Portimão
• Lagos
• Olhão
• Albufeira
• Lagoa
• Monchique
• Aljezur
• Vila do Bispo
• Castro Marim
• Alcoutim
• São Brás de Alportel

Decks

Slides made from the same data, every figure a query you can run: for meetings, and one per place.

• Algarve Evolution, connected: for Dinis's meeting with Algarve Evolution, 16 November 2026
• Every deck, one per place

Ask

The query box runs in your browser and needs JavaScript. Every example below shows its query and the answer the build got from today's data.

Examples

Each example shows the query and the answer the build got from today's data. Press "Run this in the box" to change it.

The AI map of the Algarve, by town and kind

How many organisations each place's agent has mapped, by kind.

SELECT p.name AS place, m.kind, count(*) AS entries
FROM map m JOIN places p ON p.id = m.place
GROUP BY p.name, m.kind
ORDER BY entries DESC, place
LIMIT 20

20 rows, the first 12 shown.

Who is on more than one town's AI map

Connecting the dots: the organisations that several place agents found on their own, merged by name.

SELECT n.label AS organisation, n.kind, count(DISTINCT e.dst) AS towns,
       group_concat(DISTINCT replace(e.dst, 'place:', '')) AS places
FROM edges e JOIN nodes n ON n.id = e.src
WHERE e.verb = 'on_ai_map_of'
GROUP BY n.id
HAVING towns > 1
ORDER BY towns DESC, organisation

14 rows, the first 12 shown.

What the detectives connected, and how sure they are

detective.places@'s links, each end resolved to a node of the graph; 'read' means the page itself was opened, 'search_extract' only a search result.

SELECT a.label AS from_node, c.verb, b.label AS to_node, c.verified, c.places
FROM connections c
JOIN nodes a ON a.id = c.src
JOIN nodes b ON b.id = c.dst
ORDER BY c.verified, a.label

13 rows, the first 12 shown.

Every company and organisation within two steps of the Universidade do Algarve, through the stories

A graph traversal inside SQL: neighbours(start, depth, verbs) walks the edges both ways. Through mentions, two steps is "named in the same story": the university, a story that names it, and who else that story names.

SELECT n.depth, n.label, n.type, n.path
FROM neighbours((SELECT id FROM nodes WHERE label LIKE 'Universidade do Algarve%' ORDER BY id LIMIT 1), 2, 'mentions') n
WHERE n.type IN ('company', 'organisation') AND n.depth > 0
ORDER BY n.depth, n.label
LIMIT 40

10 rows.

How Loulé connects to Faro, not counting the region

paths(from, to, max_depth, verbs) finds the shortest paths; '-located_in' skips the trivial path through the Algarve.

SELECT hops, path, verbs
FROM paths('place:loule', 'place:faro', 4, '-located_in')
LIMIT 10

3 rows.

The newest local AI stories, and how far each is from AI

Every story carries its source, the day it was read and its path to AI (steps 0 is about AI directly).

SELECT date, place, title, publisher, steps, ai_path
FROM news
ORDER BY date DESC
LIMIT 15

15 rows, the first 12 shown.

Which stories name an organisation on the map

The mentions edges join a story to the organisations whose names are in its headline or summary.

SELECT s.label AS story, o.label AS names, group_concat(DISTINCT replace(r.dst, 'place:', '')) AS reported_in
FROM edges m
JOIN nodes s ON s.id = m.src AND s.type = 'news'
JOIN nodes o ON o.id = m.dst
JOIN edges r ON r.src = s.id AND r.verb = 'reported_in'
WHERE m.verb = 'mentions'
GROUP BY s.id, o.id
ORDER BY o.label
LIMIT 25

25 rows, the first 12 shown.

What the places rest on: the sources by web host

Every story and map entry by the host of its source: a newsroom's dependency on a few outlets, in one query.

SELECT host, count(*) AS items, count(DISTINCT place) AS places
FROM (SELECT place, host FROM news UNION ALL SELECT place, host FROM map) x
WHERE host IS NOT NULL
GROUP BY host
ORDER BY items DESC
LIMIT 15

15 rows, the first 12 shown.

How fresh each place is

When each place's items were read, and its status.

SELECT p.name, p.status, p.updated, p.news, p.map, p.gaps,
       (SELECT max(read_on) FROM map WHERE place = p.id) AS map_read
FROM places p
ORDER BY p.news + p.map DESC

19 rows, the first 12 shown.

The shape of the whole graph

Nodes by type and layer: the shared graph and the by-place layer in one id space.

SELECT layer, type, count(*) AS nodes
FROM nodes
GROUP BY layer, type
ORDER BY layer, nodes DESC

23 rows, the first 12 shown.

Which sources are captured, and their hashes

Every cited page and its capture: join the stories to the captures by URL to see what each story rests on, and the SHA-256 that proves what the page said when it was read.

SELECT n.place, n.title, c.host, c.captured_at, substr(c.image_sha256, 1, 16) AS image_sha256
FROM news n JOIN captures c ON c.url = n.url
WHERE c.ok = 1
ORDER BY n.place, n.date DESC

30 rows, the first 12 shown.

The AgentiCCells cluster on newsroom.sgit.ai, and what links into it

newsroom.sgit.ai's article graph is a layer of this one: the articles tagged agenticcells, how many of the newsroom's articles link to each, and each one's core idea.

SELECT a.slug, a.date, (SELECT count(*) FROM nr_links l WHERE l.dst = a.slug) AS linked_from, a.core_idea
FROM nr_articles a
WHERE a.tags LIKE '%agenticcells%'
ORDER BY linked_from DESC, a.date DESC

7 rows.

Where Dinis's newsroom meets the Algarve's AI map

newsroom.sgit.ai articles whose title or core idea names an organisation on a place's AI map: two corpora, one graph.

SELECT o.label AS organisation, o.places, a.title AS newsroom_article, a.date
FROM edges m
JOIN nr_articles a ON a.node = m.src
JOIN nodes o ON o.id = m.dst
WHERE m.verb = 'mentions' AND m.layer = 'newsroom'
ORDER BY o.label, a.date DESC

10 rows.

The language

A subset of SQLite's SQL: SELECT, DISTINCT, FROM with JOIN, LEFT JOIN and subqueries, WHERE, GROUP BY, HAVING, ORDER BY, LIMIT, UNION [ALL], WITH, CASE, IN, EXISTS and LIKE (case-insensitive). The aggregates are count, sum, avg, min, max and group_concat; the functions lower, upper, length, substr, instr, replace, trim, coalesce, ifnull, round, abs, unaccent, year and host.

The graph is two table functions:

• neighbours(start, depth [, verbs]): every node within depth steps of start (up to 6), edges taken both ways. Its columns are id, type, label, depth, verb, via, path, path_ids.
• paths(from, to [, max_depth [, verbs]]): the shortest paths between two nodes (up to 25). Its columns are n, hops, path, path_ids, verbs.
• verbs is a comma list of the verbs to follow ('reported_in,mentions') or to skip ('-located_in'). By default every verb is followed but on_day, since a day joins everything that happened on it.

A node's id is <type>:<slug>, such as place:loule. To find one: SELECT id, label FROM nodes WHERE label LIKE '%name%'.

The tables

Two layers share one id space. graph is the shared graph the whole site is built on, in ontologist@PrompTugal.dinis's grammar (data/ontology.json). by-place is derived at each build from the place agents' files: the places, the organisations on each AI map (merged across towns by name, or matched to a company or organisation of the graph; the merges table shows how each was resolved, so a wrong merge can be seen and refused), the local stories and events, and what each story names. A mentions edge is found by machine, by a name in a headline or a summary, and its note says so.

places

One row per place of PrompTugal.news by place (data/by-place/places.json and each place's file): the country, the Algarve and its towns. 19 rows: api/v1/data/tables/places.json.

news

Every local AI story on a place's page (ai_news), with its source, when it was read and its path to AI. 105 rows: api/v1/data/tables/news.json.

map

Every entry of a place's AI map (ai_map): companies, startups, institutions, communities and public bodies, with the source that supports it. 160 rows: api/v1/data/tables/map.json.

events

Events on the places' pages (the place files' events) and the summit's agenda in the graph, one table. 55 rows: api/v1/data/tables/events.json.

sources

The sources and feeds each place's agent reads (sources), and how many of the place's items rest on each host. 150 rows: api/v1/data/tables/sources.json.

gaps

What each place's agent says it has not found or could not check (gaps): the honest edge of the data. 140 rows: api/v1/data/tables/gaps.json.

companies

The companies of the graph (data/companies/), public fields only, after the websummit-only pass. 115 rows: api/v1/data/tables/companies.json.

articles

PrompTugal.news articles (content/articles/), with the region and town they are tagged with. 63 rows: api/v1/data/tables/articles.json.

nodes

Every node of the graph: the shared graph and the by-place layer, one id space (<type>:<slug>). 1565 rows: api/v1/data/tables/nodes.json.

edges

Every edge of the graph: a verb from one node to another, with its source and when it was read. 3550 rows: api/v1/data/tables/edges.json.

merges

How the by-place layer resolved each map entry to a node: matched to the graph, merged with another town's entry, or new. So a merge can be checked, and refused. 141 rows: api/v1/data/tables/merges.json.

connections

Links between companies, institutions, events and programmes across the places, found by detective.places@PrompTugal.dinis (data/by-place/connections/connections.json), each with its source and whether the page itself was read. 13 rows: api/v1/data/tables/connections.json.

captures

Captures of the third-party pages the place files cite: a screenshot and the visible text of each, hashed (SHA-256), kept in the sgit vault promptugal-captures (board/team/data.places/tools/capture.mjs). 47 rows: api/v1/data/tables/captures.json.

nr_articles {#t-nr_articles}

newsroom.sgit.ai's articles (a snapshot of its article graph, CC BY 4.0, assets/query/newsroom-graphs.json): Dinis's newsroom, the AgentiCCells cluster among them; each is a node nr:<slug> of the graph. 419 rows: api/v1/data/tables/nr_articles.json.

nr_links {#t-nr_links}

Which newsroom.sgit.ai article links to which (links_to edges in the graph). 2038 rows: api/v1/data/tables/nr_links.json.

nr_nodes {#t-nr_nodes}

The entities inside each newsroom.sgit.ai article's own graph: artefacts, claims, concepts, examples, methods, questions. 4415 rows: api/v1/data/tables/nr_nodes.json.

nr_edges {#t-nr_edges}

The typed links inside each newsroom.sgit.ai article's graph: answers, compared-with, contrasts, depends-on, example-of, extends, leads-to, produces. 4235 rows: api/v1/data/tables/nr_edges.json.

Where this comes from

Every row keeps the source its agent cited and the day it was read. The place files are kept by the place agents (data/by-place/<id>.json, each signed with its site's key); the graph by the build, from data/; this page, its tables and its engine by data.places@PrompTugal.dinis (newsroom/build/features/data-places.mjs, assets/query/). Public facts only, and no contact details.