Domain explorer

Type a subject and press Explore. A dozen free web services are asked in parallel, straight from your browser, and the answers are put together into views of the subject. Every fact links to the page it came from, so you can check it and read further.

How to use it

Type a subject — a technology (CRDTs), a material (solid-state batteries), a craft (sourdough fermentation), an idea — and press Explore (or Enter). Examples… has a few to start with.
Wikipedia’s search decides which article the subject is. It shows the article it chose and, under the search box, the other candidates: if it picked the wrong one, click the right one. When the subject is a disambiguation page, it asks you to choose.
The coloured chips show each source: green answered, yellow still asking, grey found nothing or is switched off, orange rate-limited, red failed. Hover a chip for the detail; failures are also listed in words underneath. The views fill in as answers arrive, and whatever answered is shown even when other sources fail.
Read the views (the tabs):
- Overview — the Wikipedia introduction, key facts from Wikidata (what it is a kind of, what it is part of, dates, and so on), quick numbers (papers, repositories, discussions, books), this site’s own tools and articles on the subject, and the article’s outline and categories.
- Concept map — the subject in the middle, surrounded by the concepts it is related to: the links in the Wikipedia article’s introduction and its "See also", Wikidata’s relations in both directions (what it is a subclass of, what is a subclass of it, who invented or makes it) and OpenAlex’s research topics. Bigger nodes have more evidence. Click a node to explore it: the map re-centres on that concept, and the Path above the views records how you got there, so you can step back to any earlier subject (answers are cached, so going back is instant). The list under the map says where each node came from.
- Timeline — dated events from Wikidata (inception, discovery, first publication, start dates, and the same for related things), sentences with a year in the Wikipedia introduction, the most-cited papers, when research output began and peaked, when the leading repositories were created, the big Hacker News discussions and the books. Filter by kind.
- People & orgs — people and organisations Wikidata relates to the subject, the most prolific authors and institutions in OpenAlex, software and products, and the organisations behind the top GitHub repositories.
- Literature — works per year (a chart), OpenAlex’s research topics, the most-cited papers, notable recent papers (the last three years, by citations), the latest arXiv preprints, and a reading list of books from Open Library with their covers.
- Practice — the most-starred GitHub repositories, the top Hacker News stories, and the top-voted Stack Exchange questions with their commonest tags.
- Glossary — the terms the article leans on (the links in its introduction and its "See also"), each defined by the first sentence of its own Wikipedia article, plus Wikidata’s descriptions of related things that have no article. Filter as you type.
- Sources — everything gathered, grouped by source with its link, and every request made (with the status, the time it took, and whether this browser’s cache answered).
- Briefing — optional: a language model writes a structured, cited briefing from the gathered material. See below.
Copy link copies a link to this exploration (?q=…); Markdown report downloads everything as a Markdown document with a link on every line; JSON downloads the raw gathered data. Refresh asks every source again, ignoring the cache.

The sources, and their limits

All of them are free and asked without a key, from your browser. Answers are cached in this browser for a day (Sources, keys and cache → Clear cache empties it), so re-opening an exploration or stepping back along the path costs nothing.

Source What it gives Limits

Source	What it gives	Limits
Wikipedia (MediaWiki Action API and REST summary)	The search that picks the article; its introduction, short description, picture, categories, sections and links; the "See also" links; one-line definitions of the linked terms.	No key. Wikimedia asks for moderate use and answers bursts with HTTP 429; the page retries once. English Wikipedia only.
Wikidata (wbgetentities and the search API)	The subject’s entity and its statements (the key facts), the entities its statements point at, and the entities that point back at it (found with `haswbstatement` searches for subclass of, instance of, part of, facet of, field of work, use, industry, inventor, developer, named after, …) — with their kinds (person, organisation, software) and dates.	No key. Statements are only as complete as Wikidata is: small subjects may have one or two. The Query Service (SPARQL) is not used: it is slow for this kind of question and limits each address to a minute of query time a minute.
OpenAlex	Works whose title or abstract contain the subject’s name (the Wikipedia title, and the words you typed unless they are a short acronym like "CRDT", which matches too much else): the count, the most-cited, recent notable ones, works per year, authors, institutions and research topics.	Without a key, everyone on your network shares one free daily budget (about a hundred explorations a day: each one makes seven requests). Add your own free key from openalex.org/settings/api under Sources, keys and cache for a budget of your own. When OpenAlex is out of budget or fails, Crossref answers for the papers.
arXiv	The latest preprints on the subject.	arXiv’s own API sends no CORS headers, so a web page cannot read it, and the gitwebsite app’s relay has no arXiv upstream. The preprints come from OpenAlex instead (works with an arXiv location), and link to their arXiv pages.
Crossref	Papers by relevance, sorted here by citation count — only when OpenAlex can’t answer (its chip says standby otherwise).	No key; a polite shared pool. Its relevance search is looser than OpenAlex’s.
GitHub search API	The most-starred repositories matching the words you typed, with their language, topics and dates, and the organisations behind them.	Ten searches a minute per address without a token; the chip turns orange and says when the limit resets. A token (a fine-grained one with no permissions is enough) raises it to thirty.
Hacker News (Algolia search)	The top stories, by relevance and popularity, with points and comment counts.	No key; generous limits (thousands of requests an hour per address).
Stack Exchange	The top-voted questions on the sites that fit the subject — Stack Overflow for computing, and Seasoned Advice, Electrical Engineering, Chemistry, Physics, Mathematics, Biology, Cross Validated and others when the subject’s description matches them — and their commonest tags.	300 requests a day per address without a key. You can name the sites yourself under Sources, keys and cache (e.g. `cooking, chemistry`).
Open Library	Books for the reading list, with covers, authors, first publication and editions; fiction goes last, and results that share no word with the subject are left out.	No key; please be gentle (the page asks once per exploration).
This site	Tools and articles here about the subject, from the site’s own search index (`/search.json`).	None: it comes from this site.

Wikipedia (MediaWiki Action API and REST summary)

The search that picks the article; its introduction, short description, picture, categories, sections and links; the "See also" links; one-line definitions of the linked terms.

No key. Wikimedia asks for moderate use and answers bursts with HTTP 429; the page retries once. English Wikipedia only.

Wikidata (wbgetentities and the search API)

The subject’s entity and its statements (the key facts), the entities its statements point at, and the entities that point back at it (found with haswbstatement searches for subclass of, instance of, part of, facet of, field of work, use, industry, inventor, developer, named after, …) — with their kinds (person, organisation, software) and dates.

No key. Statements are only as complete as Wikidata is: small subjects may have one or two. The Query Service (SPARQL) is not used: it is slow for this kind of question and limits each address to a minute of query time a minute.

OpenAlex

Works whose title or abstract contain the subject’s name (the Wikipedia title, and the words you typed unless they are a short acronym like "CRDT", which matches too much else): the count, the most-cited, recent notable ones, works per year, authors, institutions and research topics.

Without a key, everyone on your network shares one free daily budget (about a hundred explorations a day: each one makes seven requests). Add your own free key from openalex.org/settings/api under Sources, keys and cache for a budget of your own. When OpenAlex is out of budget or fails, Crossref answers for the papers.

arXiv

The latest preprints on the subject.

arXiv’s own API sends no CORS headers, so a web page cannot read it, and the gitwebsite app’s relay has no arXiv upstream. The preprints come from OpenAlex instead (works with an arXiv location), and link to their arXiv pages.

Crossref

Papers by relevance, sorted here by citation count — only when OpenAlex can’t answer (its chip says standby otherwise).

No key; a polite shared pool. Its relevance search is looser than OpenAlex’s.

GitHub search API

The most-starred repositories matching the words you typed, with their language, topics and dates, and the organisations behind them.

Ten searches a minute per address without a token; the chip turns orange and says when the limit resets. A token (a fine-grained one with no permissions is enough) raises it to thirty.

Hacker News (Algolia search)

The top stories, by relevance and popularity, with points and comment counts.

No key; generous limits (thousands of requests an hour per address).

Stack Exchange

The top-voted questions on the sites that fit the subject — Stack Overflow for computing, and Seasoned Advice, Electrical Engineering, Chemistry, Physics, Mathematics, Biology, Cross Validated and others when the subject’s description matches them — and their commonest tags.

300 requests a day per address without a key. You can name the sites yourself under Sources, keys and cache (e.g. cooking, chemistry).

Open Library

Books for the reading list, with covers, authors, first publication and editions; fiction goes last, and results that share no word with the subject are left out.

No key; please be gentle (the page asks once per exploration).

This site

Tools and articles here about the subject, from the site’s own search index (/search.json).

None: it comes from this site.

Every source can be switched off under Sources, keys and cache.

Briefing (optional)

The Briefing tab sends the gathered material — up to seventy numbered sources: the article’s introduction, key facts, definitions, papers, preprints, the trend, people, books, repositories, discussions and questions, each with its link — to a language model, which writes a briefing with an overview, key concepts, history, who’s who, open problems, how to get started and a reading path. The model is told to use only those sources and to cite them as [S3]; each citation in the briefing links to its source. It can still be wrong: check the claims that matter. Nothing is sent until you press Write the briefing, and the rest of the page never needs it.

Anthropic — your own API key, sent only to api.anthropic.com, straight from your browser, with the anthropic-dangerous-direct-browser-access header that allows a web page to call the API. The model defaults to claude-sonnet-5-5; type another model ID if you want one. On the current Sonnet and Opus models the request asks for Anthropic’s server-side fallback, so a declined request is retried on another model. Requests are billed to your account; use a key with a spending limit.
OpenAI-compatible — any server with /chat/completions: Ollama (http://<host>:11434/v1), LM Studio, vLLM, llama.cpp’s server, or a hosted service. The page calls it directly, so the server must allow this site’s origin (for Ollama, start it with OLLAMA_ORIGINS=https://geekychris.github.io). Or tick Send through the gitwebsite app: the app’s companion relays the request to a host on your own network (its LAN relay; the first request to a new host asks you to allow it in the app), with no CORS set-up on the server. See the companion.

The briefing is part of the Markdown report and the JSON once written.

Privacy: what is sent where

What you type is sent to each source that is switched on, as its search query: Wikipedia, Wikidata, OpenAlex, Crossref (only when OpenAlex fails), GitHub, Hacker News’s search (Algolia), Stack Exchange and Open Library. The names of related concepts are sent to Wikipedia and Wikidata to look them up. Pictures load from Wikimedia and Open Library’s cover service.
Nothing goes to this site’s owner: the site is static, and its search index is a file it serves.
Your keys and tokens (OpenAlex, GitHub, Anthropic, an OpenAI-compatible server) are kept in this browser’s local storage and sent only to their own service, never in a link. Forget keys on the Briefing tab removes the model keys; clear the fields under Sources, keys and cache to remove the others.
The briefing sends the gathered material (public data from the sources above) and the subject to the model you choose — Anthropic, or your own server.
Answers are cached in this browser’s local storage for a day. Copied links hold only the subject (?q=) and the chosen article (&w=).

Limits

English only: English Wikipedia, English labels on Wikidata, and searches in English.
Literature is matched by the subject’s name in titles and abstracts, so a subject with a common name ("Raft") is best explored through its article ("Raft (algorithm)") — the explorer uses the Wikipedia title for this. Practice sources (GitHub, Hacker News, Stack Exchange, Open Library) are searched for the words you typed.
The concept map shows at most 34 nodes, the strongest first; the list under it has them all.
Counts are what each service reports today; citation counts differ between OpenAlex, Crossref and Google Scholar.

Linking to an example

Links can say what to explore, and the tab after #: overview (the default), map, timeline, people, literature, practice, glossary, sources or briefing — for example /tools/domain-explorer/?q=CRDTs&w=Conflict-free%20replicated%20data%20type#map.

Parameter Meaning Example

Parameter	Meaning	Example
`q`	the subject to explore, as typed	`q=solid-state%20batteries`
`w`	the English Wikipedia article to use, skipping the search	`w=Sourdough`
`example`	one of the presets below	`example=crdt`

q

the subject to explore, as typed

q=solid-state%20batteries

w

the English Wikipedia article to use, skipping the search

w=Sourdough

example

one of the presets below

example=crdt

Presets:

crdt — CRDTs, conflict-free replicated data types
solid-state-battery — solid-state batteries
sourdough — sourdough fermentation
raft — Raft, the consensus algorithm
perovskite — perovskite solar cells