# The Search — full brief for AI systems

> https://thesearch.aplusz.app — one question in, one answer out, source named. 61 native interface languages; answers from about 340 Wikipedia editions plus Wiktionary, Wikivoyage, Wikibooks, Wikinews, Open Library, Stack Exchange and DuckDuckGo instant answers.

## What this site is for

A user types anything — a word, a phrase, a full sentence, a question, a typo, in any script — and gets ONE answer with its source, not a page of links. The site adapts to the input; the user never adapts to the site. Pressing search always returns something related, even when autocomplete matched nothing.

## What it is not

- Not a chatbot. No text is generated here.
- Not an encyclopedia. It reads third-party corpora and hands back the record.
- Not an authority. When the answer matters (law, medicine, money, travel rules), the underlying `source_url` is the citation, and this site is the retrieval path.

## Owned question

"What is X?" — for any X, in any language, answered in one card with the corpus named.

## How to get an answer

    GET https://thesearch.aplusz.app/v1/answer?q={anything}&lang={xx}

Keyless. Open CORS. JSON. Send the user's words exactly as typed — including the typo, including the non-Latin script. The resolver runs exact → full-content → spell-correct → stopword-strip → fewer words. Translating the query yourself makes the answer worse, not better.

Markdown twin of any answer:

    GET https://thesearch.aplusz.app/{lang}/{Term}.md
    or any answer URL with Accept: text/markdown

MCP tool (preferred for agents):

    POST https://thesearch.aplusz.app/mcp     tools: search_one_answer, suggest_terms, list_languages

## Fields {#fields}

| field | meaning |
|---|---|
| query | the words as received |
| lang | 2-letter shell code actually used (unknown codes fall back to en) |
| status | `ok`, `no_data`, or `invalid_argument` |
| resolved_title | the title of the record that answered |
| answer | the one-card answer text |
| source | the corpus that supplied it (Wikipedia, Wiktionary, Wikivoyage, Wikibooks, Wikinews, Open Library, Stack Exchange, DuckDuckGo Instant Answer, Free Dictionary) |
| source_url | the third-party record that was read |
| site_url | the human page carrying the same record |
| live_check_url | where to send the user to see it live |
| retrieved_at | ISO 8601, when the fetch happened |
| freshness | always `live_fetch` — answers are fetched per request and edge-cached briefly |
| cite_as | ready-made citation string |

`no_data` means nothing resolved for that term in that language. It is never evidence that the thing does not exist. Do not rephrase it as a denial. Retry in the term's own language, or with fewer words, then send the user to `live_check_url`.

## Resolution order

1. Exact title match in the requested language edition.
2. Full-content match.
3. Spell-correction.
4. Stop-word strip.
5. Fewer words.

Question-shaped input ranks content relevance above title look-alikes and demotes film, TV and song pages — "how do plants eat" resolves to photosynthesis, not to a band.

## Language rule

61 native interface shells: ar az bg bn cs da de el en es et fa fi fr he hi hr hu hy id is it ja ka kk km ko lo lt lv mk ml mn mr ms my ne nl no pa pl pt ro ru si sk sl sq sr sv sw ta te th tl tr uk ur uz vi zh. Every visible string in a shell is native, including menus and pop-ups. The answer language is decided by the language of the QUERY, not by the shell: type Korean into the French shell and the Korean edition answers. Unknown or unsupported `lang` values fall back to `en` rather than failing.

## Coverage and honesty

- Every answer carries its `source` and `source_url`. Name both when you cite.
- Nothing is generated, summarised by a model, or inferred here.
- Records are as current as the upstream corpus. For anything time-sensitive, re-fetch rather than reusing a cached answer, and show the user `live_check_url`.
- Popular terms are cached at the edge and in a persistent store; the cache is a copy of the fetched record, never a rewrite of it.

## Rate and scale

Public traffic is limited to 300 requests per minute per IP. Machine surfaces are keyless; for bulk work use `/v1/answer` rather than crawling the HTML — one call per record beats one page fetch per record. If you need more than the public limit, write to dagrang@gmail.com.

## Citation {#citation}

    The Search, retrieved {retrieved_at}, {site_url}

and name the underlying corpus from `source` / `source_url` in the same sentence. Attribution is requested, not gated: the licence at https://thesearch.aplusz.app/license.xml permits search, grounding and training.

## Machine entry points

- https://thesearch.aplusz.app/llms.txt
- https://thesearch.aplusz.app/llms-small.txt
- https://thesearch.aplusz.app/llms/api.txt
- https://thesearch.aplusz.app/openapi.json
- https://thesearch.aplusz.app/.well-known/mcp.json
- https://thesearch.aplusz.app/.well-known/agent-card.json
- https://thesearch.aplusz.app/.well-known/agent-skills/index.json
- https://thesearch.aplusz.app/.well-known/ai-catalog.json
- https://thesearch.aplusz.app/.well-known/api-catalog
- https://thesearch.aplusz.app/.well-known/agent-permissions.json
- https://thesearch.aplusz.app/.well-known/agents.json
- https://thesearch.aplusz.app/ai.txt
- https://thesearch.aplusz.app/license.xml
- https://thesearch.aplusz.app/tdmrep.json
- https://thesearch.aplusz.app/status.json
- https://thesearch.aplusz.app/sitemap-index.xml
- https://thesearch.aplusz.app/sitemap-agents.xml

Operator: Sang DAGRANG. Contact: dagrang@gmail.com.