Skip to content

Documentary base (guidelines & statutes)

This guide covers Tarno's documentary base: what INAPI, BCN and the SII publish about trademark registration in Chile, stored literally and versioned and read through the same contract as the trademarks (REST ≡ MCP). It carries its own scope, legal:read (see Scope).

The complete, always-current field-level schema lives in the interactive OpenAPI reference at https://api.tarno.cl/docs (Scalar). This guide explains how these reads are requested — it does not restate the schema.

Throughout the guide the base URL is https://api.tarno.cl and every call carries the X-API-Key header.

What the documentary base is

Three sources (the SII with two runs), each with its own run:

source what it stores run
INAPI the trademark guidelines (chapter by chapter, with their page), the circulars, the Madrid System and the fee schedule inapi_documents (weekly)
BCN LeyChile the text of the statutes in the legal-norms catalog —Law 19.039, its Regulation…— in all their versions, article → letter → subsection, with the amendment notes bcn_statutes (weekly)
SII the UTM, UTA, IPC, UF and observed-dollar indicators sii_indicators (daily)
SII (roster) the legal name, subtype, start of activities and current domicile (commune and region) of legal entities with a Chilean RUT; read in the party's sii block (Parties) sii_legal_entities (daily, at the end of the sync chain)

⛔ What is stored is what the source wrote: no summaries, no paraphrase and no «relevance». Nothing is deleted: a new version closes the previous one and keeps it, so you can always read what a text said on a past date.

Today you can read the guidelines, the circulars and Madrid —the list, the outline of each document and any section, literally— (below), the text of an article on a date (below), the search by words and meaning over both (below), the fee table with pesos computed on read (below), the SII indicators (below) and the version change log (below). And, on the detail of an annotation, the explained citation: the text of each cited norm as it applied that day and the guideline chapters that discuss it (below).

Scope: legal:read

Every read of the documentary base requires legal:read, and that scope is granted explicitly:

  • an unrestricted key (empty scopes) does not have it;
  • neither does the trial key;
  • brands:read and reports:read do not include it.

Without it the call answers 403 forbidden and the message names legal:read. Ask your operator for it.

⚠ For whoever grants keys: a non-empty scopes list is a closed list. Moving a key from [] to ["legal:read"] removes every other read: write the full list (for example ["brands:read", "insights:read", "watch:read", "legal:read"]). See authentication.

The data date, per source

Every response gives the date of its source in dataAsOf (and in the Iris-Data-As-Of header when the response is an object): the last successful run of inapi_documents, bcn_statutes or sii_indicators, depending on what is read. ⛔ It is never the trademark corpus date: a trademark run this morning says nothing about when Law 19.039 was read, and vice versa. dataAsOf: null means «no successful run of that source» — we do not know, which is different from «today».

Guidelines, circulars and Madrid

Everything INAPI publishes in the documentary base: the 50 chapters of the trademark Guidelines (Directrices), the circulars, the Madrid System page and manual and the fee publications. Three reads, all with the legal:read scope:

operation REST MCP
the list of documents, each with its current version GET /guidance/documents list_guidance_documents
one document with its outline (Sumario: the tree of sections, without bodies) and its version history GET /guidance/documents/{code} get_guidance_document
one literal section, with its page, its version and the norms it cites GET /guidance/documents/{code}/section get_guidance_section

How to cite a section

  • The stable citation is (code, version, path): the document code (guideline-35 = chapter 35 of the Guidelines), the version number (version.seq) and the path of the section in INAPI's own outline (III.D, III.D.▪1.§7, III.D.sidebar1, intro). For example: «Guidelines, ch. 35, version 2, III.D.▪1.§7 (p. 18)». ⛔ Never a row id: row ids are rebuilt when the extraction of the same version improves, and we do not publish them.
  • A lawyer's citation is code + paragraph: «ch. 35 § 7» is paragraph=7. If that number repeats inside the version (INAPI sometimes numbers twice), the answer is 409 ambiguous_selector with the candidate paths; ask again by path.
  • Without version you read the current version; with version an earlier one (INAPI does not keep old versions on its website; we do). The history is in get_guidance_document → versions.

What partial, image_only and ocr mean

Every version declares in extractionStatus how it could be read:

value meaning
extracted the text and the outline were read cleanly
partial the text is literal, but the tree is incomplete: an entry of the outline INAPI prints does not appear in the body, or a labelled heading of the body is not in the outline (its text stays inside the section above). It is an inconsistency of the source, declared. A heading merely worded differently in the outline and in the body no longer counts as partial: the section is there, with the body's heading
image_only INAPI published the document as images: there is no text. get_guidance_document answers 200 with outline: [], and asking for a section is 422 out_of_coverage with extractionStatus: "image_only"
ocr INAPI published the document as images and its text was machine-read (OCR). It has one section per page (ocr.p1, …) and is found by search, but it is not literal: a word may be misread and, in a diagram (chapter 21, the application flow), the labels come in page order, not in the order of the procedure, and the arrows are not text. Quote the PDF, never this text. search_guidance hits carry it in version.extractionStatus
pending · failed not read yet / could not be read

⛔ A partial or image_only chapter is never filled in with text that is not the source's; the text of an ocr one is the source's, but machine-read, and declared as such.

Parameters

operation parameter what it is
list kind optional, one or more of guideline, circular, manual, web_page, fee_schedule (kind=guideline,circular)
list q optional, words of the title, ignoring accents and case (2–100 characters)
document, section code the document code (lower case, digits and hyphens: guideline-35, madrid-page)
document, section version optional, the version number (seq); default, the current one
section path or paragraph exactly one: the outline path (up to 200 characters, URL-encoded) or the paragraph number (1–9999)

The list is ordered by kind, chapter number (guideline-2 before guideline-10) and code, with an exact total; a document INAPI stopped linking stays in it with missingSince.

Example: § 7 of chapter 35, current version.

curl -sS "https://api.tarno.cl/guidance/documents/guideline-35/section?paragraph=7" \
  -H "X-API-Key: $TARNO_API_KEY"
# or by its path, encoded: ?path=III.D.%E2%96%AA1.%C2%A77
{ "name": "get_guidance_section",
  "arguments": { "code": "guideline-35", "paragraph": 7 } }

The response carries document, version (seq, extractionStatus…), path, kind, heading, paragraphNumber, pageFrom/pageTo, the literal body, children (the direct children, to walk down the outline) and citations: each cited norm with legalNormId (readable with get_legal_norm_article), citedNorm (the law number, also when it is outside the catalog) and lawNamed: false when the text cites an «art. 19» without saying which law. dataAsOf is the last successful read of INAPI (inapi_documents).

Errors

status code when
400 validation_error an unknown parameter; both path and paragraph, or neither; kind=statute (it is BCN's); a one-character q; a code with upper case or /
403 forbidden the key does not have legal:read
404 not_found the document, the version or the section does not exist
409 ambiguous_selector the paragraph number repeats inside the version; candidates lists the paths
422 out_of_coverage the version has no text (image_only); it carries coverage: null and its extractionStatus

The text of an article on a date

GET /legal-norms/{id}/articles/{article} ≡ get_legal_norm_article — scope legal:read.

The literal article, as BCN publishes it, in the version that applied to that norm on asOf, with its letters, subsections and amendment notes.

parameter what it is
id the norm id in GET /legal-norms ≡ list_legal_norms (opaque string)
article the article key: 20, 17 bis a, 1, único — the same one the annotation citations carry. Case, a leading «Artículo»/«Art.» and º are ignored. In REST, with spaces encoded (17%20bis%20a)
letter optional, one letter (a–z, ñ): only that letter
subsection optional, a subsection (1–99): of the letter if letter is given, of the article otherwise
asOf optional, YYYY-MM-DD: the date whose text you want. Default: today in Chile
transitional optional, true for the transitional article with that key (a statute can have a permanent article 1 and a transitional article 1). Default false. Only true or false: any other value ⇒ 400 validation_error

Example: article 20 letter h of Law 19.039 as it stood on 2021-03-15 (1 is an illustrative id; take it from GET /legal-norms).

curl -sS "https://api.tarno.cl/legal-norms/1/articles/20?letter=h&asOf=2021-03-15" \
  -H "X-API-Key: $TARNO_API_KEY"
{ "name": "get_legal_norm_article",
  "arguments": { "id": "1", "article": "20", "letter": "h", "asOf": "2021-03-15" } }

Which text applies on a date

A catalog norm can have several BCN texts over time. Law 19.039 has three:

from to text
— 2006-06-19 the original Law 19.039
2006-06-20 2022-08-05 DFL 3 (consolidated text)
2022-08-06 — DFL 4 (current consolidated text)

And each text has its own validity periods in BCN. The response says it all in text: documentCode (which text), validFrom/validTo (its validity per BCN), appliesFrom/appliesTo (the window in which it is the text of this norm) and lastSeenAt (the last time BCN served us that content). A 2015 date reads DFL 3 even if BCN gives DFL 3 an open validity: the norm's window rules.

Unsplit and transitional articles

  • split/splitStatus: split = the letters were found and verified; unsplit = they could not be cut safely (for example, a list of letters that restarts mid-article) and the article is served whole; no_letters = the article has no letters.
  • If you ask for a letter of an unsplit article, the answer is 200 with the whole article, letters: [] and selection.status: "article_not_split". ⛔ Never an invented cut.
  • transitional: true reads the transitional article with that key; without it, the permanent one.

Errors

All with the RFC 9457 body (application/problem+json), identical in REST and MCP:

status code when
400 validation_error an unknown parameter, an upper-case letter, an impossible date (2026-02-31), subsection outside 1–99, an article with / or longer than 40 characters
403 forbidden the key lacks legal:read
404 not_found the norm does not exist; or the article, letter or subsection is not in the text of that date
409 ambiguous_selector the text has two articles with the same key; candidates carries their names. We do not silently pick one
422 out_of_coverage we have no text to answer with: with coverage: null, the norm is in the catalog but its text is not loaded; with coverage {from, through}, the date falls outside the windows for which we do have it (null at one end = open)

Body shape (illustrative values):

{ "type": "https://api.tarno.cl/problems/out_of_coverage", "title": "Business rule not satisfied",
  "status": 422, "code": "out_of_coverage", "requestId": "...",
  "detail": "we have no text of legal norm 1 that applies on 1990-01-01",
  "coverage": { "from": "2006-06-20", "through": null } }

200 responses carry catalog caching (Cache-Control: public, max-age=3600, stale-while-revalidate=86400 and Vary: X-API-Key); errors, no-store.

Two searches, both with the legal:read scope, identical on REST and MCP:

what you search REST MCP
the current sections of INAPI's documents (Guidelines, circulars, Madrid, fees) POST /guidance/search search_guidance
statute articles, in the text that applied on a date or in every validity POST /legal-norms/search search_statutes

They find «what does INAPI say about coexistence» or «which article talked about renewal before 2022» without knowing the section or article number. Every hit carries the reference to read it whole (get_guidance_section / get_legal_norm_article) and a literal excerpt.

The three modes

mode what it does
hybrid (default) two channels in parallel —text (Spanish, accents and case ignored) and meaning similarity (semantic search)— fused by reciprocal rank (RRF)
text words only. Web-search syntax: "exact phrase", or, -word
semantic meaning only

Each hit says in matchedBy which channel found it (text, semantic or both). No scores are published: the order is the answer.

⭐ Declared degradation. If the meaning channel cannot run (the vector service is down or slow, or the vector search timed out), hybrid returns the text hits and says so in meta.channels:

"meta": { "dataAsOf": "2026-09-27T05:00:00.000Z", "mode": "hybrid",
          "channels": { "text": "ran", "semantic": "unavailable" } }

In mode: "semantic" there is nothing to fall back on: the answer is 503 service_unavailable. ⛔ Never an empty list saying «nothing matched» when nothing was searched.

Statutes: asOf or allVersions

  • By default the search runs on the text in force today in Chile, with the same rule as get_legal_norm_article (Which text applies on a date): searching and reading give the same text.
  • asOf: "2015-03-01" searches the text that applied that day.
  • allVersions: true searches every validity of every text —to read what an old application was examined against—. ⭐ The same text of an article repeated across validities (DFL 3 has five) comes once, with all of them in validity; a changed text is another hit.
  • asOf and allVersions together are a 400.
  • norm (the catalog id) narrows to one norm. With no text loaded —or none for that date— it is 422 out_of_coverage with its coverage, as when reading the article; an unknown one, 404.

Each statute hit carries norm (null when the text is not linked to exactly one catalog norm), document (the BCN text: code, title, bcnNormaId), article, articleLabel, transitional, headingPath, excerpt and validity ([{validFrom, validTo}], oldest first).

Parameters

parameter search what it is
q both what you look for, 2–200 characters
mode both hybrid (default), text or semantic
kind guidance one or more of guideline, circular, manual, web_page, fee_schedule
document guidance only that document (guideline-35)
norm statutes only that catalog norm (its id)
asOf statutes YYYY-MM-DD; default, today in Chile
allVersions statutes true for every validity
limit, cursor both 1–50 (default 20); cursor is the previous page's next, as is

Example: which articles talked about renewal, in any validity.

curl -sS -X POST "https://api.tarno.cl/legal-norms/search" \
  -H "X-API-Key: $TARNO_API_KEY" -H "Content-Type: application/json" \
  -d '{"q": "renovacion del registro", "allVersions": true}'
{ "name": "search_statutes",
  "arguments": { "q": "renovacion del registro", "allVersions": true } }

And in the current Guidelines:

{ "name": "search_guidance",
  "arguments": { "q": "coexistencia de marcas", "kind": ["guideline"] } }

Errors

status code when
400 validation_error q with no searchable words (only stop-words: «de la») or that only excludes (-marca); q under 2 or over 200 characters; an unknown parameter; asOf together with allVersions
403 forbidden the key lacks legal:read
404 not_found norm does not exist
422 out_of_coverage norm with no text loaded (coverage: null) or none for asOf (coverage {from, through})
503 service_unavailable mode: "semantic" and the meaning channel could not run

A search carries no catalog cache. meta.dataAsOf is the date of its source: inapi_documents for the guidance, bcn_statutes for the statutes.

Fees

INAPI's trademark fee table, in UTM and literal, with the amount in pesos computed on read:

REST MCP
GET /fee-schedule get_fee_schedule

The table is the one in chapter 13 of the trademark Guidelines (2026): the five fees (on filing, on acceptance for registration, on renewal, marginal annotations and appeal) and the six months of the late-renewal surcharge. INAPI's other two fee publications (the 2022 one and the old annex) are not served as a table: they are only compared with chapter 13, and what differs is published in the surcharge note.

How amountClp is computed

  • amountUtm is literal from INAPI (7.2, never rounded).
  • amountClp is a DERIVED value, computed on read: amountUtm × utm.value, exact and not rounded (INAPI's rounding rule is not ours to apply). No peso amount is stored, so none goes stale.
  • utmMonth (YYYY-MM) chooses the month of the UTM; by default, the current month in Chile.
  • utm says which UTM was used:
utm.basis means
requested_month the UTM of utmMonth itself
latest_published that month's UTM is not published yet: the latest earlier one was used (utm.month says which)
  • With no UTM on or before utmMonth (the SII series starts in 2013), utm is null and every amountClp is null: the fee in UTM is still there; what is missing is the conversion.
  • unit: per_class is paid once per Nice class; fixed, once per filing.

When you quote a peso amount, say the month and the source: «6 UTM per class; 430,326 pesos with the UTM of September 2026».

⭐ The fees of a given procedure (application + registration, or a renewal with its published expiry) are computed by POST /fee-schedule/calculate ≡ calculate_fees, with brands:read: see official fees.

The surcharge note

The surcharge table of the Directrices 2026 (chapter 13; chapter 23 §42) is the rule applied: the 1st month at 6 UTM, the same as an on-time renewal, up to 12 UTM in the 6th. Chapter 13's «(*)» note says the renewal increases 20 % a month after expiry, and other INAPI fee publications word it differently. That discrepancy is published as an informative note and blocks no calculation (the field keeps its name, surchargeDoubt):

  • surchargeDoubt.description states the rule and how the notes differ;
  • surchargeDoubt.notes[] gives the literal notes: chapter 13's first, then those of the other INAPI publications whose note differs (open differences only), each with its documentCode and documentTitle;
  • the rows the note concerns (the renewal and months 2 to 6; month 1 was confirmed by INAPI in writing on 2026-11-04) carry hasDoubt: true.
curl -sS "https://api.tarno.cl/fee-schedule?utmMonth=2026-09" -H "X-API-Key: $TARNO_API_KEY"
{ "name": "get_fee_schedule", "arguments": { "utmMonth": "2026-09" } }
{ "utm": { "month": "2026-09", "value": 71721, "basis": "requested_month" },
  "items": [ { "itemKey": "renewal_per_class", "amountUtm": 6, "unit": "per_class",
               "amountClp": 430326, "hasDoubt": true },
             { "itemKey": "renewal_surcharge_month_2", "amountUtm": 7.2, "surchargeMonth": 2,
               "amountClp": 516391.2, "hasDoubt": true } ] }
status code when
400 validation_error utmMonth is not a real month (2026-13, 2026-9) or an unknown parameter
403 forbidden the key lacks legal:read
422 out_of_coverage no fee table is loaded (coverage: null)

dataAsOf is the date of the INAPI documents source (inapi_documents). It carries the catalog cache.

Indicators

The indicators the SII publishes, by code and date range:

REST MCP
GET /economic-indicators list_economic_indicators
code what it is one value per
UTM monthly tax unit month (dated on the 1st)
UTA annual tax unit month (dated on the 1st)
IPC consumer price index, in points (unit: "index_points") month (dated on the 1st)
UF unidad de fomento — published in advance day
USD_OBSERVED observed dollar — up to the previous banking day banking day
  • code is required; from/to (YYYY-MM-DD, inclusive) bound the range; limit 1–366 (default 31) and next paginate. total is exact.
  • meta.coverage {code, from, through} says from and through which date we hold that indicator. A range that runs past it answers with what there is; one entirely outside is a 422 out_of_coverage with the coverage — never an empty 200 that would say «there were no values».
  • ⚠ The IPC mixes bases. The SII rebased it: the 2013 values say «base anual diciembre 2013=100» and the 2026 ones «base anual 2023=100». Every row carries its literal baseNote, and points of different bases are not comparable: never subtract or chain them into one figure.
curl -sS "https://api.tarno.cl/economic-indicators?code=UF&from=2026-09-01&to=2026-09-30" \
  -H "X-API-Key: $TARNO_API_KEY"
{ "name": "list_economic_indicators",
  "arguments": { "code": "IPC", "from": "2025-01-01" } }
status code when
400 validation_error no code or an unknown one; an impossible date; from after to
403 forbidden the key lacks legal:read
422 out_of_coverage the range is entirely outside the coverage (coverage {code, from, through}), or there is no value of the code at all (coverage: null)

meta.dataAsOf is the date of the SII source (sii_indicators).

Version changes

The log of which sections changed each time INAPI published a new version of a document, or BCN a new validity of a statute text, and when we detected it:

REST MCP
GET /guidance/changes list_guidance_changes

Each change carries the document (code, kind, publisher, title), the two versions (fromVersion, toVersion: seq and versionTag), detectedAt, effectiveDate, titleChanged, the counts (counts {added, removed, modified}) and the changed sections (sections [{path, change, headingChanged}]).

  • detectedAt is when we saw it (the run that downloaded the new version), not when the publisher made it.
  • effectiveDate exists only for BCN: the date the new text entered into force, literal as BCN gives it (one is 2222-02-02, a BCN placeholder, stored as is). INAPI publishes none: null.
  • At most 100 sections per change: sectionsTruncated: true says there are more, and counts always has the full numbers. Read a section with get_guidance_section (the old one with version).
  • Filters: document, publisher (inapi or bcn) and from/to — the detection day, Chile time. Most recent first; limit 1–100 (default 20) and next.
  • ⭐ meta.dataAsOf: with publisher (or a document), the date of its source; with neither, the older of inapi_documents and bcn_statutes — the mixed log is only complete up to there.
{ "name": "list_guidance_changes",
  "arguments": { "publisher": "inapi", "from": "2026-09-01" } }
status code when
400 validation_error from after to; an impossible date; an unknown publisher; limit outside 1–100
403 forbidden the key lacks legal:read

The explained citation

An INAPI resolution says «refused under art. 20 letter h) of Law 19.039». With expand=citations, the detail of that annotation carries, for each norm it cites, the literal text of the article or letter that applied on the annotation's date and the signals from the guidelines: the INAPI documents that cite that same article/letter, with their counts.

REST MCP
GET /annotations/{id}?expand=citations get_annotation with expand: ["citations"]

⛔ It requires legal:read IN ADDITION to brands:read. Without it, 403 forbidden naming legal:read — and the request consumes no quota. Without expand, the annotation is exactly what it always was, with brands:read: the citations key is absent.

⛔ Only on the detail. The lists (GET /trademarks/{n}/annotations, POST /annotations/search, list_trademark_annotations, search_annotations) and GET /trademarks/{n}?expand= reject expand=citations with 400: a hundred annotations with their texts would be a huge response.

Each element of citations[] (at most 50, in the order the resolution cites them; citationsTruncated: true when there are more):

  • what the resolution cites: normKind (ley | reglamento), legalNormId (the GET /legal-norms catalog), citedNorm (19039), name, article, letter, subsection and span (the literal words);
  • text — the text of that day, with the same rule as get_legal_norm_article: asOf (the date used), chosenBy (what chose it: act_date, or one of the two exceptions below), documentCode and bcnNormaId (the BCN text), validFrom/validTo (that version's validity), articleLabel, split, scope (letter: body is the cited letter; article: body is the whole article) and body, literal;
  • textStatus — why there is or is not text;
  • guidance — the signals (below).
textStatus what it means text
found the article (or its letter) that applied that day yes
letter_not_split the article is not split into letters (a cut is never invented) the whole article, scope: "article"
law_not_identified the resolution cites an article without naming any law null, and guidance: []
no_text_loaded the norm is in the catalog, but its text is not loaded null
no_date the annotation has no date null
out_of_window no text of that norm applied on that date null
article_not_found that day's text has no such article null
ambiguous_article that day's text has two permanent articles with that key (get_legal_norm_article answers 409) null
letter_not_found the article has no such letter null

⚠ The date used is the annotation's, taken as the resolution's. Near a text change (2022-05-09, 2022-08-06) an annotation published days after the resolution would read the following text: that is why text.asOf says which date was used. And a transitional article cannot be told apart in the citation: the permanent article with that number is read.

⚠ Two exceptions, both in the Regulation of Law 19.039 and both because of the template INAPI copies.

1 · Art. 9 after 2022-05-09 (chosenBy: "former_regulation_template"). On 2022-05-09 Decree 82 repealed Decree 236, and its art. 9 changed subject: in DS 236 it lists what a trademark application must contain («Toda solicitud para el registro de marcas comerciales contendrá…», letters a-d); in DS 82, what a patent application must contain. INAPI kept quoting the DS 236 template in its trademark resolutions. When the annotation's prose carries the template's textual signature —the heading «registro de marcas comerciales contendrá» for any letter; «especificación clara de la marca» or «transliteración» for b; «Resolución Exenta N° 391» for c— the citation is read against DS 236 on its last day in force, and text.asOf is 2022-05-08. Without the signature, art. 9 follows the date like any other citation. Measured in production on 2026-09-29: 254 citations (230 with a letter, 24 without).

2 · Art. 6 and art. 7 before 2022-05-09 (chosenBy: "later_regulation_template"). The mirror case: from December 2017 INAPI cited «art. 6 letter a/b/c/d» and «art. 7 letter b/c» of the Regulation in its trademark resolutions with the numbering and wording of DS 82, which came into force on 2022-05-09; art. 6 and 7 of DS 236, in force on that date, have no letters, so the date rule could not find the letter (letter_not_found). When the citation is earlier than 2022-05-09 and the prose carries the textual signature of its letter —art. 6: a) «marcas comerciales contendrá»; b) «especificación clara» or «especificar con claridad»; c) «Res. Ex. N° 391 de 2017»; d) «idioma distinto al/del español»; art. 7: b) «nombre propio» or «nombre de una persona»; c) «acompañe poder» or «artículos 15 y 22»— the citation is read against DS 82 on its first day, and text.asOf is 2022-05-09, after the act: ⚠ that text did not yet apply on the resolution's date, and chosenBy says so, so that it can be flagged. Art. 7 a) and d) (the «letras a) y d)… nuevos ejemplares» formula, which does not match art. 7 d) of DS 82) and any citation without the signature follow the date. Measured in production on 2026-09-29: 15,353 citations (art. 6: 9,679; art. 7: 5,674), 2017-12-07 → 2022-05-05.

Signals, not a verdict

guidance lists the current INAPI documents (guidelines, circulars, manual, page) that cite that article and that letter, the ones that cite it most first (at most 5), each with:

  • citingCount — how many citations of that article/letter its current version has, and unnamedCount — how many do not name the law;
  • up to 3 sections (path, heading, paragraphNumber, pageFrom, lawNamed): numbered paragraphs first, then the shortest. Read them with get_guidance_section (code + path).

⛔ It is not «the» section that interprets the citation. It is a signal of where INAPI talks about that article, with its count; which section applies to the case is for the reader to decide. (A heuristic that picked one picked chapter 43 instead of 42 in production.)

⭐ Citations without a law. In the trademark guidelines, a bare «art. 20 letter h)» is Law 19.039 (their default law): those citations count only for citations of Law 19.039, and their section says so with lawNamed: false. For the Regulation or any other law it does not apply.

curl -H "X-API-Key: $TARNO_API_KEY" \
  "https://api.tarno.cl/annotations/11397483?expand=citations"
{ "name": "get_annotation",
  "arguments": { "id": "11397483", "expand": ["citations"] } }
status code when
400 validation_error expand other than citations; expand on a list
403 forbidden with expand=citations, the key lacks legal:read (or lacks brands:read)
404 not_found the annotation does not exist

Annotations by party and by current status

POST /annotations/search ≡ search_annotations (scope brands:read) accepts, since 2026-09-28, five more filters. Without them the answer is byte for byte the usual one.

field what it filters
partyId the acts of that party's trademarks (its id from /parties); an unknown one ⇒ 404
taxId the same by RUT, with or without dots and hyphen (at most one of partyId/taxId); not RUT-shaped ⇒ 400 invalid_rut
partyRole holder or representative (absent: both); requires a party
relation current (default), former (retired relation) or all; requires a party
trademarkStatus a list of statuses: only acts of trademarks whose published status today is one of them — the same meaning as status of the trademark search, never INAPI's raw value
# «What happened this month to this RUT's trademarks?»
curl -s -X POST "$TARNO_BASE_URL/annotations/search" -H "X-API-Key: $TARNO_API_KEY" \
  -H 'Content-Type: application/json' \
  -d '{"taxId": "76.086.428-5", "dateFrom": "2026-09-01"}'

The party is first resolved to its trademark numbers (up to 5,000; a larger party takes another form of the same query), so the filter does not walk the corpus by date. Measured in production on 2026-09-28: every path under 2 s. The order (date, id) and the cursor are the usual ones. The batch of annotations by a list of trademarks and the filter by detection date: below.

The term INAPI prints in the act's name: inapiTermLabel

Since 2026-09-29, every annotation of the resource —POST /annotations/search ≡ search_annotations, GET /trademarks/{n}/annotations, GET /annotations/{id} ≡ get_annotation and the daily bulletin— carries the field inapiTermLabel. Some INAPI acts carry in their name the term the resolution itself sets («Fondo - Res. Observaciones Escrito (30 días)»); the field quotes that text:

{
  "section": "M11",
  "act": "661B",
  "actName": "Fondo - Res. Observaciones Escrito (30 días)",
  "inapiTermLabel": {
    "basis": "inapi_label",
    "text": "30 días",
    "code": "661B",
    "quotedFrom": "actName"
  }
}
  • Only a closed list of 14 codes: the administrative, form, substance and registration observations (661, 661B, 661D, 813, 813A, 813B, 813D, 931, 931B, 931D) and those of opposition proceedings (738, 964, 925, 926). Any other act: null; also null when the INAPI name does not carry the term in parentheses.
  • The act first: if the annotation has an act, the quote comes from actName (quotedFrom: "actName"); otherwise from sectionName.
  • Left out on purpose: the appeal (W146, W148), the payment (W570), W015/W023 and 664B.

⛔ A label is not a deadline. inapiTermLabel is quoted INAPI text: it has no date, it is not a milestone and it never appears in milestones[]. No due date is computed from it. Write an alert as a fact with its date: «On 2026-09-11 INAPI published substance observations; INAPI indicates 30 días». The term to answer an opposition, the grant-fee payment and the observations stay without a computed deadline until there is a verified rule.

By a list of trademarks and by detection date

Since 2026-09-29, POST /annotations/search ≡ search_annotations (scope brands:read) answers in one call «which new acts do these trademarks have since last time».

field what it does
applicationNumbers the annotations of up to 100 trademarks, by cursor; duplicates count once. More than 100 distinct ⇒ 422 batch_too_large; more than 1,000 ⇒ 400. Not combinable with applicationNumber, partyId or taxId (400)
detectedAfter only acts this corpus learned of strictly after this instant (ISO 8601 with offset and seconds, 2026-10-01T10:00:00Z; without seconds ⇒ 400; exclusive). With applicationNumber or applicationNumbers, the changes of those trademarks; on its own, the corpus-wide feed
detectedUntil the inclusive end, same format (seconds required). Requires detectedAfter, applicationNumber or applicationNumbers; before detectedAfter ⇒ 400

Every annotation carries denomination (the denomination of its trademark; an empty string when the trademark has none) and detectedAt: when this corpus learned of it, not when INAPI published it (that is date). Imported legacy history carries its load date (2026-06-28/29, 2026-08-22, 2026-08-27): it is for «from now on», not for rebuilding history. With dedupe (the default) a fact counts from its first row: a twin version arriving today does not make an old fact new.

With a list or detection, the answer declares meta.detectedThrough, the safe horizon: the earlier of now and the start of any ingestion run in progress. Nothing detected after it is served, so no row of a half-done run is skipped.

# First time: my list up to now (keep meta.detectedThrough)
curl -s -X POST "$TARNO_BASE_URL/annotations/search" -H "X-API-Key: $TARNO_API_KEY" \
  -H 'Content-Type: application/json' \
  -d '{"applicationNumbers": ["1510102", "1672581"], "detectedAfter": "2026-09-01T00:00:00Z"}'
# Next time: «changes since X», X = the previous detectedThrough
curl -s -X POST "$TARNO_BASE_URL/annotations/search" -H "X-API-Key: $TARNO_API_KEY" \
  -H 'Content-Type: application/json' \
  -d '{"applicationNumbers": ["1510102", "1672581"], "detectedAfter": "2026-09-29T13:24:04.323Z"}'

The cursor pattern: ask with detectedAfter = the previous detectedThrough; page with cursor and the SAME detectedUntil = the detectedThrough of the first page; when done, keep that value for next time. No row is skipped or repeated: whatever the corpus stores after that detectedUntil stays out of the sweep, even as a twin version of a fact already served. Lists over 100: split into batches of 100. Measured in production on 2026-09-29: lists of 100 trademarks in 3–15 ms, no new index.

The corpus-wide feed

detectedAfter without applicationNumber, applicationNumbers or a party (partyId, taxId) returns every act the corpus learned of since that instant, of all trademarks: whoever follows thousands of trademarks reads the feed and filters on their side, instead of asking batch by batch.

  • Detection order, not act-date order: the instant the corpus stored the row and, on a tie, its id. Paged by cursor. With dedupe (the default) each fact comes once, at the position of the row that presents it, and its detectedAt is that of its first row: it can be earlier than its neighbours'.
  • No content filters. It only takes detectedUntil, maxActAgeDays, dedupe, limit and cursor. family, section, citedNorm, article, letter, dateFrom, dateTo, niceClass and trademarkStatus are a 400 naming the field: filter on your side, or name the trademarks. detectedAfter with partyId or taxId and no trademark is a 400 too.
  • Every row carries what an alert needs without another call: applicationNumber, denomination, section with sectionName and sectionLabel, date and detectedAt.
  • Act age: maxActAgeDays (integer 0..3650, requires detectedAfter). Only acts that were at most N days old when the corpus stored them: their date is not before the UTC day of detection minus N. An act dated after its detection always passes; one without date never does. It keeps a historical load (old acts the corpus adds today) out of the feed: the server does not send it to you. The order, the cursor and meta.detectedThrough do not change. In the feed each page reads a bounded stretch of the detection order: it can bring fewer than limit results, or none, with next not null — keep following next until it is null. It also works with applicationNumber or applicationNumbers.
# First page: everything learned since X (keep meta.detectedThrough and next)
curl -s -X POST "$TARNO_BASE_URL/annotations/search" -H "X-API-Key: $TARNO_API_KEY" \
  -H 'Content-Type: application/json' \
  -d '{"detectedAfter": "2026-10-03T13:24:04.323Z", "limit": 100}'
# Following pages: the cursor and, ALWAYS, detectedUntil = the detectedThrough of the first page
curl -s -X POST "$TARNO_BASE_URL/annotations/search" -H "X-API-Key: $TARNO_API_KEY" \
  -H 'Content-Type: application/json' \
  -d '{"detectedAfter": "2026-10-03T13:24:04.323Z", "detectedUntil": "2026-10-04T13:20:11.870Z", "limit": 100, "cursor": "<next>"}'
# Next time: detectedAfter = that detectedThrough

The feed reads through its own index, idx_anotacion_created_at_id (migration …513).