Wire your service to the corpus

# what we ship you (systemd, one command to install)
lexatlas-tunnel@yourservice.service        # the tunnel itself, auto-restart
lexatlas-tunnel-check@yourservice.timer    # liveness check: is the socket answering, not just running

# after install, from your host:
curl -s -H "X-API-Key: $LEXATLAS_KEY" http://127.0.0.1:8788/v1/health

curl -s -H "X-API-Key: $LEXATLAS_KEY" \
  "http://127.0.0.1:8788/v1/norm/ru:gk_rf_1:395"

curl -s -H "X-API-Key: $LEXATLAS_KEY" http://127.0.0.1:8788/v1/limits

{
  "key_name": "your-service",
  "rate_per_min": 300,
  "active": true,
  "org": { "slug": "your-org", "name": "Your Company" },
  "remaining_this_minute": 296,
  "batch": {
    "norms_max_cites": 50,
    "citecheck_batch_max_documents": 20,
    "answer_pack_max_fragments": 20,
    "changes_max_limit": 1000,
    "max_body_bytes": 1048576
  },
  "rate_model": {
    "kind": "token_bucket",
    "refill_per_min": 300,
    "burst": 300
  },
  "on_limit": "HTTP 429 + Retry-After (seconds until the next token, not until the minute ends)"
}

curl -s -H "X-API-Key: $LEXATLAS_KEY" \
  "http://127.0.0.1:8788/v1/norm/ru:gk_rf_1:395"

{
  "cite": "ru:gk_rf_1:395",
  "title": "Ответственность за неисполнение денежного обязательства",
  "act_title": "Кодекс Российской Федерации от 30.11.1994 № 51-ФЗ «Гражданский кодекс …»",
  "status": {
    "value": "in_force",
    "operative": true,
    "confirmed": true,
    "basis": "source_status",
    "evidence": "статус акта у источника: «Действует c изменениями»"
  },
  "language": {
    "lang": "rus",
    "status": "authentic",
    "authentic": true,
    "note": "язык принятия — текст имеет юридическую силу"
  },
  "confidence": "official_consolidated",
  "provenance": {
    "source": "pravo.gov.ru/ips",
    "official_url": "http://pravo.gov.ru/proxy/ips/?docbody=&nd=102033239",
    "license": { "reuse": "reconstruct", "regime": "public_domain", "quote_verbatim_allowed": true }
  },
  "knowledge_cursor": 1691710
}

# the corpus clock right now
curl -s -H "X-API-Key: $LEXATLAS_KEY" http://127.0.0.1:8788/v1/knowledge
{ "at": "2026-08-20T20:38:50Z", "changes_cursor": 1691710 }

# the rule as it stood on a past date
curl -s -H "X-API-Key: $LEXATLAS_KEY" \
  "http://127.0.0.1:8788/v1/norm/ru:gk_rf_1:395?as_of=2019-01-01"

GET/v1/norm/{cite}

One article: text, act, status, freshness, provenance. Parameters: lang, as_of.

curl -s -H "X-API-Key: $LEXATLAS_KEY" \
  "http://127.0.0.1:8788/v1/norm/ru:gk_rf_1:395?lang=rus"

The cite format is jurisdiction:act_code:article. Act codes are ours and have changed before — if you store anything long-term, store the requisites and resolve through /v1/act instead.

POST/v1/norms

Up to 50 cites in one call — use it instead of a loop; it is one rate-limit token, not fifty.

curl -s -X POST -H "X-API-Key: $LEXATLAS_KEY" -H "Content-Type: application/json" \
  -d '{"cites":["ru:gk_rf_1:395","ru:gk_rf_1:196"],"lang":"rus"}' \
  http://127.0.0.1:8788/v1/norms

GET/v1/verify/{cite}

The cheap question: does this citation exist, is it in force, may it be quoted verbatim?

curl -s -H "X-API-Key: $LEXATLAS_KEY" \
  "http://127.0.0.1:8788/v1/verify/ru:gk_rf_1:196"
{ "cite": "ru:gk_rf_1:196", "exists": true, "in_force": true, "status": "in_force" }

GET/v1/history/{cite}

Amendment history of an article: when it changed and by which law. This is the layer that cannot be reconstructed after the fact — it exists because we recorded it as it happened.

curl -s -H "X-API-Key: $LEXATLAS_KEY" \
  "http://127.0.0.1:8788/v1/history/ru:gk_rf_1:395"

GET/v1/act · GET/v1/act/toc

Resolve an act by its requisites (date + number), then list its structure. Requisites do not age; codes do.

curl -s -H "X-API-Key: $LEXATLAS_KEY" \
  "http://127.0.0.1:8788/v1/act?jur=ru&date=1994-11-30&no=51-%D0%A4%D0%97"

{
  "jurisdiction": "ru",
  "query": { "date": "1994-11-30", "number": "51-ФЗ", "number_matched_as": "51" },
  "found": true,
  "count": 3,
  "acts": [
    { "cite_prefix": "ru:gk_rf_1", "is_canonical": true, "in_force": true,
      "kind": "codex", "text_available": true, "articles": 594, "articles_indexed": 594 }
  ]
}

GET/v1/search

Hybrid retrieval — full-text and vector channels fused by RRF, across jurisdictions. Parameters: q, jur, k, plus act to restrict to a single act.

curl -s -H "X-API-Key: $LEXATLAS_KEY" -G \
  --data-urlencode "q=ответственность за неисполнение денежного обязательства" \
  --data "jur=ru&k=5" \
  http://127.0.0.1:8788/v1/search

Each row carries cite, act_code, act_title, act_kind, status, language, signal (which channels matched) and rrf. Two response fields are worth wiring into your code from day one:

• act_filter.known — present always when you filter by act. An unknown act code returns an empty list with known:false: that is a fact about our corpus, not about the law.
• situation_used — present when the question matched a verified institution mapping and we therefore asked our index a different, shorter question. Your text is never rewritten; match tells you whether it was recognised exactly or generalised.

POST/v1/answer_pack

Everything an answer needs in one call: ranked fragments, mandatory citations, the outcome verdict and the envelope. This is what a legal assistant should call instead of assembling five requests.

curl -s -X POST -H "X-API-Key: $LEXATLAS_KEY" -H "Content-Type: application/json" \
  -d '{"q":"оплата товара по договору поставки просрочка оплаты покупателем","jur":"ru","k":8}' \
  http://127.0.0.1:8788/v1/answer_pack

{
  "outcome": "supported",
  "situation_used": { "primary": "оплата товара покупателем", "match": "exact" },
  "must_cite": [ { "cite": "ru:gk_rf_2:486" }, { "cite": "ru:gk_rf_2:516" }, { "cite": "ru:gk_rf_1:395" } ],
  "fragments": [
    { "cite_base": "ru:gk_rf_2:486", "relevance": { "grade": "strong" } },
    { "cite_base": "ru:gk_rf_2:516", "relevance": { "grade": "companion" }, "companion_of": "расчёты за поставляемые товары" }
  ]
}

GET/v1/changes

The changefeed. Ask by cursor (recommended) or by since date; jur and limit narrow it.

curl -s -H "X-API-Key: $LEXATLAS_KEY" \
  "http://127.0.0.1:8788/v1/changes?cursor=1691000&jur=ru&limit=2"

{
  "since_cursor": 1691000,
  "summary": { "article_added": 1269587, "article_modified": 300006, "article_removed": 17798 },
  "cause_summary": { "source_change": 1384240, "artifact_fix": 136758, "reparse": 66845 },
  "cause_note": "source_change — the law itself changed; any other cause is OUR correction of the
                 same original text. If you only want legal changes, filter on `cause`.",
  "items": [ { "cite": "ru:…", "change": "article_modified", "at": "…", "cursor": 33910 } ],
  "order": "cursor_asc",
  "page": { "limit": 2, "returned": 2, "truncated": true }
}

Filter on cause. Most integrations want source_change only — otherwise our own re-parsing shows up in your product as "the law changed", which it did not.

GET/v1/calendar/upcoming · GET/v1/calendar/stats

What comes into force next, with the sentence that says so. A predictive fact, quoted, not inferred.

curl -s -H "X-API-Key: $LEXATLAS_KEY" \
  "http://127.0.0.1:8788/v1/calendar/upcoming?jur=ru&months=3"

{
  "from": "2026-08-20", "to": "2026-11-19", "jur": "ru", "count": 544,
  "events": [
    { "date": "2026-09-01", "cite": "ru:ru_fz105_2026:3",
      "snippet": "Настоящий Федеральный закон вступает в силу с 1 сентября 2026 года." }
  ]
}

GET/v1/coverage/{jur}

Provable completeness, not a marketing number: the source's own count, our count inside the same slice, and the evidence URL the count was read from.

curl -s -H "X-API-Key: $LEXATLAS_KEY" http://127.0.0.1:8788/v1/coverage/kz

{
  "jurisdiction": "kz",
  "known": true,
  "universe": [ { "doc_type": "документы", "source_total": 229237, "our_total": 228271,
                  "measured_at": "2026-08-20", "evidence": "https://adilet.zan.kz/… → regex" } ],
  "catalog": { "rows": 228271, "with_source_status": 228244, "pct_of_universe": 99.6 },
  "corpus": { "acts": 74, "articles": 19821, "articles_indexed": 19821 }
}

catalog is what we have catalogued; corpus is what we have with full text and index. They are different numbers and we never merge them.

GET/v1/transparency/{jur} · GET/v1/authority/{jur} · GET/v1/observed/{jur} · GET/v1/brief/{jur}

The observation layer around the corpus:

• transparency — publication lag and how much of a jurisdiction is machine-observable;
• authority — the state body map and who is the official publisher for each act type;
• observed — what our observation points actually saw (issues, acts, bills), before parsing;
• brief — the daily country brief in three planes: legislation, media, pipeline.

curl -s -H "X-API-Key: $LEXATLAS_KEY" "http://127.0.0.1:8788/v1/brief/kz?date=2026-08-20"

GET/v1/screen

Multipolar sanctions screening — US OFAC, UK OFSI and EAEU lists, matched by name, alias or registration number, quoted verbatim from the list.

curl -s -H "X-API-Key: $LEXATLAS_KEY" \
  "http://127.0.0.1:8788/v1/screen?q=Rosneft&limit=1"

{
  "query": "Rosneft", "hits": 1, "total_matches": 6, "truncated": true,
  "by_jurisdiction": { "us": 1 },
  "matches": [
    { "jurisdiction": "us", "authority": "us_ofac",
      "list_entry": "OPEN JOINT-STOCK COMPANY ROSNEFT OIL COMPANY",
      "subject_type": "entity", "status": "in_force",
      "restriction_kind": "financial_designation",
      "match_via": "alias_exact", "match_score": 1.0,
      "legal_basis": "UKRAINE-EO13662, RUSSIA-E…" }
  ]
}

POST/v1/citecheck · POST/v1/citecheck/batch

Give it a document; it finds every legal reference in the text and checks each one against the corpus: does it exist, is it still in force, was it renumbered. Batch takes up to 20 documents.

curl -s -X POST -H "X-API-Key: $LEXATLAS_KEY" -H "Content-Type: application/json" \
  -d '{"text":"Согласно ст. 395 ГК РФ проценты начисляются на сумму долга.","default_jur":"ru"}' \
  http://127.0.0.1:8788/v1/citecheck

{
  "total": 1,
  "tally": { "ok": 1, "repealed": 0, "not_found": 0, "unresolved": 0 },
  "flagged": 0,
  "findings": [
    { "raw": "ст.395 ГК РФ", "article_no": "395", "jurisdiction": "ru",
      "code_abbrev": "ГК", "cite": "ru:gk_rf_1:395", "verdict": "ok" }
  ]
}

This is the cheapest integration with the highest visible value: run it over drafts your product already produces and you catch dead references before your user signs them.

GET/v1/pipeline

The legislative pipeline across the CIS: bills and laws by stage, with links to the resulting act when it appears. Parameters: jur, stage, since, linked_only, limit.

curl -s -H "X-API-Key: $LEXATLAS_KEY" "http://127.0.0.1:8788/v1/pipeline?jur=ru&limit=2"

GET/v1/notary/tariffs

Regional notary tariffs of the Russian Federation, by subject and scope, with the approving document behind each figure. Parameters: subject, scope, changed_since, limit.

curl -s -H "X-API-Key: $LEXATLAS_KEY" \
  "http://127.0.0.1:8788/v1/notary/tariffs?subject=77&scope=all"

GET/v1/rates

Central bank policy rates — the value that was in force on a given date, plus the full history of change points. A policy rate is an observation with a date, not a norm: it is set by a board decision, so it has no articles and no editions. What we add is not the number — anyone can read that off the central bank site — but the basis on which it applies, returned with a verbatim quote from the law.

as_of is the point of this endpoint: an interest claim under art. 395 of the Russian Civil Code needs the rate of the period of default, not today's. in_force_to is derived from the next change point; null means "no end yet", not "forever". An as_of before the first known point returns found: false and says where the series starts — returning the earliest value instead would pass "we do not know" off as "that is what it was". Parameters: jur, series, as_of, history.

curl -s -H "X-API-Key: $LEXATLAS_KEY"   "http://127.0.0.1:8788/v1/rates?jur=ru&as_of=2022-03-15"

{
  "jurisdiction": "ru",
  "found": true,
  "rates": [
    {
      "id": "ru_key_rate",
      "title": "Ключевая ставка Банка России",
      "unit": "% годовых",
      "as_of": "2022-03-15",
      "value": "20.00",
      "in_force_from": "2022-02-28",
      "in_force_to": "2022-04-10",
      "is_current": false,
      "agency": { "short_name": "ЦБ РФ", "url": "https://www.cbr.ru/" },
      "basis": {
        "act": "ru:fz86_rf",
        "article": "37",
        "verbatim": "Банк России может устанавливать одну или несколько процентных ставок…"
      },
      "coverage": {
        "first_change_point": "2013-09-17",
        "last_change_point": "2026-07-27",
        "change_points": 66
      },
      "license": "unknown"
    }
  ],
  "knowledge": { "at": "2026-08-30T01:52:10Z", "changes_cursor": 1812164 }
}

Source licence is unknown until we review the terms of use of the source — the field ships with every row so that it does not surface at procurement time.

GET/v1/deploys?since=<cursor>

Our deployment journal. Every change we ship that can affect your answers appears here with a monotonic cursor, so your regression suite can start itself when — and only when — something moved.

curl -s -H "X-API-Key: $LEXATLAS_KEY" "http://127.0.0.1:8788/v1/deploys?since=108"

{
  "deploys": [
    { "cursor": 110,
      "deployed_at": "2026-08-20T19:19:02Z",
      "closes": [ { "id": "P-2026-08-19-038",
                    "check": "ops/test_situation_layer.py + ops/check_enumerations.py",
                    "available_via_api": true } ],
      "retracts": [],
      "note": "…what changed, in plain words…" }
  ]
}

• closes is always present, empty list included — "nothing closed" and "field missing" are different facts;
• available_via_api sits on each identifier: "fixed" and "reachable by you" are separate events, and the difference once cost a consumer a day of downtime;
• retracts carries statements we withdraw — knowledge that travelled by machine has its cancellation travel the same way.

POST/v1/findings · GET/v1/findings?since=<cursor>

Send us a defect with the same key you read with. We triage it hourly, automatically.

curl -s -X POST -H "X-API-Key: $LEXATLAS_KEY" -H "Content-Type: application/json" \
  -d '{
        "finding": {
          "id": "YOURPREFIX-2026-08-20-001",
          "blocking": false,
          "title": "expected norm does not reach top-5 for a supply-contract question",
          "class": "corpus.rank.norm_not_reached"
        },
        "cases": [
          { "task": "T001",
            "corpus_query": "оплата товара по договору поставки просрочка оплаты покупателем",
            "jur": "ru",
            "expected": ["ru:gk_rf_2:486", "ru:gk_rf_2:516"],
            "observed": ["ru:gk_rf_2:489"] }
        ]
      }' \
  http://127.0.0.1:8788/v1/findings

Three fields are mandatory, and the reason is the same for all three — a finding we cannot reproduce by machine is a finding we cannot close:

1. finding.id — your identifier. We close it under your number, not ours;
2. finding.blocking — boolean, explicitly;
3. cases[].corpus_query — verbatim, exactly the string you sent us.

Then read the state back with GET /v1/findings?since=<cursor>: statuses move through получено → воспроизведено → закрыто, and the closing deploy appears in /v1/deploys with your id in closes.

curl -s http://127.0.0.1:8788/v1/errors   # no key needed: the contract describes itself