Find the news, filings and calls behind any market question.
curl -sS "https://data.finks.ai/v1/search" \
-H "Authorization: Bearer $FINKS_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"query": "Nvidia short interest, holders and insider selling",
"limit": 10,
"symbols": [
"NVDA"
]
}'
# Next page: the same body plus "cursor": <next_cursor>, while "has_more" is true.YYYY-MM-DD, point in time (behind the account flag api_point_in_time): the search as published by the end of that New York day. Results published at or after the start of the next New York day are not served, reported_facts reads only facts filed on or before it (each period's latest version filed by then), and every other data-block dataset is not read (not_point_in_time). Not after today.
The next_cursor of the previous page, sent with that search's body unchanged except limit. Leave it out for page 1.
asked (the default) or all: which company and fund datasets page 1's data block reads.
standard (the default) or deep, which also ranks each result by its best passage and returns it. Needs the deep_search plan right.
RFC 3339, exclusive: only items whose metadata was computed after this instant. Articles don't carry it yet, so this leaves them out.
Event-type slugs (earnings, m_and_a, ...): only items carrying any of them.
Pools (brief, article, filing, transcript) or related datasets to leave out. GET /v1/search/fields lists them.
Results per page, up to your plan's page cap.
RFC 3339, inclusive.
RFC 3339, exclusive.
Leave it out. false skips re-ranking, for comparisons.
relevance (best match first, the default) or newest.
Tickers: only items about any of them.
The pools to search. Prefer exclude; naming a pool in both is a 400.
What this search read, and why it skipped any. Later pages repeat page 1's.
True while another page follows.
retrieval; jev on accounts whose results are re-ranked; or judge when a deep search's judge ranked them.
Ranked best first, at most limit items, may be empty.
Deep search's answer: a quote of a result's own text, with offsets into the text GET /v1/content/{result_id} serves (a brief's or article's body, a filing's text of section section_id, a transcript's body).
asked (the default) or all: which company and fund datasets page 1's data block reads.
Page 1 only: what the data block is about, in query order: the companies and funds the query names, and the macro series it asks for. Absent on a later page.
Only when a standard search dropped low-scoring results: how many. results can then be empty on a 200, meaning no confident match.
How a search naming 2 or more companies was read, and what page 1 holds about each.
Opaque token for the next page of the same ranked set. null on the last page.
The cutoff a point-in-time search (as_of) applied.
YYYY-MM-DD, as sent: facts filed on or before it.
RFC 3339, exclusive.
depth: "standard", page 1 only: present when the query embedding did not answer within its 2.5 s budget and the page was ranked on keyword matching alone, without semantic similarity (and without filing excerpts). Absent otherwise.
sort: newest, page 1 only: why the page holds fewer than limit results. Present when fewer than limit items matched the query and the read that fills the page with its best relevance matches did not complete (timed out or failed), so the page serves the matches alone. Absent otherwise.
request.invalid naming the field, request.out_of_plan naming the allowed value, invalid_cursor, or cursor_expiredrequest.too_large, the body is over the 2 MB limitcontent.unresolved, a pool types names, or every pool searched, could not answerA deep search records search_deep instead of search_standard.
Page with cursor: resend the same body with the previous next_cursor (only limit may change) while has_more is true. A cursor from another request is 400 invalid_cursor; an old one is 400 cursor_expired.
Timestamps are RFC 3339 in UTC. *_after is inclusive (except enriched_after) and *_before exclusive.
curl -sS "https://data.finks.ai/v1/search/fields?format=tool_schema" \
-H "Authorization: Bearer $FINKS_API_KEY"tool_schema returns the catalog as a tool-call schema an agent framework registers directly.
Result fields: the types that carry it.
Result fields: the types that carry it.
One entry per type whose pool is live right now, with its corpus depth.
format=tool_schemarequest.invalid, an unknown parameter or format valuecontent.unresolved, no pool could report its depthFiling: permanent identifier of the cited source span.
YYYY-MM-DD: the calendar row's report date, within a day of the filing.
The calendar's label: earnings_calendar_feed.
Per share, US dollars: the company's adjusted EPS where it reports one, else GAAP.
gaap or adjusted: the verified actual's basis. Verified actuals only.
An actual against its estimate: EPS to the cent, revenue within ±0.5% is inline.
eps_actual - eps_estimate, to the cent. Verified actuals only.
The surprise as a percent of |estimate|, one decimal. Verified actuals only, and absent when |estimate| < 0.10 or the actual and the estimate differ in sign.
Q1..Q4, the reported quarter as the company names it.
The reported quarter's fiscal year as the company names it.
YYYY-MM-DD: the last close before the reaction session.
reaction_close / base_close - 1 in percent, two decimals; split-adjusted closes.
YYYY-MM-DD: the reaction session's date.
The session that priced the news first: pre (before the open on the reaction date), intraday, or after (the reaction is a later trading day). The price_* fields are all present or all absent.
Total revenue, US dollars.
An actual against its estimate: EPS to the cent, revenue within ±0.5% is inline.
As eps_surprise_pct.
Filing: the event's type, one of the event-type allowlist's slugs (earnings, insider_transaction for a Form 4, insider_sale_notice for a Form 144, beneficial_ownership for a Schedule 13D/13G, periodic_report, ...). Absent on other types and when the filing's type is not on the allowlist.
Filing: verbatim text from the source document.
Filing, beside excerpt only: the excerpt passage's citation id.
Filing, beside excerpt only: lower-hex SHA-256 of excerpt exactly as served (its UTF-8 bytes, no normalization).
Filing, beside excerpt only: the excerpt passage's document URL.
The reporting owner, as filed (Parekh Kevan).
Whether the filing marks its transactions as made under a Rule 10b5-1 plan. Absent when the filing does not say.
Filing: the event ledger id (le1:) of the real-world event this filing proves, shared by an 8-K and its 8-K/As. Absent when the ledger holds no row for it. On a company-timeline brief or article: the event of the 8-K folded into it.
Deep search: which half ranked a result.
Deep search: the passage of a result that best matches the query: a brief's or article's body, or one of a filing's sections.
CommonMark. Brief, article: Finks-written, never a publisher's words. Filing: verbatim filing text.
Unicode code-point offset, exclusive. Absent with start.
Filing only: the section the passage is from, one of the sections[].section_id GET /v1/content/{id} serves.
Unicode code-point offset, inclusive, into the text GET /v1/content/{id} serves: a brief's or article's body, a filing's text of section section_id, a transcript's body. Absent when the offsets could not be verified against text.
Brief, article: publisher domains behind the text.
The document type: press_release, litigation_release, administrative_proceeding, enforcement_action, corporate_decision, conditional_approval, interpretive_letter, crs_report. Open vocabulary.
Transcript: what url links to, e.g. sec_8k (the SEC EDGAR index page of the earnings-release 8-K the call discusses), never a transcript page. Other values may appear: treat an unknown one as another kind of source link. Absent without a url.
Unicode code-point offset, inclusive.
Exactly the served text between start and end. Brief, article: Finks-written; filing, transcript: the source's words.
Filing only: the section quoted, one of the record's sections[].section_id.
A private company's company_key (anduril), the company of GET /v1/funding and GET /v1/federal-awards; present on private_company only.
The name a typo in matched was read as (nvdia -> nvidia).
The listing's, fund's or series' name, when known.
A macro series' /v1/macro/series id (cpi_u_sa); absent otherwise.
The ticker the data block reads a company or fund by; absent on a macro series and a private company.
Page-1 results about it (its subject band, a filing or call of its).
What a query naming 2 or more companies asks for.
Page-1 results about 2 or more of entities.
What decided a MultiEntityIntent: the word lists, or nothing (default: they could not tell, so the query ranks as one topic, unbalanced).
Closed vocabularies: the live values.
Closed vocabularies: the live values.
briefarticlefilingtranscriptregulator