deepxiv/deepxiv_sdk/deepxiv-fa-filings

DeepXiv · Filings (filings)

~1.32M corporate filings and regulatory documents across US and China; the vector layer and the Postgres layer are kept in sync (semantic plus lexical): ~395k EDGAR 8-K events, 154k 6-K, 188k 13D/13G ownership filings, 47k structured 10-K/20-F, 105k 10-Q quarterlies, 32k DEF 14A, 28k Form 425, 7k S-1, 89k SEC comment-letter documents (CORRESP and UPLOAD), 119k Chinese periodic reports, 142k Chinese interim announcements, 10k exchange inquiry letters and discipline decisions, 2.4k CSRC penalty decisions (measured 10-15; Synced daily; exact counts are in corpus (live) — per-source counts in corpus.sources). Default search excludes 10-Q (numerous and topically similar, they would crowd out everything else); filter on form or doc_family to include them (e.g. form=10-Q, doc_family=periodic). One-liners, counterparties, amounts, MD&A drivers, inquiry points and violation types are model-extracted; company name / tickers / SIC come from EDGAR submissions.

Corpus: 1,318,627 documents, updated 2026-10.

Setup

pip install -U deepxiv-sdk      # deepxiv registers a token on first use

With a registered key (https://data.rag.ac.cn/register), search and read are fair use: 60/min, 5 at once, 5,000/day; on the auto-registered token they spend the deepxiv daily limit at the credits listed below. ask draws on the daily ask allowance (resets 00:00 UTC; a site-wide daily total can run out early at peak times), or 3 lifetime trial asks on the auto-registered token. On 429, back off for Retry-After seconds. Data goes to stdout; cost, quota and sources go to stderr. Exit 0 ok, 1 API error, 2 usage error. The general CLI guide is the deepxiv-cli skill.

Start here

deepxiv fa search filings "CEO resignation" -F market=us -F event_types=officer_director_change -F "filed_year>=2023" -k 5
deepxiv fa search filings "*" -F form=10-Q -F cik=320193 --sort date -k 4   # 10-Q is excluded unless you filter on it
deepxiv fa read filings us:861838:0001104659-23-092550 -l brief
deepxiv fa ask "Which US companies disclosed a CEO departure in 2024 and why?" -d filings

Free samples: ids us:861838:0001104659-23-092550; queries CEO resignation.

Search

deepxiv fa search filings "query" [-F FILTER]... [-k N] [--offset N] [--mode M] [--sort S] [--head] [--json]

Modes: hybrid, dense, bm25. Up to 100 hits. Cost 2 (with --head: 3). Sorts: relevance, date (default relevance). An empty query is rejected; use "*" to list by filters only. Filters combine with AND; an undeclared field exits 2.

Advanced filters (see deepxiv fa spec filings): items, sic, letter_kind, exchange, issue_categories, violation_types, company_name, cik, sec_code, tickers, counterparties, risk_categories, party_names, has_reply, has_ban.

Read

deepxiv fa read filings ID [ID...] [-l brief|detail|full] [-e EXTRA] [-p k=v]... [-f json|md|html]

At -l full, -p section=NAME reads one section and -p characters=N the first N characters.

Facets

deepxiv fa facets filings FIELD lists the legal values and counts, free. Use it instead of guessing a spelling. Fields: market, doc_family, form, event_types, report_kind, letter_kind, exchange, issue_categories, violation_types, sic.

Ask

deepxiv fa ask "question" -d filings [-e low|medium|high] [-p k=v]... [--no-stream] [--json]

Agentic Q&A over this domain: the model turns the question into filters plus keywords from the domain's own self-description, reads the hits and answers with [filings:<doc_id> <title>] citations; /stream is NDJSON with real token deltas

Always pass -d filings: automatic domain picking is still early.

What this corpus will and will not tell you

• Fields differ by doc_family: event (8-K) has items / event_types / counterparties / amounts; periodic has main_business / segments / key_numbers; 10-Q quarterlies have mdna_drivers / liquidity / legal_updates / new_or_changed_risks / outlook; inquiry_letter has issues / deadline / has_reply; penalty has violation_types / facts / legal_basis / parties. Filter by doc_family before reading fields.
• The ~105k 10-Q quarterlies are in the vector layer but excluded from default search: with no form / doc_family filter the results contain no 10-Q. To search quarterlies, filter form=10-Q (semantic search) or doc_family=periodic (ranked together with 10-K, 20-F and Chinese periodic reports).
• Sorting (sort=date) and day-precision filed_at ranges exist only in the Postgres layer: either one routes the whole query there (lexical search), and that path does not apply the 10-Q exclusion (a company's filings listed by date should include its quarterlies). Every hit is tagged with its layer and the response says which layer answered.
• Lexical search in the Postgres layer also matches model-written retrieval keywords (kw_en / kw_zh, folded into the full-text index), so a Chinese query such as 持续经营 finds English going-concern language and synonyms absent from the source text. The trade-off: the score includes that contribution and the response does not distinguish a source-text match from a keyword-only match. The keywords themselves are not returned (the store marks them recall-only).
• Financial statements from 10-K / 20-F / 10-Q are not here — use XBRL; this corpus carries structured highlights only.
• Source documents are on NAS (raw_path in level=detail); the index holds no full text.
• The index holds no full text: level=full is not available yet (source documents are on NAS (raw_path); neither layer holds the body text, only structured highlights).

Citing

Cite records as [8-K 0001104659-23-092550]. Say plainly when the corpus has nothing.

Generated by scripts/gen_fa_skills.py from deepxiv fa spec filings --json on 2026-10-19. Field names and prices can change; deepxiv fa spec filings is the live source.