xbrlkit

Work with XBRL filings above Arelle: fetch a filing, parse it once into a neutral typed model, and project that model into whichever portable representation you need.

EDGAR ──▶ Arelle ──▶ XbrlModel ──┬──▶ holon.jsonld   (RDF / JSON-LD)
                                 ├──▶ Tavi           (compiled model)
                                 ├──▶ xBRL-JSON      (OIM)
                                 └──▶ …

Arelle stays the parser — nobody should reimplement DTS resolution. What it does not give you is anything ergonomic to hold: ModelXbrl is a large mutable object graph tied to a controller you have to close. XbrlModel is the answer to that — stateless, single-filing, lossless, and the waist every projection hangs off.

The one architectural rule: everything goes through XbrlModel. A feature that reaches into Arelle's ModelXbrl directly is bypassing the waist, and that is the change that turns a kit into a junk drawer.

Projections

The OIM projection is the one with a released reference implementation to check against: Arelle's saveLoadableOIM writes the same document from the same filing. A second writer is redundant as a feature — its value is that every difference is a fidelity bug in the parse or the model, and those same bugs are otherwise silent in the holon output, which has nothing to check it. Current parity is every fact on 3M FY2024 (3,150) and Boeing FY2024 (2,688), and all but one on Microsoft FY2024 (1,855 of 1,856); footnotes are the one construct the model does not carry.

Tavi is a public working draft and its name is explicitly a working title, so treat that projection as tracking a moving target. It has been diffed, object class by object class, against the compiled model Arelle's unreleased XbrlModel plugin (Arelle PR #2418) writes for 3M FY2024; the two agree on every fact outside that plugin's own defects and on every cube. Where the draft left a choice open, the choice and its reason are recorded in SPEC_AMBIGUITIES and carried in the .tavi.gaps.json sidecar --format tavi writes alongside the document — the sidecar also records what the filing carries that the model has nowhere to put, and that file is the point of the projection, not a by-product of it.

Property graph

xbrlkit build --format lpg (with the lpg extra: pip install "xbrlkit[lpg]") writes the filing as a single-file LadybugDB database with the tables the RoboSystems sec graph is built from — the same node labels, relationship types, columns and ids, declared once in xbrlkit.schema — so Cypher written against the shared graph runs on the file and a fact in either is the same row. What the platform adds after projection is not in the file: text blocks stay inline in Fact.value, and the enrichment columns and tables (canonical_concept, canonical_type, FactSet, Classification) are empty. The projection is checked row for row against the platform's own processor on the Filing Ladder's 26-filing corpus; the two explained differences are association ids (random on the platform, derived from the arc here) and exact duplicate arcs inside Arelle's aggregate XBRL-dimensions network, which the derived ids collapse.

from xbrlkit.serialize import to_graph_tables, write_parquet, build_lbug

tables = to_graph_tables(model)          # node and relationship rows, schema order
write_parquet(tables, Path("out/mmm"))   # nodes/*.parquet, relationships/*.parquet
build_lbug(tables, Path("out/mmm.lbug")) # CREATE TABLE … + COPY FROM, one file

Text

xbrlkit.text reads the filing's primary HTML document — no Arelle, no network — and returns its text as sections:

Both render HTML tables as markdown pipe tables and split a long section into balanced parts at paragraph boundaries (part, part_count, and a label like "MD&A (2/6)") instead of truncating it. Measured on a 26-filing corpus of 2024–2025 10-Ks and 10-Qs: every text block's full text is carried, where a map of outermost continuations alone lost 15–29% of the note text on nine of the filings, and every target Item starts at its body heading.

from xbrlkit.text import iXBRLParser, NarrativeExtractor

html = open("mmm-20241231.htm").read()
for s in iXBRLParser().parse(html):
  print(s.section_id, s.label, s.word_count, s.xbrl_elements[:3])
for s in NarrativeExtractor().extract(html, form_type="10-K"):
  print(s.section_id, s.label, s.word_count)

Install

As a package

pip install xbrlkit

Exposes the xbrlkit CLI (xbrlkit build …, xbrlkit fetch …, xbrlkit query …) and the library — use this to consume it from another project. Set your SEC User-Agent via the environment (see SEC User-Agent).

From source (development)

# Install the toolchain
brew install uv just

# Install dependencies and provision .env from the template
just install

just install creates .env from .env.example on first run — then set your SEC User-Agent in it.

SEC User-Agent

SEC EDGAR requires a descriptive User-Agent on every request, or it throttles you (empty responses / HTTP 429). just install already created your .env — set your details there:

# .env
SEC_GOV_USER_AGENT="Your Name your@email.com"

.env is loaded automatically by every command. Outside the just workflow, export SEC_GOV_USER_AGENT="Your Name your@email.com" or pass --user-agent.

Usage

# Build a holon.jsonld from a specific filing (-> ./output/)
xbrlkit build --cik 320193 --accno 0000320193-23-000106

# Fetch the latest filing for a ticker (-> ./output/)
xbrlkit fetch --ticker NVDA

# Query consolidated facts in a built holon (in-memory SPARQL)
xbrlkit query --in output/0000320193-23-000106.holon.jsonld --element us-gaap:Assets

From a source checkout, just wraps the same CLI as a shorthand: just build 320193 0000320193-23-000106 and just fetch NVDA.

View & explore

Built holons render in the RoboSystems Holon Viewer — a browser-based reader that renders the financial statements and lets you ask questions of the report with AI:

• Hosted: https://holon.robosystems.ai/ — open a holon.jsonld and explore the statements, notes, and dimensional facts, or chat with the report.
• Source: https://github.com/RoboFinSystems/robosystems-holon-viewer — run it locally or self-host.

The viewer reads a holon entirely client-side, so a single holon.jsonld is a complete, portable, self-describing report.

License

This project is licensed under the MIT License - see the LICENSE file for details.

MIT © 2026 RFS LLC