omnidoc-excel-extractor 0.1.0

Comprehensive Excel extraction SDK — structured chunks for AI/LLM/RAG pipelines

omnidoc-excel-extractor

Comprehensive Excel extraction SDK — 57 structured chunk types for AI, LLM, and RAG pipelines.

Supports .xlsx · .xlsm · .xlsb · .xls

---

Table of Contents

• Overview
• Architecture
• Installation
• Quick Start
• Extraction Options
• Convenience Methods
• ExtractionResult API
• All 57 Chunk Types
  • Structural
  • Semantic
  • Analytical
  • Validation / QA
  • Visual — Charts
  • Visual — Pivot Tables
  • Visual — Other Objects
  • VBA / Macro
  • Cross-Reference / Linkage
  • Connectivity / Power Features
  • Operational / Metadata
• Format Support Details
• RAG / LLM Integration
• Advanced Usage
• Development Setup
• Publishing to PyPI
  • One-time Setup
  • Build the Package
  • Publish to TestPyPI
  • Publish to PyPI (Production)
  • Versioning
• License

---

Overview

omnidoc-excel-extractor converts any Excel file into a structured list of chunks — typed, JSON-serialisable Pydantic objects that capture every layer of meaning in the workbook: raw data, schema, formulas, KPIs, charts, pivot tables, VBA macros, cross-sheet relationships, Power Query, and more.

Each chunk carries:

Field	Description
chunk_id	UUID — globally unique identifier
chunk_type	String name of the chunk class
layer	structural / semantic / analytical / validation / visual / vba / cross_ref / connectivity / operational
priority	must / should / nice — extraction importance tier
(type-specific fields)	All fields defined per chunk type

Why chunks? LLM context windows are finite. Chunking lets you embed, retrieve, and inject only the Excel knowledge relevant to each query — instead of dumping raw cell values into a prompt.

---

Architecture

omnidoc_excel_extractor/
│
├── __init__.py                  ← public API (ExcelExtractor, ExtractionResult)
│
├── chunks/
│   └── models.py                ← 57 Pydantic v2 chunk models + CHUNK_REGISTRY
│
├── core/
│   ├── extractor.py             ← orchestrator — runs every builder in order
│   ├── workbook_parser.py       ← workbook metadata, named ranges, external links
│   ├── sheet_parser.py          ← rows, merged cells, groups, validations, comments
│   ├── formula_extractor.py     ← formula detection, classification, lookup/agg detection
│   └── vba_extractor.py         ← VBA module extraction via oletools
│
├── builder/
│   ├── structural.py            ← chunks 1-9
│   ├── semantic.py              ← chunks 10-15
│   ├── analytical.py            ← chunks 16-23
│   ├── validation.py            ← chunks 24-28
│   ├── visual.py                ← chunks 29-41
│   ├── vba.py                   ← chunks 42-46
│   ├── crossref.py              ← chunks 47-50
│   ├── connectivity.py          ← chunks 51-54
│   └── operational.py           ← chunks 55-57
│
├── utils/
│   ├── cell_utils.py            ← cell address parsing, range iteration
│   ├── type_inference.py        ← dtype detection, semantic role, normalisation
│   └── stats.py                 ← aggregations, outlier detection, trend, frequency
│
└── _adapters/
    ├── xlrd_adapter.py          ← openpyxl-compatible wrapper for .xls (xlrd)
    └── xlsb_adapter.py          ← openpyxl-compatible wrapper for .xlsb (pyxlsb)

Extraction pipeline:

Excel file
    │
    ▼
_load_workbook()  ──► openpyxl (xlsx/xlsm)
                  ──► xlrd adapter  (xls)
                  ──► pyxlsb adapter (xlsb)
    │
    ▼
builder/* ──► 57 chunk builders run in priority order
    │
    ▼
ExtractionResult (chunks: list[BaseChunk], index: ChunkIndexChunk)

---

Installation

pip install omnidoc-excel-extractor

Optional — VBA/macro extraction (.xlsm, .xlsb, .xls):

pip install oletools

Full install with all optional extras:

pip install omnidoc-excel-extractor oletools networkx

Python version: 3.9+

---

Quick Start

from omnidoc_excel_extractor import ExcelExtractor

extractor = ExcelExtractor("report.xlsx")
result = extractor.extract()

# High-level summary
print(result.summary())
# {
#   'total_chunks': 412,
#   'type_counts': {
#     'WorkbookChunk': 1, 'SheetChunk': 4, 'TableChunk': 2,
#     'SchemaChunk': 4, 'RowChunk': 1800, 'FormulaDefinitionChunk': 37,
#     'KPIChunk': 5, 'ChartChunk': 3, ...
#   },
#   'workbook': 'report',
#   'created_at': '2026-05-16T10:00:00'
# }

# All chunks as JSON-serialisable dicts
dicts = result.to_dicts()

# Filter by type
schemas  = result.by_type("SchemaChunk")
formulas = result.by_type("FormulaDefinitionChunk")

# Filter by sheet
sales_chunks = result.by_sheet("Sales")

# Access the master manifest
index = result.index   # ChunkIndexChunk
print(index.type_counts)
print(index.all_chunk_ids)

---

Extraction Options

result = extractor.extract(
    # Selectively include only specific chunk types (list of class name strings)
    chunk_types=["WorkbookChunk", "SchemaChunk", "FormulaDefinitionChunk"],

    # Include RowChunks — can produce thousands of chunks for large sheets
    include_rows=True,

    # Include analytical chunks: AggregationChunk, TrendChunk, OutlierChunk,
    # KPIChunk, TemporalChunk, FormulaDefinitionChunk, FormulaResultChunk
    include_analytics=True,

    # Include visual chunks: ChartChunk, PivotTableChunk, SparklineChunk,
    # SlicerChunk, TimelineChunk, ShapeChunk, ImageChunk, FormControlChunk
    include_visual=True,

    # Include VBA chunks (only activates for .xlsm / .xlsb / .xls files
    # and requires oletools to be installed)
    include_vba=True,

    # Number of data rows per RowChunk batch
    row_batch_size=500,

    # Filter by extraction priority tier:
    #   "must"   — structural core + key analytics + validation
    #   "should" — adds trends, outliers, pivots, slicers, VBA events
    #   "nice"   — adds frozen panes, shapes, images, ribbon XML
    #   None     — include everything (default)
    priority=None,
)

Priority tiers at a glance

Priority	Count	Includes
must	27 types	WorkbookChunk, SheetChunk, SchemaChunk, RowChunk, SummaryChunk, KPIChunk, FormulaDefinitionChunk, AggregationChunk, ValidationChunk, ErrorChunk, ChartChunk, PivotTableChunk, MacroChunk, VBAModuleChunk, NamedRangeChunk, RelationshipChunk, PowerQueryChunk, ChunkIndexChunk, and more
should	22 types	TrendChunk, OutlierChunk, EntityChunk, CellAnnotationChunk, TemporalChunk, LookupMapChunk, DataQualityChunk, ProtectionChunk, PivotFieldChunk, SlicerChunk, TimelineChunk, SparklineChunk, FormControlChunk, ActiveXControlChunk, VBAEventChunk, CustomFunctionChunk, ExternalLinkChunk, DependencyGraphChunk, DataModelChunk, PowerPivotMeasureChunk, ConnectionChunk, PivotCacheChunk
nice	8 types	FrozenPaneChunk, ShapeChunk, ImageChunk, RibbonCustomizationChunk, PrintSettingsChunk, ChangeLogChunk, ChartAnnotationChunk, GroupChunk

---

Convenience Methods

# Workbook overview — just metadata and sheet info
result = extractor.extract_sheets()
# Produces: WorkbookChunk, SheetChunk, SummaryChunk

# Schema discovery — column types and header normalisation
result = extractor.extract_schema()
# Produces: SchemaChunk, TableChunk, ColumnSemanticChunk, HeaderAliasChunk

# Formula audit — every formula, its type, and cross-sheet dependencies
result = extractor.extract_formulas()
# Produces: FormulaDefinitionChunk, FormulaResultChunk, LookupMapChunk

# VBA/macro inventory
result = extractor.extract_vba_only()
# Produces: MacroChunk, VBAModuleChunk, VBAEventChunk, CustomFunctionChunk

---

ExtractionResult API

result = extractor.extract()

result.chunks                         # list[BaseChunk] — all extracted chunks
result.index                          # ChunkIndexChunk — master manifest
result.summary()                      # dict with total_chunks, type_counts, workbook, created_at

result.to_dicts()                     # list[dict] — all chunks as plain dicts (JSON-safe)
result.by_type("SchemaChunk")         # list[BaseChunk] — filter by chunk_type string
result.by_sheet("Revenue")            # list[BaseChunk] — filter by sheet_name attribute

# Each chunk also has .to_dict()
chunk = result.chunks[0]
chunk.chunk_id                        # UUID string
chunk.chunk_type                      # "WorkbookChunk"
chunk.layer                           # "structural"
chunk.priority                        # "must"
chunk.to_dict()                       # dict

---

All 57 Chunk Types

---

Structural — priority: must

Capture the physical layout and raw data of the workbook.

1. WorkbookChunk

One per file. Top-level workbook metadata.

Field	Type	Description
name	str	File stem (without extension)
path	str	Absolute resolved file path
sheet_names	list[str]	All sheet names in order
author	str | None	Creator property
created_at	datetime | None	File creation timestamp
modified_at	datetime | None	Last modified timestamp
file_size_kb	float | None	File size in kilobytes
app_version	str | None	Excel application version
has_macros	bool	True for .xlsm/.xlsb/.xls
is_shared	bool	Shared workbook flag

2. SheetChunk

One per worksheet/chartsheet.

Field	Type	Description
sheet_name	str	Sheet tab name
sheet_index	int	0-based position
sheet_type	str	worksheet / chart / dialog
is_visible	bool	Sheet visibility state
used_range	str | None	e.g. "A1:F100"
tab_color	str | None	RGB or theme colour
is_protected	bool	Sheet protection enabled
zoom_level	int | None	Zoom percentage

3. TableChunk

One per Excel ListObject (Table).

Field	Type	Description
table_name	str	Table display name
sheet_name	str	Host sheet
address	str	Range reference e.g. "A1:F50"
style	str | None	Table style name
has_total_row	bool	Totals row present
has_header_row	bool	Header row present
col_count	int	Number of columns
row_count	int	Number of data rows

4. SchemaChunk

One per sheet — inferred column schema.

Field	Type	Description
sheet_name	str	Source sheet
table_name	str | None	Source table if applicable
columns	list[dict]	[{name, data_type, null_rate, unique_rate, sample_values}]
header_row_addr	str | None	e.g. "A1:F1"
inferred_pk	str | None	Column with 100% unique non-null values

Column data_type values: numeric / date / boolean / text / mixed

5. RowChunk

One per data row (batched for large sheets).

Field	Type	Description
row_index	int	1-based Excel row number
values	dict[str, Any]	{column_name: cell_value}
parent_table	str | None	Table name if row belongs to a table
parent_sheet	str	Sheet name
batch_start	int | None	First row index in this batch
batch_end	int | None	Last row index in this batch
is_total_row	bool	Flagged as a totals/summary row

6. GroupChunk

One per row/column outline group level.

Field	Type	Description
sheet_name	str	Host sheet
direction	str	row or col
level	int	Outline level (1 = outermost)
start_index	int	First row/column index
end_index	int	Last row/column index
is_collapsed	bool	Group collapsed state
label	str | None	Optional group label

7. ParentContextChunk

Breadcrumb injected per sheet — anchors RowChunks in their hierarchy.

Field	Type	Description
workbook	str	Workbook name
sheet	str	Sheet name
table	str | None	Table name
group_path	list[str]	Outline group path
row_range	str | None	Row range of this context

8. MergedCellChunk

One per merged cell region.

Field	Type	Description
sheet_name	str	Host sheet
merge_range	str	e.g. "B2:D4"
merged_value	Any	Value from top-left cell
row_span	int	Number of rows spanned
col_span	int	Number of columns spanned

9. FrozenPaneChunk (priority: nice)

One per sheet with frozen panes.

Field	Type	Description
sheet_name	str	Host sheet
freeze_row	int | None	Number of rows frozen from top
freeze_col	int | None	Number of columns frozen from left
split_horizontal	float | None	Horizontal split position
split_vertical	float | None	Vertical split position

---

Semantic — priority: must/should

Add meaning and context on top of raw data.

10. SummaryChunk

Statistical overview of each sheet/table.

Field	Type	Description
sheet_name	str	Source sheet
table_name	str | None	Source table
row_count	int	Data row count
col_count	int	Column count
numeric_cols	list[str]	Columns detected as numeric
date_cols	list[str]	Columns detected as date/time
text_cols	list[str]	Columns detected as text
null_rate	float	Fraction of null/blank values (0–1)
dupe_rate	float	Fraction of duplicate values (0–1)
size_kb	float | None	Approximate data size

11. SemanticNarrativeChunk

Auto-generated plain-English description of a sheet or table.

Field	Type	Description
scope	str	sheet or table
scope_name	str	Sheet/table name
narrative	str	Human-readable description
subject	str | None	Main topic inferred
time_scope	str | None	Detected time range
grain	str | None	Row grain (e.g. "daily transaction")
confidence	float	0–1 confidence in narrative
generated_by	str	Generator identifier

12. ColumnSemanticChunk

Semantic role for each column.

Field	Type	Description
sheet_name	str	Host sheet
col_name	str	Column header
role	str	id / measure / dimension / date / flag / freetext
unit	str | None	Detected unit (e.g. %, kg, MB)
currency	str | None	Detected currency code (e.g. USD, EUR)
is_pk	bool	Inferred primary key
is_fk	bool	Inferred foreign key

13. HeaderAliasChunk

Header normalisation and alias variants.

Field	Type	Description
sheet_name	str	Host sheet
original_header	str	Raw header text
aliases	list[str]	Common alternative names
abbreviations	list[str]	Short-form variants
normalized_name	str	Snake-case normalised name

14. EntityChunk (priority: should)

Detected named entities across text columns.

Field	Type	Description
workbook	str	Workbook name
entity_type	str	person / org / product / location / date
value	str	Entity string
frequency	int	Occurrence count
source_columns	list[str]	Columns where entity appears
canonical	str | None	Canonical / normalised form

15. CellAnnotationChunk (priority: should)

Cell comments, notes, and tooltips.

Field	Type	Description
sheet_name	str	Host sheet
cell_address	str	e.g. "B3"
annotation_type	str	comment / note / tooltip
text	str	Annotation text content
author	str | None	Comment author
timestamp	datetime | None	When comment was added
resolved	bool	Whether comment is resolved

---

Analytical — priority: must/should

Deep analytics derived from the data.

16. FormulaDefinitionChunk

Every formula cell — source of truth for formula logic.

Field	Type	Description
sheet_name	str	Host sheet
cell_address	str	Cell reference e.g. "B5"
formula_string	str	Full formula text e.g. "=SUM(A1:A10)"
formula_type	str	scalar / array / dynamic / lambda
named_refs	list[str]	Function names used
precedents	list[str]	All cell/range references in formula

17. FormulaResultChunk

Cached computed value paired with its definition.

Field	Type	Description
sheet_name	str	Host sheet
cell_address	str	Cell reference
computed_value	Any | None	Last saved computed value
value_type	str | None	Python type name of value
has_error	bool	True if cell contains an error
error_type	str | None	#REF / #DIV0 / #NA / #VALUE / #NAME / #NULL / #NUM
linked_definition_id	str | None	chunk_id of matching FormulaDefinitionChunk

18. KPIChunk

Key performance indicators detected from label+value cell pairs.

Field	Type	Description
sheet_name	str	Host sheet
kpi_name	str	KPI label (e.g. "Total Revenue")
value	Any | None	Numeric value
unit	str | None	Unit of measure
target	Any | None	Target / budget value
variance	float | None	Actual minus target
variance_pct	float | None	Percentage variance
source_cell	str | None	Cell address of value
period	str | None	Period label if detected

19. AggregationChunk

Pre-computed aggregations for every numeric column.

Field	Type	Description
sheet_name	str	Host sheet
col_name	str	Column
agg_type	str	SUM / AVG / MIN / MAX / COUNT / MEDIAN / STDEV
value	Any | None	Computed result
source_range	str | None	Cell range used
filter_conditions	list[str]	Any filter conditions applied

20. TemporalChunk

Time series profile for date columns.

Field	Type	Description
sheet_name	str	Host sheet
date_col	str	Date column name
frequency	str | None	daily / weekly / monthly / quarterly / yearly
start_date	str | None	First date
end_date	str | None	Last date
gap_count	int	Missing period count
is_sorted	bool	Whether dates are ascending
fiscal_year_start	int | None	Month number for fiscal year start

21. TrendChunk (priority: should)

Statistical trend over a numeric column.

Field	Type	Description
sheet_name	str	Host sheet
col_name	str	Column
direction	str | None	up / down / flat
delta_abs	float | None	Absolute change (last − first)
delta_pct	float | None	Percentage change
periods_compared	int | None	Number of non-null values
regression_slope	float | None	Linear regression slope

22. OutlierChunk (priority: should)

Statistical outlier in a numeric column.

Field	Type	Description
sheet_name	str	Host sheet
col_name	str	Column
outlier_type	str	zscore / iqr / blank_spike / dupe
cell_addr	str | None	Cell address of outlier
value	Any | None	Outlier value
z_score	float | None	Z-score (zscore type only)
severity	str | None	low / medium / high

23. LookupMapChunk (priority: should)

Lookup formula detected in a cell.

Field	Type	Description
sheet_name	str	Host sheet
cell_address	str	Cell containing lookup
lookup_type	str	VLOOKUP / HLOOKUP / INDEX-MATCH / XLOOKUP
key_col	str | None	Lookup key column
return_col	str | None	Return column
lookup_range	str | None	Source lookup range
match_type	str | None	Exact / approximate match

---

Validation / QA — priority: must/should

Data quality and constraint enforcement.

24. ValidationChunk

Data validation rules applied to cell ranges.

Field	Type	Description
sheet_name	str	Host sheet
range_addr	str	Validated range
validation_type	str	list / decimal / date / whole / custom
allowed_values	list[Any]	Dropdown values (list type)
formula	str | None	Custom formula
error_msg	str | None	User-facing error message

25. ErrorChunk

Excel error values found in cells.

Field	Type	Description
sheet_name	str	Host sheet
cell_addr	str	Cell address
error_type	str	#REF / #DIV0 / #NA / #VALUE / #NAME / #NULL / #NUM
formula	str | None	Formula that produced the error
likely_cause	str | None	Human-readable explanation

26. ConditionalFormatChunk

Conditional formatting rules.

Field	Type	Description
sheet_name	str	Host sheet
range_addr	str	Formatted range
rule_type	str | None	Rule type (e.g. colorScale, dataBar)
condition_formula	str | None	Rule formula
threshold	Any | None	Threshold value
format_applied	str | None	Format description
business_meaning	str | None	Inferred business meaning

27. ProtectionChunk (priority: should)

Sheet or range protection settings.

Field	Type	Description
sheet_name	str	Host sheet
scope	str	sheet or range
is_password_protected	bool	Password set
locked_ranges	list[str]	Protected ranges
allowed_edit_ranges	list[str]	Ranges exempt from protection

28. DataQualityChunk (priority: should)

Per-column data quality metrics.

Field	Type	Description
sheet_name	str	Host sheet
col_name	str	Column
blank_count	int	Blank/null cell count
dupe_count	int	Duplicate value count
type_mismatch_count	int	Cells with unexpected type
out_of_range_count	int	Values outside expected bounds
flagged_cells	list[str]	Cell addresses with issues

---

Visual — Charts — priority: must/should

29. ChartChunk

One per embedded chart.

Field	Type	Description
sheet_name	str	Host sheet
chart_name	str	Chart title or generated name
chart_type	str	bar / line / pie / scatter / area / combo / waterfall / funnel / treemap / map / stock
title	str | None	Chart title text
x_axis	str | None	X-axis label
y_axis	str | None	Y-axis label
series	list[dict]	[{name, range}] for each data series
sheet_anchor	str | None	Top-left cell anchor
source_table	str | None	Source table if identifiable

30. ChartSeriesChunk

One per data series in a chart.

Field	Type	Description
chart_id	str	chunk_id of parent ChartChunk
series_name	str | None	Series legend name
source_range	str | None	Data range reference
series_type	str | None	Override type for combo charts
color	str | None	Series colour
trendline_type	str | None	linear / exponential / polynomial / etc.
trendline_formula	str | None	Trendline equation

31. ChartAnnotationChunk (priority: should)

Text boxes, callouts, and data labels on charts.

Field	Type	Description
chart_id	str	Parent chart chunk_id
annotation_type	str	data_label / callout / textbox
text	str | None	Annotation text
cell_ref	str | None	Linked cell
position	dict | None	Position metadata

---

Visual — Pivot Tables — priority: must/should

32. PivotTableChunk

One per pivot table.

Field	Type	Description
sheet_name	str	Host sheet
pivot_name	str	Pivot table name
source_range	str | None	Source data range
source_sheet	str | None	Source data sheet
row_fields	list[str]	Fields in row area
col_fields	list[str]	Fields in column area
value_fields	list[dict]	[{field, agg}] — value fields with aggregation
filter_fields	list[str]	Report filter fields
report_filter_values	dict	Active filter values

33. PivotFieldChunk

One per field in a pivot table.

Field	Type	Description
pivot_id	str	Parent PivotTableChunk chunk_id
field_name	str	Field name
field_type	str	row / col / value / filter
agg_function	str | None	SUM / COUNT / AVERAGE / etc.
sort_order	str | None	asc / desc
subtotal_enabled	bool	Subtotals shown
grouped_by	str | None	Grouping interval

34. PivotCacheChunk (priority: should)

Pivot table cache metadata.

Field	Type	Description
pivot_id	str	Parent PivotTableChunk chunk_id
cache_range	str | None	Cached data range
last_refreshed	datetime | None	Last refresh timestamp
record_count	int | None	Number of cached records
unique_items_per_field	dict[str, int]	Unique value counts per field

35. SlicerChunk (priority: should)

Slicer objects connected to pivot tables or tables.

Field	Type	Description
sheet_name	str	Host sheet
slicer_name	str	Slicer name
field_name	str | None	Filtered field
connected_objects	list[str]	Connected pivot/table names
active_filters	list[str]	Currently selected filter values
style	str | None	Slicer style
sheet_anchor	str | None	Position anchor

36. TimelineChunk (priority: should)

Date timeline slicers.

Field	Type	Description
sheet_name	str	Host sheet
timeline_name	str	Timeline name
date_field	str | None	Date field being filtered
connected_pivots	list[str]	Connected pivot table names
active_period	str | None	Currently selected period
granularity	str | None	year / quarter / month / day

---

Visual — Other Objects — priority: should/nice

37. SparklineChunk (priority: should)

Mini in-cell charts.

Field	Type	Description
sheet_name	str	Host sheet
sparkline_type	str	line / column / win_loss
host_cell	str	Cell containing the sparkline
source_range	str	Data range
color	str | None	Sparkline colour
markers	bool	Markers shown
axis_min	float | None	Axis minimum
axis_max	float | None	Axis maximum

38. ShapeChunk (priority: nice)

Drawing shapes and text boxes.

Field	Type	Description
sheet_name	str	Host sheet
shape_type	str | None	Shape type name
text_content	str | None	Text inside the shape
cell_anchor	str | None	Top-left cell anchor
position	dict	{left, top, width, height}
linked_cell	str | None	Cell linked to shape
shape_name	str | None	Shape object name

39. ImageChunk (priority: nice)

Embedded images.

Field	Type	Description
sheet_name	str	Host sheet
image_type	str | None	png / jpeg / gif / etc.
anchor_cell	str | None	Anchor cell
alt_text	str | None	Accessibility alt text
width_px	int | None	Width in pixels
height_px	int | None	Height in pixels
is_linked	bool	External linked image
source_url	str | None	Source URL (linked images)

40. FormControlChunk (priority: should)

Legacy form controls (button, checkbox, listbox, etc.).

Field	Type	Description
sheet_name	str	Host sheet
control_type	str	button / checkbox / listbox / combobox / spinner / scrollbar
linked_cell	str | None	Linked output cell
value	Any | None	Current value
range_source	str | None	Input range for listbox/combobox
macro_assigned	str | None	Assigned macro name

41. ActiveXControlChunk (priority: should)

ActiveX controls.

Field	Type	Description
sheet_name	str	Host sheet
control_type	str | None	ActiveX class identifier
name	str | None	Control object name
linked_cell	str | None	Linked cell
properties	dict	Raw property bag
event_macro	str | None	Attached event macro
sheet_anchor	str | None	Position anchor

---

VBA / Macro — priority: must/should

Requires oletools and a macro-enabled file (.xlsm, .xlsb, .xls).

42. MacroChunk

One per VBA procedure (Sub / Function).

Field	Type	Description
workbook	str	Workbook name
macro_name	str	Procedure name
module_name	str | None	Containing module
trigger_type	str | None	button / event / auto / ribbon
trigger_event	str | None	e.g. Workbook_Open
description	str | None	Docstring or comment
line_count	int	Number of lines
scope	str	workbook or sheet

43. VBAModuleChunk

One per VBA module.

Field	Type	Description
workbook	str	Workbook name
module_name	str	Module name
module_type	str	standard / class / sheet / workbook
procedure_names	list[str]	All Subs/Functions in this module
line_count	int	Total lines of code
references	list[str]	Library references declared
source_code	str | None	Full VBA source text

44. VBAEventChunk (priority: should)

Event handler procedures.

Field	Type	Description
module	str	Module containing the event
event_name	str	e.g. Workbook_Open, Worksheet_Change
trigger_condition	str | None	When the event fires
affected_range	str | None	Range affected by event
summary	str | None	What the event handler does

45. CustomFunctionChunk (priority: should)

User-defined functions (UDFs) callable from cells.

Field	Type	Description
module	str	Containing module
function_name	str	UDF name
parameters	list[str]	Parameter names
return_type	str | None	Return type if annotated
description	str | None	Function description
usage_cells	list[str]	Cells calling this UDF
is_udf	bool	Always True

46. RibbonCustomizationChunk (priority: nice)

Custom Ribbon UI (customUI XML).

Field	Type	Description
workbook	str	Workbook name
tab_name	str | None	Custom tab label
group_name	str | None	Group within tab
button_label	str | None	Button display label
macro_assigned	str | None	OnAction macro
icon	str | None	Button icon reference

---

Cross-Reference / Linkage — priority: must/should

47. RelationshipChunk

Cross-sheet formula dependencies.

Field	Type	Description
source_sheet	str	Sheet containing the formula
source_cell	str	Cell with the cross-reference
target_cell	str	Referenced cell
target_sheet	str	Referenced sheet
rel_type	str	vlookup / index_match / direct_ref / power_query_feed
formula	str | None	Source formula

48. NamedRangeChunk

Workbook or sheet-scoped named ranges.

Field	Type	Description
range_name	str	Defined name
refers_to	str	Formula or range reference
scope	str	workbook or sheet name
usage_count	int	Number of formulas referencing this name
used_in	list[str]	Cell addresses using this name

49. ExternalLinkChunk (priority: should)

References to other workbooks.

Field	Type	Description
workbook	str	Source workbook
source_cell	str | None	Cell containing the link
target_file	str	Referenced file path
target_range	str | None	Referenced range
last_updated	datetime | None	Last update timestamp
is_broken	bool	Link cannot be resolved
update_mode	str | None	Automatic / manual / on-open

50. DependencyGraphChunk (priority: should)

Formula precedent/dependent graph per sheet.

Field	Type	Description
sheet_name	str	Host sheet
nodes	list[str]	All cell addresses in the graph
edges	list[dict]	[{from, to, type}] directed edges
max_depth	int	Longest dependency chain
has_circular_ref	bool	Circular reference detected

---

Connectivity / Power Features — priority: must/should

51. PowerQueryChunk

Power Query (Get & Transform) queries.

Field	Type	Description
workbook	str	Workbook name
query_name	str	Query name
source_type	str | None	file / db / api / web / sharepoint
source_path	str | None	Connection source path
transformation_steps	list[str]	Applied step names
output_table	str | None	Output table name
last_refreshed	datetime | None	Last refresh timestamp

52. DataModelChunk (priority: should)

PowerPivot data model tables and relationships.

Field	Type	Description
workbook	str	Workbook name
tables	list[str]	Table names in the model
relationships	list[dict]	[{from_table, to_table, on_col}]
dax_measures	list[str]	DAX measure names
dax_columns	list[str]	DAX calculated column names

53. PowerPivotMeasureChunk (priority: should)

Individual DAX measure definitions.

Field	Type	Description
workbook	str	Workbook name
measure_name	str	Measure name
dax_expression	str | None	Full DAX expression
format_string	str | None	Display format
source_table	str | None	Host table
used_in_pivots	list[str]	Pivot tables using this measure

54. ConnectionChunk (priority: should)

External data connection definitions.

Field	Type	Description
workbook	str	Workbook name
connection_name	str	Connection name
connection_type	str | None	ODBC / OLE / web / sharepoint / odata
connection_string_sanitized	str | None	Redacted connection string
refresh_schedule	str | None	Auto-refresh schedule

---

Operational / Metadata — priority: must/nice

55. PrintSettingsChunk (priority: nice)

Print configuration per sheet.

Field	Type	Description
sheet_name	str	Host sheet
print_area	str | None	Print area range
page_break_rows	list[int]	Manual page break row indices
header_text	str | None	Page header text
footer_text	str | None	Page footer text
orientation	str | None	portrait or landscape
fit_to_pages	bool | None	Fit-to-page enabled

56. ChangeLogChunk (priority: nice)

Tracked changes from shared workbook revision history.

Field	Type	Description
sheet_name	str	Host sheet
cell_addr	str	Changed cell
old_value	Any | None	Previous value
new_value	Any | None	New value
author	str | None	Who made the change
changed_at	datetime | None	When the change was made
change_type	str | None	edit / insert / delete

57. ChunkIndexChunk (priority: must)

Master manifest — always the last chunk produced.

Field	Type	Description
workbook_ref	str	Workbook name
all_chunk_ids	list[str]	Every chunk_id in extraction order
type_counts	dict[str, int]	Count of each chunk type produced
source_address_map	dict[str, str]	{chunk_id: "SheetName!A1"} for cell-level chunks
created_at	datetime	UTC timestamp of extraction

---

Format Support Details

Format	Parser	Tables	Charts	Pivot	VBA	Power Query	Notes
.xlsx	openpyxl	✓	✓	✓	—	✓	Full support
.xlsm	openpyxl	✓	✓	✓	✓	✓	Macro-enabled
.xlsb	pyxlsb + adapter	✓	—	—	✓	—	Binary format; limited visual support
.xls	xlrd + adapter	✓	—	—	✓	—	Legacy format (Excel 97–2003)

VBA extraction requires oletools: pip install oletools

Dependency graphs require networkx: pip install networkx

---

RAG / LLM Integration

Feeding chunks into a vector store

from omnidoc_excel_extractor import ExcelExtractor

extractor = ExcelExtractor("financials.xlsx")
result = extractor.extract(include_rows=True, priority="must")

documents = []
for chunk in result.chunks:
    d = chunk.to_dict()

    # Build a text representation for embedding
    text_parts = [f"type: {d['chunk_type']}", f"layer: {d['layer']}"]
    for k, v in d.items():
        if k not in {"chunk_id", "chunk_type", "layer", "priority"} and v is not None:
            text_parts.append(f"{k}: {v}")
    text = "\n".join(text_parts)

    documents.append({
        "id":       d["chunk_id"],
        "text":     text,
        "metadata": {
            "chunk_type": d["chunk_type"],
            "layer":      d["layer"],
            "priority":   d["priority"],
            "sheet":      d.get("sheet_name"),
        },
    })

# e.g. with LangChain, LlamaIndex, Pinecone, Qdrant, Weaviate, etc.
# vector_store.add_documents(documents)

Selective extraction for query routing

# Fast schema-only retrieval for "what columns does this file have?"
schema_result = extractor.extract_schema()
schema_dicts = schema_result.to_dicts()

# KPI retrieval for "what is the total revenue?"
kpi_result = extractor.extract(
    chunk_types=["KPIChunk", "AggregationChunk"],
    include_analytics=True,
)

# Formula audit for "explain this formula"
formula_result = extractor.extract_formulas()

Serialising to JSON

import json

result = extractor.extract()
with open("chunks.json", "w") as f:
    json.dump(result.to_dicts(), f, default=str, indent=2)

---

Advanced Usage

Batch processing multiple files

from pathlib import Path
from omnidoc_excel_extractor import ExcelExtractor

results = {}
for path in Path("./reports").glob("*.xlsx"):
    extractor = ExcelExtractor(str(path))
    result = extractor.extract(include_rows=False)
    results[path.name] = result.summary()

print(results)

Accessing CHUNK_REGISTRY

from omnidoc_excel_extractor import CHUNK_REGISTRY

print(list(CHUNK_REGISTRY.keys()))  # all 57 chunk type names
ChunkClass = CHUNK_REGISTRY["SchemaChunk"]

Constructing a chunk manually

from omnidoc_excel_extractor.chunks.models import KPIChunk

kpi = KPIChunk(
    sheet_name="Dashboard",
    kpi_name="Total Revenue",
    value=1_250_000.0,
    unit="USD",
    target=1_500_000.0,
    variance=-250_000.0,
    variance_pct=-16.67,
    source_cell="B2",
)
print(kpi.to_dict())

---

Development Setup

# Clone the repository
git clone https://github.com/ganeshkinkargiri/omnidoc-excel-extractor-sdk.git
cd omnidoc-excel-extractor-sdk

# Create a virtual environment
python3 -m venv .venv
source .venv/bin/activate          # Windows: .venv\Scripts\activate

# Install in editable mode with dev dependencies
pip install -e ".[dev]"

# Optional extras
pip install oletools networkx

# Run tests
pytest

# Run tests with coverage
pytest --cov=omnidoc_excel_extractor --cov-report=term-missing

Project structure for contributors

omnidoc_excel_extractor/
├── chunks/models.py      ← Add new chunk class here first
├── builder/<category>.py ← Add builder function here
├── builder/__init__.py   ← Export the builder function
└── core/extractor.py     ← Call the builder inside extract()
tests/
├── conftest.py           ← Shared fixtures (programmatic .xlsx)
└── test_extractor.py     ← Tests

Adding a new chunk type:

1. Define the Pydantic class in chunks/models.py and add it to CHUNK_REGISTRY.
2. Write a build_<name>_chunks() function in the appropriate builder file.
3. Export it from builder/__init__.py.
4. Call it inside extract() in core/extractor.py.
5. Add a test in tests/test_extractor.py.

---

Publishing to PyPI

One-time Setup

1. Create accounts

Registry	URL	Purpose
TestPyPI	https://test.pypi.org/account/register/	Test uploads before going live
PyPI	https://pypi.org/account/register/	Production — public installs

2. Generate API tokens

TestPyPI token:

1. Log in to https://test.pypi.org
2. Account Settings → API tokens → Add API token
3. Scope: "Entire account" for first upload, or per-project after
4. Copy the token (shown only once) — starts with pypi-

PyPI token:

1. Log in to https://pypi.org
2. Account Settings → API tokens → Add API token
3. Copy the token

3. Configure ~/.pypirc

Create or edit ~/.pypirc:

[distutils]
index-servers =
    pypi
    testpypi

[pypi]
repository = https://upload.pypi.org/legacy/
username = __token__
password = pypi-YOUR_PYPI_TOKEN_HERE

[testpypi]
repository = https://test.pypi.org/legacy/
username = __token__
password = pypi-YOUR_TESTPYPI_TOKEN_HERE

chmod 600 ~/.pypirc    # keep token private

4. Install publishing tools

pip install build twine

---

Build the Package

# Clean previous builds first
rm -rf dist/ build/ *.egg-info

# Build both sdist (.tar.gz) and wheel (.whl)
python -m build

Expected output:

Successfully built omnidoc_excel_extractor-0.1.0.tar.gz
            and omnidoc_excel_extractor-0.1.0-py3-none-any.whl

Validate the build before uploading:

twine check dist/*

You should see PASSED for both files. Fix any warnings before uploading.

---

Publish to TestPyPI

Always publish to TestPyPI first to catch packaging issues.

twine upload --repository testpypi dist/*

You will be prompted for credentials if ~/.pypirc is not configured.

Verify the TestPyPI install:

# Use a fresh virtual environment
python3 -m venv /tmp/test-install-venv
source /tmp/test-install-venv/bin/activate

# Install from TestPyPI (note: dependencies come from real PyPI)
pip install --index-url https://test.pypi.org/simple/ \
            --extra-index-url https://pypi.org/simple/ \
            omnidoc-excel-extractor

# Quick smoke test
python -c "
from omnidoc_excel_extractor import ExcelExtractor, CHUNK_REGISTRY
print(f'Package imported OK. {len(CHUNK_REGISTRY)} chunk types registered.')
"

Expected: Package imported OK. 57 chunk types registered.

---

Publish to PyPI (Production)

Once TestPyPI install is confirmed working:

twine upload dist/*

Or explicitly specifying the repository:

twine upload --repository pypi dist/*

Verify the production install:

pip install omnidoc-excel-extractor
python -c "from omnidoc_excel_extractor import ExcelExtractor; print('OK')"

Your package will be live at:

https://pypi.org/project/omnidoc-excel-extractor/

---

Versioning

Version is declared in pyproject.toml:

[project]
version = "0.1.0"

Follow Semantic Versioning:

Change	Version bump	Example
Bug fix, no API change	Patch	0.1.0 → 0.1.1
New chunk type or feature, backward-compatible	Minor	0.1.0 → 0.2.0
Breaking change to chunk fields or API	Major	0.1.0 → 1.0.0

Release workflow:

# 1. Bump version in pyproject.toml
#    version = "0.2.0"

# 2. Update CHANGELOG (optional but recommended)

# 3. Commit and tag
git add pyproject.toml
git commit -m "chore: bump version to 0.2.0"
git tag v0.2.0
git push origin master --tags

# 4. Build
rm -rf dist/
python -m build

# 5. Validate
twine check dist/*

# 6. TestPyPI first
twine upload --repository testpypi dist/*

# 7. Verify, then ship
twine upload dist/*

---

License

MIT — see LICENSE

---

Built with openpyxl, xlrd, pyxlsb, oletools, pydantic, networkx, and pandas.