iwork is a Rust crate for opening Apple iWork packages (.numbers, .pages, and .key), inspecting a small set of stable metadata, and extracting semantic content from supported document types.

Numbers (.numbers)

• sheet names and table membership (Sheet, TableModel)
• cell values: text, rich text, numbers (decimal128), dates, booleans, durations, errors, cached formula results with formula ids, currency, percentages
• table geometry (row/column counts, header rows/columns)
• write support: build minimal .numbers files from scalar cell data

Pages (.pages)

• document template name
• all text fragments from the body (TSWP storage, paragraphs split and cleaned)
• image alt-text (media descriptions)

Keynote (.key)

• presentation theme name
• per-slide: title, text fragments, image alt-text, template vs. content distinction

All formats

• document UUID and file-format version from Metadata/Properties.plist
• document kind detection from file extension

Add the crate to your project, then open a document and inspect it:

use iwork::Document;

fn main() -> Result<(), iwork::Error> {
    let document = Document::open("examples/numbers/personal_budget.numbers")?;
    let report = document.inspect("personal_budget.numbers")?;

    println!("kind: {}", report.kind.as_str());
    println!(
        "file format version: {}",
        report.properties.file_format_version.as_deref().unwrap_or("<unknown>")
    );

    Ok(())
}

List a Numbers document's tables with their names, geometry, and cells:

use iwork::numbers;

fn main() -> Result<(), iwork::Error> {
    let document = numbers::Document::open("examples/numbers/my_stocks.numbers")?;

    for (model, table) in document.spreadsheet()?.decoded_tables() {
        println!(
            "{} — {}x{}, {} rows decoded",
            model.name().unwrap_or("(unnamed)"),
            model.row_count(),
            model.column_count(),
            table.rows().len(),
        );
    }

    Ok(())
}

List a Numbers document's sheets and the table models they contain:

use iwork::numbers;

fn main() -> Result<(), iwork::Error> {
    let spreadsheet = numbers::Document::open("examples/numbers/my_stocks.numbers")?
        .spreadsheet()?;

    for sheet in spreadsheet.sheets() {
        println!(
            "{}: {:?}",
            sheet.name().unwrap_or("(unnamed)"),
            sheet.table_model_ids()
        );
    }

    Ok(())
}

Read cell values from a Numbers spreadsheet:

use iwork::numbers::{self, CellValue};

fn main() -> Result<(), iwork::Error> {
    let document = numbers::Document::open("examples/numbers/my_stocks.numbers")?;

    for table in document.spreadsheet()?.tables() {
        for row in table.rows() {
            for cell in &row.cells {
                match cell {
                    CellValue::Text(s) => println!("text:   {s}"),
                    CellValue::Number(n) => println!("number: {n}"),
                    CellValue::Date(secs) => println!("date:   {secs} s since 2001-01-01"),
                    CellValue::Empty => {}
                    _ => {}
                }
            }
        }
    }

    Ok(())
}

Extract UTF-8 string fields from a Pages document:

use iwork::pages;

fn main() -> Result<(), iwork::Error> {
    let document = pages::Document::open("examples/pages/term_paper.pages")?;
    let content = document.document()?;

    println!("title: {:?}", content.title());     // None until title fields are structurally decoded
    println!("headings: {:?}", content.headings()); // empty until heading fields are structurally decoded
    println!("first fragments: {:?}", &content.text_fragments()[..3.min(content.text_fragments().len())]);
    Ok(())
}

Extract UTF-8 string fields from a Keynote deck:

use iwork::keynote;

fn main() -> Result<(), iwork::Error> {
    let document = keynote::Document::open("examples/keynote/blueprint.key")?;
    let presentation = document.presentation()?;

    for slide in presentation.slides() {
        println!("{:?} {:?}", slide.layout_name(), slide.title());
        println!("text: {:?}", slide.text_fragments());
        println!("media: {:?}", slide.media_descriptions());
    }

    Ok(())
}

The Numbers API surface:

• Spreadsheet::sheets() — sheet names and which tables each sheet contains
• Spreadsheet::table_models() — authoritative table list with name and geometry (row/column counts, header counts)
• Spreadsheet::decoded_tables() — one (TableModel, Table) per real table; cells resolved per-table so string keys never collide across tables
• Spreadsheet::formula_records() — formula record objects; field 2 matches formula ids stored on some cached formula-result cells
• Spreadsheet::header_storage_bucket() — row- and column-indexed header storage referenced by each table model

Writing Numbers files:

• numbers::Workbook and numbers::WritableTable assemble table rows from scalar cell values
• Workbook::to_numbers_bytes() and Workbook::save_numbers() produce a .numbers package

The generated package round-trips through this crate. Full Apple Numbers compatibility still requires the complete document object graph, view state, styles, and calculation-engine references.

Example tools help map those missing structures without turning fixture content into parser heuristics. They decode the IWA/Snappy framing and hand the raw protobuf payloads to the protorev workbench rather than re-implementing wire decoding:

# IWA framing + object stream overview, and a structural diff between packages
cargo run -p iwork-workbench --bin dump_iwa_graph -- path/to/document.numbers
cargo run -p iwork-workbench --bin diff_iwa_graph -- left.numbers right.numbers
cargo run -p iwork-workbench --bin inspect_numbers -- path/to/document.numbers

# Schema-infer one object type across packages (protorev schema/explain/values/diff/experiments)
cargo run -p iwork-workbench --bin iwa_corpus -- schema       6001 examples/numbers/*.numbers
cargo run -p iwork-workbench --bin iwa_corpus -- diff         6001 before.numbers after.numbers
cargo run -p iwork-workbench --bin iwa_corpus -- experiments  7    layout_investigation.protorev

# Explore the cross-object reference graph
cargo run -p iwork-workbench --bin iwa_refs -- edges 6001 examples/numbers/personal_budget.numbers

dump_iwa_graph/diff_iwa_graph work at the IWA framing and object-graph level (entries, descriptor refs, object-type counts) with a protorev corpus shape per archive. iwa_corpus gathers every object of one message type and runs protorev's confidence-gated schema inference, per-field explanations, value sampling, and corpus diffing on it. iwa_refs recovers the cross-object reference graph structurally (which object types reference which). Together they let reverse engineering focus on controlled one-edit deltas, such as adding one table, renaming one sheet, or changing one cell style.

Known date encoding: date cell values are f64 seconds since the Cocoa epoch (2001-01-01T00:00:00Z). Decimal values may use IEEE 754-2008 decimal128 and are converted to f64.

What is decoded today:

• Body::template_name() — iWork template identifier (e.g. "04B_Term_Paper")
• Body::text_fragments() — all paragraph text in document order; paragraphs split on \n, control-character block markers and object-replacement characters filtered
• Body::media_descriptions() — alt-text for each image

Known gaps:

• Body::title() returns None — paragraph-style title classification not yet implemented
• Body::headings() returns empty — heading style classification not yet implemented
• page/section structure and inline tables are not yet modeled

What is decoded today:

• Presentation::theme_name() — theme name (e.g. "Blueprint")
• Slide::is_template() — distinguishes layout masters from real slides
• Slide::title() — slide title from the title placeholder
• Slide::text_fragments() — all text on the slide in archive order
• Slide::media_descriptions() — alt-text for each image on the slide

Known gaps:

• Slide::layout_name() returns None — not yet located in the format
• Slide::speaker_notes() not yet implemented (pattern confirmed, ready to implement)
• slide ordering is by archive path, not a full slide-graph traversal
• animations, transitions, and exact object placement are not modeled

The codebase depends on a small set of reverse-engineered format assumptions. Those notes live in docs/file-format.md, with the most important details also documented directly in src/package.rs and src/plist.rs.

• src/: library implementation and the small inspection CLI
• examples/: sample .numbers, .pages, and .key fixtures
• tests/: fixture-driven integration coverage
• docs/: format notes and implementation-oriented documentation

MIT. See LICENSE.