Create Documents from IR
create_from_ir writes a brand-new DOCX, XLSX, or PPTX from a DocumentIR — the same structured schema you get out of to_ir(). One IR, three target formats. The IR is the natural sink for LLM-generated content: produce structured JSON, then materialize the right Office format on demand.
Build and write
Rust
use office_oxide::{DocumentFormat};
use office_oxide::create::create_from_ir;
use office_oxide::ir::{DocumentIR, Section, Element, Run};
let ir = DocumentIR {
sections: vec![Section {
title: Some("Quarterly Report".into()),
elements: vec![
Element::Heading { level: 1, text: "Highlights".into() },
Element::Paragraph { runs: vec![Run::plain("Revenue grew 18%.")] },
Element::Table {
rows: vec![
vec!["Region".into(), "Revenue".into()],
vec!["NA".into(), "$1.2M".into()],
vec!["EU".into(), "$820K".into()],
],
},
],
}],
};
create_from_ir(&ir, DocumentFormat::Docx, "report.docx")?;
Switch the target format and you get an XLSX or PPTX from the same IR:
create_from_ir(&ir, DocumentFormat::Xlsx, "report.xlsx")?;
create_from_ir(&ir, DocumentFormat::Pptx, "report.pptx")?;
Python
from office_oxide import create_from_ir
ir = {
"sections": [{
"title": "Quarterly Report",
"elements": [
{"kind": "Heading", "level": 1, "text": "Highlights"},
{"kind": "Paragraph", "runs": [{"text": "Revenue grew 18%."}]},
{"kind": "Table", "rows": [
["Region", "Revenue"],
["NA", "$1.2M"],
["EU", "$820K"],
]},
],
}],
}
create_from_ir(ir, "docx", "report.docx")
create_from_ir(ir, "xlsx", "report.xlsx")
create_from_ir(ir, "pptx", "report.pptx")
JavaScript
import { createFromIr } from 'office-oxide';
const ir = {
sections: [{
title: 'Quarterly Report',
elements: [
{ kind: 'Heading', level: 1, text: 'Highlights' },
{ kind: 'Paragraph', runs: [{ text: 'Revenue grew 18%.' }] },
{ kind: 'Table', rows: [
['Region', 'Revenue'],
['NA', '$1.2M'],
['EU', '$820K'],
]},
],
}],
};
createFromIr(ir, 'docx', 'report.docx');
createFromIr(ir, 'xlsx', 'report.xlsx');
createFromIr(ir, 'pptx', 'report.pptx');
Go
Go accepts the IR as a JSON payload:
import (
"encoding/json"
officeoxide "github.com/yfedoseev/office_oxide/go"
)
ir := map[string]any{
"sections": []any{
map[string]any{
"title": "Quarterly Report",
"elements": []any{
map[string]any{"kind": "Heading", "level": 1, "text": "Highlights"},
map[string]any{"kind": "Paragraph", "runs": []any{map[string]any{"text": "Revenue grew 18%."}}},
map[string]any{"kind": "Table", "rows": [][]string{
{"Region", "Revenue"},
{"NA", "$1.2M"},
{"EU", "$820K"},
}},
},
},
},
}
payload, _ := json.Marshal(ir)
officeoxide.CreateFromIR(string(payload), "docx", "report.docx")
officeoxide.CreateFromIR(string(payload), "xlsx", "report.xlsx")
officeoxide.CreateFromIR(string(payload), "pptx", "report.pptx")
C#
using OfficeOxide;
using System.Text.Json;
var ir = new {
sections = new[] {
new {
title = "Quarterly Report",
elements = new object[] {
new { kind = "Heading", level = 1, text = "Highlights" },
new { kind = "Paragraph", runs = new[] { new { text = "Revenue grew 18%." } } },
new { kind = "Table", rows = new[] {
new[] { "Region", "Revenue" },
new[] { "NA", "$1.2M" },
new[] { "EU", "$820K" },
}},
},
},
},
};
var payload = JsonSerializer.Serialize(ir);
OfficeOxide.CreateFromIr(payload, "docx", "report.docx");
OfficeOxide.CreateFromIr(payload, "xlsx", "report.xlsx");
OfficeOxide.CreateFromIr(payload, "pptx", "report.pptx");
How each target format renders the IR
| IR element | DOCX | XLSX | PPTX |
|---|---|---|---|
Section |
Word section break | Worksheet (one section → one sheet) | Slide |
Heading { level } |
Paragraph w/ Heading{level} style |
Bold cell at row top | Slide title placeholder |
Paragraph |
Body paragraph | Cells in successive rows | Body text placeholder |
List { ordered, items } |
Numbered or bulleted list | Cells in column A | Body bullets |
Table { rows } |
Word table | Cells starting at the next free row | PowerPoint table |
Image { filename, data } |
Inline image | (skipped — XLSX needs anchor coords) | Slide image |
The mapping is conservative: every IR element produces deterministic output, but each format has features the IR doesn’t model (XLSX merges, PPTX animations, DOCX comment threads). For richer output, use the format-specific builders.
Round-trip — read, modify, write
Combine to_ir() with create_from_ir to do structural edits that survive across formats:
from office_oxide import Document, create_from_ir
with Document.open("legacy.doc") as doc:
ir = doc.to_ir()
# Append a new section
ir["sections"].append({
"title": "Appendix",
"elements": [
{"kind": "Heading", "level": 2, "text": "Updated terms"},
{"kind": "Paragraph", "runs": [{"text": "Effective 2026-04-19."}]},
],
})
create_from_ir(ir, "docx", "modernized.docx")
This is also the cleanest way to do legacy → OOXML conversion (DOC → DOCX, XLS → XLSX, PPT → PPTX). See Conversion: legacy → OOXML.
Why use the IR for creation
- Format-agnostic templates. Generate the same content as DOCX (for Word users), XLSX (for analysts), or PPTX (for presentations) with no rewriting.
- LLM-native. Have the model produce JSON matching the IR schema; materialize the file. Far more reliable than asking it to emit Office XML directly.
- Diffable. IR is JSON; commit it. The Office file becomes a build artifact.
Limitations
- Images in
Element::Imageare written when the target format supports inline images (DOCX, PPTX). XLSX requires anchor coordinates, which the simple IR doesn’t carry — skip or use the format-specific builder. - Cell formatting (number formats, currency, bold) isn’t part of the IR. Add it via the per-format editor after creation.
- Charts, smart art, and embedded objects aren’t part of the IR.
For everything beyond the IR’s expressive range, drop into format-specific builders (docx::create::DocxBuilder, xlsx::create::XlsxBuilder, pptx::create::PptxBuilder). They give you full access to OOXML features at the cost of being format-specific.
See also
- Structured IR — the schema in detail
- Conversion: legacy → OOXML — uses
create_from_irunder the hood - Editing overview — modify existing files in place