Skip to content

v0.13.0

Choose a tag to compare

@github-actions github-actions released this 02 Nov 21:05
· 5 commits to main since this release

Adds getDocxBytes() for saving the current editor state without a browser download, repairs legacy Word and VML fidelity, and accepts DOC, DOT, DOCX, DOTX, DOCM, and DOTM consistently. The playground now recompiles WASM from deployed Rust sources and exposes verifiable build metadata. The main CI lockfile mismatch is fixed.

Install:

pnpm add @extend-ai/react-docx@0.13.0

Fresh benchmarks on merged main covered 1,329 distinct inputs against docx-rs 0.4.22, office_oxide 0.1.15, and undoc 0.27.1. Each cell is body-text F1 · median / p95 milliseconds; DOCX uses the separate 1,000-input corpus, and other formats pool both multi-format cohorts.

Format (inputs) docx-core docx-rs office_oxide undoc
DOCX (1,000) 99.983% · 0.800 / 6.641 99.740% · 1.212 / 9.593 99.917% · 0.816 / 5.793 99.988% · 1.204 / 13.712
DOC (217) 98.748% · 0.268 / 5.664 Unsupported 97.979% · 0.063 / 1.285 95.293% · 0.027 / 0.417
DOT (2) 100.000% · 1.008 / 1.564 Unsupported 99.902% · 0.339 / 0.486 100.000% · 0.131 / 0.191
DOTX (6) 100.000% · 0.565 / 7.232 99.756% · 0.589 / 3.724 99.813% · 0.357 / 1.970 99.911% · 0.461 / 3.886
DOCM (6) 100.000% · 0.180 / 2.463 100.000% · 0.257 / 1.823 100.000% · 0.142 / 1.031 100.000% · 0.083 / 2.569
DOTM (12) 100.000% · 0.546 / 1.076 100.000% · 1.640 / 2.587 100.000% · 0.618 / 1.716 100.000% · 0.724 / 3.439

Accuracy retains failures where references exist; timing uses common successful inputs. Native extraction scores do not establish visual fidelity, and template/macro cohorts are small. No accuracy metric or parsing status regressed after cleanup. Legacy field destinations, headers/footers, lists, and some table/text-box details still have gaps. Macros are never executed.

Full comparison, methodology, and remaining gaps · Supported formats and MIME types · Playground build provenance