Jamailar/Weave-Office

What it is

Weave Office bundles four office editors — documents, spreadsheets, presentations and PDF — plus a Markdown/code editor and a work-in-progress HTML rich-text editor into a single Rust application. Files of different kinds sit in one row of tabs, and caret position, selection, scroll, zoom, undo history and unsaved changes persist when switching away and back. Shared host systems handle open, save, save as, recent files, autosave, printing and dialogs, and fonts are loaded once for all editors.

The second pillar is the command line: the same binary works as weave-office <verb>, reads and writes real office files through the same engines, and returns JSON with --json. That makes it usable by AI agents and scripts without opening a window.

Getting started

The GUI build currently targets macOS and needs Rust 1.95.0, which is pinned in rust-toolchain.toml, and Python 3. Large build inputs such as fonts and translations are expected on an external volume pointed to by WEAVE_DEV_VOLUME:

export WEAVE_DEV_VOLUME=/Volumes/MyDrive

python3 scripts/prepare-fonts.py
python3 scripts/prepare-document-fonts.py
python3 scripts/prepare-i18n.py
python3 scripts/cargo.py run --locked --offline

Build output goes to $WEAVE_DEV_VOLUME/DevCache/WeaveOffice/target. If you only want to try the CLI, font preparation can be skipped:

cargo run -p weave-cli --example weave -- help
cargo test -p weave-cli

Once running, a first CLI workflow looks like this: create a spreadsheet, set a cell, then export a document to PDF.

weave-office create quote.xlsx
weave-office set quote.xlsx /Sheet1/B2 value=1280
weave-office export contract.docx contract.pdf

Addresses are paths such as /body/p[3], /Sheet1/A1:C10, /slide[2]/shape[1] and /page[4]. The full command reference is in the CLI guide, and the agent-oriented guide ships as AGENT.md. For a macOS .app, see the packaging notes.

Formats and editors

• Documents open .docx, .docm, .doc, .odt and .rtf, and new files save as .docx. The WordCraft engine covers formatting, styles, tables, images, headers/footers, page breaks, WMF/EMF pictures, PDF export and printing.
• Spreadsheets open .xlsx, .xlsm, .xls, .ods and .csv, and save as .xlsx. GridCraft handles multiple sheets, formulas, number formats, borders, merged cells and PDF export with embedded CJK fonts.
• Presentations open .pptx, .pptm, .ppt and .odp, saving as .pptx, with slideshow and export.
• PDF is an editor rather than only a viewer: existing text and images can be edited, annotations added, and pages inserted, deleted, reordered, rotated, cropped, extracted or split. Details are in the PDF capabilities doc.
• Markdown has Obsidian-style live preview, task checkboxes, [[wikilinks]], backlinks, tags, outline, a command palette (⇧⌘P) and quick open (⌥⌘O).
• Code and plain text covers 20+ languages; files are written back in their original encoding and line endings, and unchanged files are saved byte for byte. Rich text (HTML) is WYSIWYG but still in development.

Constraints

The project describes itself as under active development and explicitly not yet a day-to-day replacement for a mature office suite; fidelity on complex documents, input methods and some advanced workflows are listed as still being polished, with current status in the acceptance log. The platform badge and packaging notes point to macOS, so other desktop platforms are not documented here. Building the GUI requires the external-volume font and translation assets and the preparation scripts; only the CLI path avoids that. Most project documentation is currently in Chinese. Weave Office is MIT-licensed; the four office engines are vendored pinned snapshots of WordCraft, GridCraft, DeckCraft and PdfCraft under their own MIT OR Apache-2.0 terms.

HotRepo's take

The repository was recorded on the Rust new-project list on 4 November 2026, with 109 stars and a reported 9-star gain over a roughly two-hour window starting 05:45 UTC; it has no recorded GitHub daily rank. That is a small, early sample, so it says little about adoption. What does stand out in the materials is the architecture: a single host with vendored engine snapshots and a CLI that addresses document content by path and returns structured JSON, which is a more concrete AI-agent story than a plain "AI office" label. The honest development-status note and the published acceptance log also make it easier to judge current limits.