genspark-ai/genoffice

GenOffice – Open‑source AI‑enhanced Office suite

What it is – GenOffice is a free, cross‑platform (macOS, Windows, Linux) office suite that can open, edit, and save native Microsoft Office files (.docx, .xlsx, .pptx) as well as PDFs and Markdown. The distinguishing claim is that AI‑driven editing is built into each editor rather than being tacked on as a chat window.

Core apps

App	File type	Main tech
Docs	Word (.docx)	In‑house docx engine that parses OOXML, tracks dirty blocks, and writes only the changed paragraphs back – preserving every untouched byte so Word never notices layout changes.
Sheets	Excel (.xlsx)	UI built on the open‑source Univer core; a Rust side‑car (calamine + IronCalc) handles import/export, charts, pivot tables, slicers, and formula tracing.
Slides	PowerPoint (.pptx)	Custom .pptx parser/render engine with masters, layouts, smart guides, and non‑destructive image cropping.
PDF	PDF	Viewer/editor based on pdf.js + pdf-lib; true text‑and‑image editing rewrites page content streams via PDFium compiled to WebAssembly, preserving original fonts. Also provides local PDF → DOCX / PPTX / XLSX conversion (including OCR on macOS/Windows).
Markdown	.md	Tiptap/ProseMirror block editor that reads/writes plain Markdown files.
Shell	Suite launcher	Tabbed host for the five editors, theme handling, auto‑updates.

AI capabilities

• Block‑level edit suggestions with snapshot/diff history.
• Document‑aware agents that can call tools (web/image search, image generation, media analysis, transcription).
• Bring‑Your‑Own‑Key (BYOK) support for dozens of hosted models (Claude, OpenAI, Gemini, DeepSeek, etc.) or any OpenAI‑compatible endpoint, plus a keyless Genspark sign‑in that proxies calls through the Genspark service.
• Agents can be extended via the packages/ai-search module.

Installation

• Pre‑built installers for macOS (Intel & Apple Silicon), Windows (x64), and Linux (deb, rpm, AppImage). All builds are signed and come from the main branch.
• Linux packages install via apt or dnf/zypper; the AppImage requires FUSE 2.
• Development workflow uses a monorepo with npm workspaces; the Sheets side‑car needs a Rust toolchain (cargo).

Architecture highlights

• Electron provides the desktop runtime for every app.
• Shared TypeScript packages (packages/*) contain pure‑logic engines (docx, pptx, pdf2docx, agent core, AI provider abstraction, UI kit, i18n, etc.) that are unit‑tested and have no Electron dependency.
• Document round‑trip strategy: the original file is kept as a zip archive; only edited blocks are regenerated and spliced back, leaving the rest byte‑for‑byte unchanged.
• PDF conversion uses PDFium (WASM) for character‑level extraction and geometry‑based layout analysis, feeding the same docx engine to produce editable Office files.

Privacy & security

• The app works fully offline for document handling; AI features require network access.
• Optional usage analytics are opt‑out and never transmit document content or personal identifiers.
• Security notes (renderer sandboxing, IPC validation) are documented in SECURITY.md.

License

• Apache‑2.0 for the whole codebase, except the ee/ directory which is reserved for future enterprise modules under a separate commercial license.

Who might use it

• Users who need a privacy‑preserving Office alternative with AI assistance.
• Developers interested in extending an open‑source AI‑augmented document pipeline.
• Organizations looking for a cross‑platform suite that can edit Office files without losing fidelity and that can run AI models locally or via their own API keys.

---

All information above is taken directly from the repository’s README; no additional features have been inferred.