What Is GenOffice? Features, License, and Alternatives (2026)
GenOffice is an open-source, AI-native office suite from Genspark covering word processing, spreadsheets, presentations, PDF and Markdown. It runs on macOS, Windows and Linux under the Apache-2.0 licence. The repository went public on 31 July 2026 and had passed 2,300 stars and 400 forks within ten days.
The interesting part is not that someone built an office suite. It is what they chose to build in-house: a Rust sidecar for xlsx, their own pptx engine, and byte-preserving docx editing. Those are the parts that usually break when a tool round-trips your files.
This piece covers what each app actually does and what the Apache licence does and does not cover here. It closes with where the suite sits if your work starts as a dataset rather than a document.
What happened
Genspark published the genspark-ai/genoffice repository on 31 July 2026. Ten days later it showed 2,374 stars, 422 forks and 25 open issues, with commits landing daily. The current release is v0.6.13, and the macOS and Windows installers are signed.
Two of those numbers are worth pausing on. A 422-to-2,374 fork-to-star ratio is unusually high for a desktop application, and it says people are modifying the code rather than only bookmarking it. Twenty-five open issues on a ten-day-old project of this size suggests the maintainers are keeping up.
A small ecosystem has already formed around it. Forks include a build that swaps the AI provider for DeepSeek, a Nix packaging flake, and a variant driven by local coding agents. There is also an MCP server that exposes byte-preserving docx and pptx editing to other tools. That last one is a good signal: people are treating GenOffice as an engine, not just an app.
All figures here were read from the GitHub API on 10 August 2026. Star counts move; the repository is the source of truth.
What GenOffice actually is
Six Electron apps sharing one engine layer, with AI editing designed in rather than bolted on as a chat sidebar. The engine packages are pure TypeScript with no Electron dependency. That is why the fork turning them into an MCP server was possible at all.
| App | Format | What stands out |
|---|---|---|
| Docs | .docx |
Byte-preserving round trip — only edited paragraphs are regenerated, the rest of the file is kept byte-for-byte |
| Sheets | .xlsx |
Univer core plus in-house extensions; Rust sidecar for import and export; pivot tables, slicers, conditional formatting, formula tracing |
| Slides | .pptx |
In-house parse and render engine with masters, charts, cropping and HarfBuzz text shaping |
.pdf |
Real text editing with in-block reflow and original-font preservation, not cover-up annotations | |
| Markdown | .md |
Tiptap block editor saving back as plain Markdown, with no proprietary wrapper around the file |
| Shell | — | Tabbed host for the five editors, with themes and auto-update |
The docx approach deserves a note. The original file is archived by hash and never touched. Edited blocks become OOXML fragments spliced back into the original XML, and untouched blocks keep their original bytes. That is why Word does not notice a round trip. Sheets and slides follow the same philosophy.
The licence question: what open source covers here
The code is Apache-2.0. The AI is not self-hosted.
Per the project's own README, the apps sign in to a Genspark account through a device-code flow. No model API key is entered or stored by the user. Model calls route through the Genspark proxy across the Claude, GPT and Gemini families. The same account unlocks the search, image-generation and media-analysis tools the agents are built on.
So "free and open source" is accurate about the code and incomplete about the product. You can read, fork and self-build every editor. The AI features depend on someone else's account system and proxy. That is a reasonable trade for a free suite. It is also precisely why one of the first forks replaced the provider with a DeepSeek endpoint.
If your reason for wanting an open-source office suite is auditability of the document engines, GenOffice delivers. If it is avoiding a dependency on a vendor's inference service, read the AI backend section before you commit.
Who it is actually for
Three groups have an obvious reason to look at this.
Teams that need format fidelity without a Microsoft licence. The byte-preserving round trip is the differentiator. Most alternatives re-render a document on save, which is why a file that goes through them comes back subtly different.
Developers who want the engines rather than the apps. The docx, pptx and parsing packages are plain TypeScript and separately consumable, which is what the MCP fork demonstrated within days.
Anyone whose objection to cloud suites is where files live. Editing happens locally, though the AI layer does not — see the previous section.
Who it is not for, yet: teams that depend on real-time co-editing. There is no shared collaboration layer, and the project does not claim one.
What it means if your work starts with data
GenOffice is an editing environment. It assumes a document exists, or that you know what document should exist.
Reporting work usually starts earlier than that. There is an export, and three questions come before any document. Which of its forty columns matter. What the number becomes once you settle the grouping and the period. Which chart states that honestly. None of those are editing operations.
The pivot table engine is a good illustration. Having pivot tables in an open-source spreadsheet is genuinely useful, and it still requires you to know which field goes where. That prerequisite is what makes summarizing without pivot tables a common search in the first place.
Where a data agent still fits
The split is analysis versus production, and both halves are real work.
Powerdrill Bloom sits on the analysis half. You upload the Excel, CSV or PDF export and ask in natural language what the numbers say. The analysis and the chart arrive together, before any document exists. The canvas then exports to PowerPoint. Our Excel to PPT page covers that route.
These are complements rather than competitors. Work out what the deck should argue, then edit it wherever you like, including in a suite you can fork.
The order matters more than the tooling. A deck assembled before anyone has interrogated the data is a well-formatted guess, and no editor fixes that. A deck assembled after it is just production work, which is exactly what an office suite should be good at.
Alternatives worth knowing
LibreOffice. The mature open-source option, with far broader format coverage and decades of hardening. No AI layer, and no ambition to have one. If format breadth matters more than AI, this remains the safer choice, and it has a support ecosystem GenOffice has not had time to build.
Microsoft 365 with Copilot agent mode. Agent mode reaches Word, Excel and PowerPoint inside the Microsoft boundary. The right default if your organization already licenses it.
Agent-first tools. Claude Cowork and ChatGPT for PowerPoint come at the same problem from the agent side rather than the editor side. Both work across the apps you already run.
The short version
GenOffice is the most credible open-source AI office suite so far, and the engineering choices are the reason. Byte-preserving docx, an in-house pptx engine, a Rust xlsx sidecar. Ten days and 422 forks suggest other developers agree.
Worth repeating the caveat once more, because it is the thing most coverage skips. Read the AI backend section before treating it as fully self-hosted. Open code and open inference are different claims, and only the first one is being made here.
Read that section before treating it as fully self-hosted. And if your bottleneck is working out what the numbers say rather than producing the file, that step comes first. Try Powerdrill Bloom on the export, then take the answer into whichever editor you prefer.
Frequently asked questions
Is GenOffice free?
Yes. The code is published under the Apache-2.0 licence and the installers are free to download. The AI features require signing in to a Genspark account. The apps handle that through a device-code flow rather than by asking for a model API key.
What platforms does GenOffice run on?
macOS 11 and later, Windows 10 and later, and Linux on x86_64 with glibc 2.34 or newer. Linux ships as both a .deb and an AppImage. The macOS and Windows installers are signed.
Can GenOffice open and save Word files without breaking them?
That is its central design claim. The original file is archived and only edited paragraphs are regenerated, with untouched content kept byte-for-byte, so formatting survives a round trip. The same approach applies in Sheets and Slides.
Does GenOffice have pivot tables?
Yes. GenOffice Sheets is built on the Univer core with in-house extensions and includes pivot tables, slicers, conditional formatting and formula tracing. Import and export run through a Rust sidecar rather than a JavaScript parser.
Is GenOffice a replacement for Microsoft Office?
For document editing it is a serious open-source option, particularly if format fidelity matters. Coverage is narrower than Microsoft 365, there is no shared cloud collaboration layer, and the AI features depend on Genspark's service rather than your own.