GenOffice is open source. The AI inside it isn't.

Genspark open-sourced GenOffice on August 3 — a desktop office suite for macOS and Windows with docs, sheets, slides, and PDF, AI wired into each one, free and ad-free. The headline writes itself: somebody finally open-sourced Microsoft Office. The README says something more interesting, and it's the part worth ten minutes before you install it on a work machine.

What actually happened

GenOffice ships as native apps for Apple Silicon macOS and x64 Windows, currently at version 0.4.110. The four editors read and write the formats you already have: .docx, .xlsx, .pptx, and PDF. Genspark's Super Agent runs inside the documents rather than in a side panel.

The licensing is a dual model. Most of the code is under the Apache License 2.0; an ee/ directory is carved out under a separate GenOffice Enterprise License for enterprise features. That's a normal open-core arrangement, and it's stated plainly — but "fully open source" and "open core" are not the same sentence.

The line that actually matters is in the GitHub README: the apps sign in to a Genspark account and route model calls through the Genspark service side, and no model API key is stored locally. There's no BYOK path in the shipped build. The editor is yours. The intelligence is a hosted dependency with an account attached.

Why open-source AI tools still need a vendor review

We like this release. A local-first editor that speaks Office formats and doesn't demand a browser tab is a genuinely good thing, and the code being readable means the file handling can be audited by someone other than the vendor. That's more than most AI tools offer.

But route the question the way you'd route it for any SaaS: where does the document text go when the AI touches it? Under this architecture, the answer is Genspark's servers, under Genspark's terms, on Genspark's account. Open-sourcing the client doesn't change that — it just means you can read the code that sends it. If the document is a client contract, a payroll sheet, or an unannounced pricing model, the license file is not the control you need. The data processing terms are.

The practical read: treat GenOffice as a promising AI tool with a cloud dependency, not as the self-hosted escape hatch it sounds like. It's Alpha-tier by its own version number. Pilot it on material you'd be fine emailing. Keep the sensitive documents on whatever you already govern. And if the appeal was owning the whole stack, watch the repo for a local-model path — because that's the commit that would change the answer.

Sources: genspark-ai/genoffice on GitHub, Genspark — GenOffice.