TL;DR: GenOffice is an Apache 2.0 office suite (Docs, Sheets, Slides, PDF, Markdown, HTML) from Genspark that treats AI editing as the core workflow and can point at any OpenAI-compatible endpoint — including a local Ollama server. Document editing is fully local; the AI layer needs a Genspark account or your own backend. It is alpha software (v0.11.0, September 28 2026) and moves fast.

	GenOffice	LibreOffice	OnlyOffice
Best for	AI-first editing on your own model backend	Maximum maturity, zero AI dependencies	Best .docx/.xlsx fidelity in a mature FOSS suite
License	Apache 2.0 (ee/ dir reserved for enterprise)	MPL 2.0	AGPL v3
AI assistant	Built in: Genspark cloud, 15+ BYOK providers, or any OpenAI-compatible local endpoint	None native; third-party extensions with API keys	Plugin-based, cloud API keys
Maturity	Alpha, launched Aug 3 2026	25+ years	10+ years

Honest take: If you want AI inside your documents without a subscription, GenOffice pointed at a local Ollama server is the only suite that does it natively today — but keep LibreOffice installed for anything you can’t afford to have an alpha mangle.

What is GenOffice and what license does it use?

GenOffice is Apache 2.0, with one carve-out: the repository’s ee/ directory is reserved for future enterprise modules under a separate GenOffice Enterprise License. Everything shipping today — all six editors — is under the plain Apache 2.0 license, which permits commercial use, modification, and redistribution with no MAU caps or attribution beyond the standard notice. That makes it one of the very few genuinely permissive AI-integrated office suites, and the license check that usually kills these projects passes here.

The project comes from Genspark (Mainfunc), open-sourced August 3 2026 at github.com/genspark-ai/genoffice. Press coverage repeated the launch story — an alpha reportedly built in 7 days by a single engineer for about $10,000 in LLM API costs — which is a fun anecdote but also a fair warning about how young the codebase is. As of September 30 2026 the repo sits at roughly 8.2k GitHub stars, and the release cadence is aggressive: v0.10.1038 on September 22, v0.10.1467 on September 27, v0.11.0 on September 28 2026.

What you actually get is six Electron apps sharing one engine layer: Docs (.docx), Sheets (.xlsx), Slides (.pptx), plus PDF, Markdown, and HTML editors, wrapped in a tabbed shell with local file search. The engine is TypeScript with a Rust sidecar for xlsx operations, and the spreadsheet builds on the open-source Univer core, according to the project’s own materials. Installers ship for macOS (arm64/x64), Windows (x64/ARM64), and Linux (deb, rpm, AppImage), so “Windows and Mac” in the launch headline undersells it — Linux is a first-class target.

Can GenOffice run its AI on a local model with Ollama?

Yes — GenOffice accepts any OpenAI-compatible endpoint (base URL + key), which explicitly includes local model servers, so an Ollama instance on the same machine works as the AI backend. This is the detail that matters for the aifoss audience, because the default path does not run offline: the keyless “sign in with Genspark” option routes requests through Genspark’s cloud proxy (Claude, GPT, and Gemini model families), and the BYOK list (OpenAI, Anthropic, Gemini, DeepSeek, Kimi, GLM, Qwen, Mistral, OpenRouter, and others) is all cloud APIs. Local is a third, manual option.

The setup: run Ollama, then point GenOffice’s custom endpoint at Ollama’s OpenAI-compatible API.

$ ollama pull qwen3
$ curl http://localhost:11434/v1/models
{"object":"list","data":[{"id":"qwen3:latest","object":"model", ...}]}

In GenOffice’s model settings, set the base URL to http://localhost:11434/v1 and the model name to whatever tag ollama list shows.

One snag you will hit: like most OpenAI-compatible clients, GenOffice expects a non-empty API key, while Ollama ignores authentication entirely. The fix is to enter any placeholder string (ollama works) in the key field. If the assistant returns connection errors instead, the usual cause is Ollama binding only to localhost while GenOffice runs sandboxed or on another machine — start the server with OLLAMA_HOST=0.0.0.0 ollama serve and use the LAN IP in the base URL.

Two boundaries to be clear about. First, document editing is fully local regardless of AI configuration — files are opened, edited, and saved on your machine and never uploaded. Second, some AI features are inherently online: web search and image generation go through network services even in BYOK mode. A local Ollama backend covers the text work — rewriting, summarization, in-cell agents — which is most of what an office AI does all day.

How well does GenOffice handle .docx, .xlsx, and .pptx files?

GenOffice opens and saves native .docx, .xlsx, and .pptx files, and its README makes an unusual engineering promise: editing is byte-preserving, meaning parts of the file you didn’t touch are written back byte-for-byte. If that holds up, it is a genuinely better compatibility story than round-tripping a document through a conversion layer — the classic way office alternatives corrupt tracked changes, embedded objects, and exotic formatting.

The caveat is the word “alpha.” The v0.10.1467 release (September 27 2026) was billed as a major Word-parity update, adding spelling suggestions, AutoCorrect, Find & Replace with regex, a navigation pane, and rich header/footer editing — table-stakes features that were missing three weeks earlier. v0.11.0 fixed multi-section documents pinning a CPU core. That is the trajectory of a project sprinting toward parity, not one that has reached it. For clean, mainstream documents the format support is reportedly solid; for a 200-page contract with nested tracked changes, verify the round-trip yourself before trusting it, and keep the original.

There’s also a notable extra for developers: recent releases added MCP (Model Context Protocol) support, so coding agents can drive GenOffice headless — create, edit, and audit documents from a CLI. If you run agentic coding tools, that turns the suite into a scriptable document backend; see our sister site’s coding agent comparison for the client side of that equation.

How does GenOffice compare to LibreOffice and OnlyOffice?

GenOffice wins on exactly one axis — native, configurable AI — and loses on most others, for now. That is the honest comparison:

	GenOffice v0.11.0	LibreOffice 25.x	OnlyOffice 9.x
Native AI with local model support	Yes — OpenAI-compatible base URL	No — third-party extensions	Plugin, cloud keys
MS format fidelity	Byte-preserving claim, alpha track record	Good, occasional layout drift	Strongest among FOSS suites
Stability	Alpha, weekly breaking pace	Very high	High
Footprint	Six Electron apps	Native toolkit	Native/web hybrid
Telemetry	On by default, can disable	None	None in desktop editors

LibreOffice (MPL 2.0) remains the default recommendation for anyone whose priority is opening whatever file lands in their inbox. Its AI story, however, is genuinely weak: integrations are third-party extensions that mostly assume an OpenAI API key, with none of them woven into the editing model. OnlyOffice (AGPL v3) has the best .docx fidelity of the established FOSS suites but the same bolted-on AI posture.

GenOffice’s differentiator is architectural: the AI operates on document state — block-level rewrites in Docs, agents that build models inside Sheets cells, deck generation in Slides — rather than living in a sidebar chatbot that pastes text at your cursor. Whether that design survives contact with real documents is the alpha-stage question, but it is the right design, and no incumbent FOSS suite has it.

What hardware do you need to run GenOffice with local AI?

The suite itself runs on any recent machine — the real hardware question is the local model behind it. Electron apps are memory-hungry in aggregate, but a modern laptop handles the editors fine. Budget your hardware for the Ollama backend instead:

• 7B-8B class models (fine for rewriting, summarizing, spreadsheet formulas): 5-6GB VRAM at Q4. A used RTX 3060 12GB ($260-$296, September 2026) is the cheapest comfortable entry.
• CPU-only works for 7B models at usable-but-not-snappy speeds — the constraint is RAM, and DDR5 is brutal right now: a 32GB kit runs $399-$479 as of September 2026, roughly 4x its mid-2025 price. If you already have 32GB, CPU inference costs you nothing to try.
• Larger models (27B-35B class) want a 24GB card; that decision is bigger than this article, and our full breakdown of what self-hosting actually costs is in FOSS AI vs SaaS AI: the real cost math, with hardware picks at runaihome.com.

If you’d rather not run a server process at all, LM Studio also exposes an OpenAI-compatible endpoint that GenOffice can target — though LM Studio itself is freeware, not open source.

When should you NOT use GenOffice?

Skip GenOffice, at least for now, in four situations:

1. Documents you cannot afford to break. This is alpha software shipping near-daily releases. The byte-preserving design limits the blast radius, but “limits” is not “eliminates.” Legal documents, theses, and anything with heavy tracked changes belong in LibreOffice or OnlyOffice until GenOffice has a longer track record.
2. Strict-privacy environments without configuration effort. Official builds send usage analytics by default (the project states document content, file names, paths, and identity are never sent, and reporting can be disabled in Settings → General). The convenient keyless AI mode routes through Genspark’s cloud. Air-gapped or policy-bound users must disable analytics and configure a local endpoint — both possible, neither the default.
3. You need macros or deep legacy compatibility. There is no VBA story here, and organizations built on spreadsheet macros will hit that wall immediately.
4. You just want a stable office suite and don’t care about AI. Then GenOffice’s one differentiator is worthless to you, and LibreOffice is the boring, correct answer.

Verdict

GenOffice is the first office suite where “open source,” “native AI,” and “your own local model” are all true at once, and Apache 2.0 licensing means that stays true commercially. As a daily driver it is not there yet — the alpha label is accurate, and the format-fidelity claims need months of real-world round-trips before they’re proven. As a second suite, pointed at an Ollama server for AI-heavy drafting and spreadsheet work while LibreOffice handles the mission-critical files, it is already worth the install. Watch the release page; at this development pace, the alpha caveats in this review have a short shelf life.

FAQ

Does GenOffice work completely offline? Document editing, saving, and local file search work fully offline. AI text features work offline only if you configure a local OpenAI-compatible backend such as Ollama at http://localhost:11434/v1. Web search and image generation always require a network connection.

Is GenOffice really free for commercial use? Yes. The suite is Apache 2.0 with no paid tier for the apps themselves. The only exception is the ee/ directory, reserved for future enterprise modules under a separate license — nothing you install today depends on it.

What Microsoft Office formats can GenOffice open and save? Native .docx, .xlsx, and .pptx, plus PDF, Markdown, and HTML editing. The project claims byte-preserving writes — untouched parts of a file are written back byte-for-byte — but as alpha software, complex documents deserve a verification pass after round-tripping.

Sources

• GenOffice GitHub repository — license, AI backend options, format support, telemetry policy (accessed September 30 2026)
• GenOffice releases — v0.11.0 (Sep 28 2026), v0.10.1467 (Sep 27 2026), v0.10.1038 (Sep 22 2026)
• Genspark open-sources GenOffice — Open Source For You — launch date and background
• Genspark Released GenOffice — Thinkfree blog — architecture notes (TypeScript engine, Rust xlsx sidecar, Univer core)

What self-hosting actually costs

Real cost breakdowns for self-hosted AI: hardware floors, power, maintenance hours, and the honest comparison against paying for it. No spam, unsubscribe anytime.

Email address Subscribe