dream-num/univer

The problem

If you need spreadsheet or document editing inside a SaaS product, internal tool, or an agent that must produce real Office files, the options are awkward: host Google Sheets or Microsoft 365 and fight the iframe, permissions, and terms; or adopt a viewer-only library that cannot edit. Univer is an SDK that gives you spreadsheet, document, slide, base and PDF surfaces rendered on canvas, with the same APIs available in the browser and in Node.js for headless work.

What's different

Compared with hosted office suites, Univer is not a destination app but a framework: you compose plugins, replace UI, and keep data in your own backend. Compared with file-viewer libraries, it is editable and includes a formula engine and a rendering layer built for large surfaces. Two things genuinely stand out: a plugin-first architecture where every capability can be added, removed or lazy-loaded, and an isomorphic runtime so the same workbook logic can run server-side for agent or batch workflows. The README frames it as an "Office Harness for AI Agents", and the companion projects (CLI, workspace, MCP-style integrations) follow that line.

Who it's for

Product and platform teams building a collaboration tool, BI surface, or document editor into their own app, and teams building agent pipelines that must read and write real Office content. It fits teams that want control over UI and storage and are willing to integrate deep. It is a poor fit if you just want a working online spreadsheet today, a simple preview widget, or if you need full Google-Docs parity out of the box; the README is explicit that capability coverage varies and some collaboration features are licensed separately.

Cost of adoption

TypeScript SDK with a large multi-package surface; the plugin-mode install alone pulls a dozen @univerjs/* packages, and the README itself distinguishes plugin mode from preset mode, so expect meaningful bundling and integration work. Running it on the server requires building and deploying the headless runtime. The repo is Apache-2.0, but the README warns that some AI and collaboration capabilities have separate licensing, so check the capability matrix before committing. There is no short install command in the visible README; start from the docs.

Verdict

Univer is not new: created in September 2022, it has accumulated roughly 22.7k stars and 1.9k forks, and it drew Hacker News attention back in 2024, so this is an established project rather than a spike. The recent signal is a monthly trending appearance with about 6.8k stars gained, which looks like renewed interest around the AI-agent positioning more than a sudden launch. That momentum is real but the README is a marketing surface: licensing for AI and collaboration features is vague, so budget evaluation time. Worth a serious look if you need an embeddable Office runtime; skip it if you want a finished product.