Univer: The Unified Office SDK for AI Agents Across Spreadsheets, Docs, and Slides
Explore Univer, an Apache 2.0 open-source office SDK enabling humans and AI agents to co-edit spreadsheets, documents, slides, and infinite canvases within a si
Artificial intelligence agents are rapidly extending beyond software engineering into everyday operational workflows such as document composition and quantitative data analysis. Univer, an open-source project by dream-num, is an Apache 2.0 developer SDK specifically architected to provide humans and AI agents with a shared, unified environment where they can collaborate, create, and inspect spreadsheets, documents, presentations, and infinite canvases within a single runtime.
Image source: dream-num / Univer
By collapsing previously isolated office tools into a unified computational environment, the framework resolves the fragmented context that traditionally complicated agent-driven document automation.
Six Core Office Tools in One Runtime and the Unified .univer Format
Rather than relying on disjointed standalone applications that trap data in separate silos, Univer integrates six foundational office tools into a single coherent runtime: Spreadsheets, Documents (Docs), Presentations (Slides), Infinite Canvas, Relational Tables, and PDF (currently under active development).
All components share state and computational resources through a single unified '.univer' file format instead of maintaining disparate disconnected documents. This architecture unlocks deep composition and live embedding across blocks. For example, a presentation slide can embed a live spreadsheet with active formula calculations, while an analytical report can cite source tables where all references update dynamically whenever the underlying data changes. By unifying the storage layer and execution runtime, Univer enables AI agents to manipulate rich compound documents without negotiating inconsistent external APIs across different office media.
High-Performance 10M Cell Formula Engine and Single Facade API
At the engine level, Univer pairs a Canvas-based rendering pipeline with a formula computation engine capable of handling up to 10 million (10M) cells. This allows developers to deliver responsive, enterprise-grade data calculation and display without suffering performance degradation or browser thread lockups during large-scale data manipulation.
Developer ergonomics are grounded in a single Facade API and an extensible plugin architecture that functions identically across browser client environments and Node.js server runtimes. Engineering teams can embed modular spreadsheet, document, or presentation capabilities directly into custom SaaS products and internal tooling without being bound to third-party hosted software or rigid user interface layouts. Furthermore, native real-time human and agent co-editing ensures that every programmatic modification made by an autonomous agent remains traceable, versioned, and inspectable within the shared workflow.
Ecosystem Adoption, DeepSeek Harness, and Implementation Caveats
Univer has demonstrated rapid adoption across the open-source software ecosystem, gathering more than 15,500 GitHub stars and 1,380 forks. Its official documentation portal (docs.univer.ai) provides thorough multilingual guides, including dedicated Korean (ko-KR) documentation and architectural references. The project also connects into conversational agent workflows through specialized integrations such as 'dsh-univer-office' for DeepSeek Harness, allowing agents to generate, inspect, and preview office assets in isolated worktrees directly during a session.
When evaluating Univer for production deployment, teams should account for several verified boundaries. While six core productivity tools are planned, the PDF rendering and viewing module remains in a 'coming soon' state awaiting official release. In addition, importing and exporting native Microsoft Office formats (.xlsx, .docx, .pptx) or configuring deep integration with specific conversational agents requires configuring dedicated harness plugins such as dsh-univer-office alongside the core SDK.