dream-num/univer

The Office Harness for AI Agents — Spreadsheets, Docs, Slides, Canvas, Relational Tables, and PDF in one runtime.

What it solves

Univer provides a high-performance, open-source SDK for developers to embed fully customizable office productivity tools—such as spreadsheets, documents, and presentations—directly into their own applications. It eliminates the need to rely on hosted third-party apps or fixed UIs, allowing for a seamless integration of productivity surfaces into SaaS products, BI workflows, or AI-driven applications.

How it works

Univer uses a plugin-first architecture where every capability is delivered as a composable plugin. This allows developers to add, remove, or lazy-load only the features they need. It employs a Canvas-based rendering engine for high performance on large surfaces and a dedicated formula engine. The SDK is isomorphic, meaning it can run as a UI application in the browser or as a headless runtime in Node.js for server-side processing and automation.

Who it’s for

It is designed for developers building SaaS products, internal tools, or AI agents that require embedded office editing capabilities. It is particularly useful for those who need a unified API (Facade API) to manage workbooks, documents, and formulas across both client and server environments.

Highlights

  • Isomorphic Runtime: Works in both the browser and Node.js for headless automation.
  • Plugin-based Architecture: Highly modular system for composing custom productivity experiences.
  • Canvas Rendering: Ensures responsiveness even when handling complex, large-scale workbooks.
  • AI Agent Ready: Provides programmatic APIs for agents to inspect and modify content, with support for isolated worktrees for human review.
  • Unified Facade API: A consistent interface for interacting with office elements across different environments.

Related

  • Project
    davisking/dlib

    A modern C++ toolkit containing machine learning algorithms and tools for creating complex software to solve real-world problems.

  • Project
    casdoor/casdoor

    An open-source, self-hosted identity and access management platform providing single sign-on (SSO) and multi-protocol authentication for applications.

  • Project
    NVIDIA/TensorRT-LLM

    An optimization library for LLMs and visual generation models that uses specialized kernels and an efficient runtime to maximize inference throughput and minimize latency on NVIDIA GPUs.

  • Project
    horovod/horovod

    A distributed deep learning training framework that makes it is easy to scale single-GPU training scripts to multiple GPUs and hosts across TensorFlow, PyTorch, Keras, and MXNet.