Univer

Overview

Univer occupies a specific niche in the agent tooling landscape: it is the office document layer. Most agent frameworks handle reasoning, tool calls, and orchestration, but when an agent needs to actually read a spreadsheet, write a document, or generate a presentation, it typically shells out to a hosted service (Google Docs API, Microsoft Graph) or uses limited file-format libraries. Univer provides an open-source, self-hostable alternative — a full office runtime that agents can drive programmatically.

The “Office Harness for AI Agents” tagline reflects its dual nature: it works as a traditional embedded office suite (users editing spreadsheets in a browser) and as an infrastructure layer for AI workflows (agents reading, modifying, and creating office documents through structured APIs).

Key Capabilities

Unified Document API: Spreadsheets, documents, presentations, relational tables, and PDFs share a consistent API surface. An agent that knows how to interact with one document type can interact with all of them without learning separate interfaces.

Isomorphic Runtime: Univer runs the same code in browsers and Node.js. An agent running server-side can manipulate documents with the same APIs a browser-based user interface uses. This eliminates the common problem of server-side document manipulation libraries that produce different results than the browser rendering.

Worktree Collaboration: Borrowed from Git’s model, worktree collaboration lets agents propose changes in isolated draft documents. A human reviewer can inspect the diff, approve or reject changes, and merge — providing a review gate between agent action and production documents. This is particularly important for financial spreadsheets, legal documents, or any content where unreviewed agent edits carry risk.

Plugin Architecture: The core system handles rendering, data models, and collaboration. Everything else — specific formula functions, chart types, import/export formats, AI integrations — is a plugin. This means you can customize the office experience for your domain without forking the core.

Use Cases

Univer fits wherever agents need to produce or consume structured office documents. Financial agents that generate reports in spreadsheet format, legal agents that draft and revise documents, data agents that produce presentation decks from analysis results, or any workflow where the output is a document humans will read, edit, and share. The self-hosted model means document data never leaves your infrastructure.

Who It’s For

Teams building AI agent workflows that produce or consume office documents, and want an open-source, self-hostable alternative to Google Docs or Office APIs. Also valuable for product teams embedding office capabilities (spreadsheet views, document editors) into their own applications.