# Office SDK overview

> Learn about the three parts of Office SDK: Web SDK, Server SDK, and AI SDK, and how they work together in an Office application.

- Human documentation: [https://docs.univer.ai/overview](https://docs.univer.ai/overview)

- Agent Markdown: [https://docs.univer.ai/overview.md](https://docs.univer.ai/overview.md)

- Requested language: `en-US`

- Content language: `en-US`

- Documentation version: `1.3.0`

- Source: [index.mdx](https://github.com/dream-num/documentation/blob/dev/content/docs/index.mdx)

---

Univer Office SDK includes Web SDK, Server SDK, and AI SDK. It supports spreadsheets, documents, slides, Relational Table, Canvases, and PDFs.

Web SDK provides the shared content engine, running both browser editors and headless content processing. Server SDK and AI SDK both depend on this headless runtime. Server SDK adds collaboration and file processing, while AI SDK adds content operations and verification for agents.

## Explore features

Browse the [feature catalog](https://docs.univer.ai/overview/features.md) by product and category, including support levels, open-source availability, and integration scope. Shared capabilities include charts, file conversion, collaboration, Worktree, and embedding.

[Get started](https://docs.univer.ai/overview/get-started.md): start with one requirements template and three runnable examples, then connect, build, and deploy your application.

## SDK architecture and responsibilities

- [Web SDK](https://docs.univer.ai/guides/sheets.md)
- [Server SDK](https://docs.univer.ai/server.md)
- [AI SDK](https://docs.univer.ai/ai.md)

### Web SDK: the shared content engine

For developers displaying, editing, or processing Office content in their application. Use Facade APIs to work with content, data, and UI in browser editors or headless Node.js applications.

Web SDK uses a plugin architecture: product capabilities, editor UI, and individual features such as find and replace can be composed through plugins. Start with a preset, or select the product and feature plugins your application needs and assemble them with their dependencies.

You can also develop custom plugins for business-specific features without modifying Univer's core code. Different plugin combinations adapt the same content engine to each environment: browser applications load editor UI, while headless applications omit browser UI plugins. See [Univer Architecture: plugin design](/blog/univer#plugins).

**Shared Canvas rendering.** Office content uses a shared Canvas-based rendering architecture, allowing content types to reuse layout and drawing capabilities and support nested content. For example, spreadsheet cell text reuses document layout and rendering for consistent content presentation. See [Architecture of Rendering Engine](/blog/rendering).

**Hybrid computation.** The formula engine supports a hybrid computation model in which frontend calculation works with a server-side Rust engine, combining responsive editing with performance for large formula workloads. Explore [formula capabilities](https://docs.univer.ai/guides/sheets/features/core/formula.md) and [server-side calculation](https://docs.univer.ai/overview/features/shared/server-calculation.md).

[Start with spreadsheets](https://docs.univer.ai/guides/sheets/getting-started/quickstart.md), or choose [documents](https://docs.univer.ai/guides/docs.md), [slides](https://docs.univer.ai/guides/slides.md), [relational tables](https://docs.univer.ai/guides/bases.md), [boards](https://docs.univer.ai/guides/boards.md), and [PDFs](https://docs.univer.ai/guides/pdfs.md).

### Server SDK: connect collaboration and files

For developers adding shared editing, file conversion, and backend extensions. Build on Web SDK's headless content capabilities, integrate the modules into your own backend, and use your application's identity, permissions, and storage.

**Real-time collaboration.** OT (Operational Transformation) handles concurrent operations such as row/column changes and content edits. Edits apply locally first, then converge across replicas while preserving user intent where possible. Add history, comments, and Worktree as needed. See [collaborative editing design](/blog/ot) and the [collaboration quick start](https://docs.univer.ai/server/collaboration/quick-start.md).

**File conversion.** Convert between Office files and Univer document data to connect import, editing, and export, with or without collaboration. See [Office file import and export](https://docs.univer.ai/server/import-export/node.md) for conversion APIs and [file exchange integration](https://docs.univer.ai/server/import-export.md) for uploads, tasks, and downloads.

**Deployment and scaling.** Assemble the modules in your own Node.js application. Start on one machine, separate file conversion as load grows, then scale to multiple collaboration nodes. See [application architecture](https://docs.univer.ai/server/architecture.md) and the [deployment guide](https://docs.univer.ai/server/deployment.md).

[Explore Server SDK](https://docs.univer.ai/server.md).

### AI SDK: give agents an entry point

For developers adding Office capabilities to existing agents, automation, and application-specific CLIs. Build on Web SDK's headless runtime and Facade APIs for content operations, calculation, and inspection.

**Connect your existing agent stack.** Expose Office tools through TypeScript APIs or an application CLI assembled from the SDK. Your application owns model selection, task orchestration, and Skills. Agents receive execution results and inspection feedback, then continue their workflows.

**Create content that stays editable.** Agents read document structure, locate ranges or paragraphs, then use Facade APIs to create and modify spreadsheets, documents, slides, and other content, including formula calculation. Results retain their content structure and editing capabilities so people can continue refining them. See [content operations](https://docs.univer.ai/ai/content-operations.md).

**Inspect, then revise.** Read back values and formula results, and inspect screenshots of the actual page. For slides, layout checks also identify off-page, overflowing, and overlapping text. Agents use the feedback to revise and verify again, forming a read → edit and calculate → inspect → revise loop. See [visual inspection](https://docs.univer.ai/ai/visual-inspection.md).

Connect to Server SDK and use [Worktree](https://docs.univer.ai/ai/worktree.md) to isolate intermediate agent changes before human review and merging in the Web application. [Explore AI SDK](https://docs.univer.ai/ai.md).

## Application workflow

```mermaid
flowchart TB
    Human[Human user] --> Browser
    Agent[AI agent / Automation] --> AI[AI SDK · Operations and verification]
    subgraph WebSDK[Web SDK · Shared content engine]
      Browser[Browser editors]
      Headless[Headless · Content model and Facade APIs]
    end
    Server[Server SDK · Collaboration, conversion, and extensions] -->|Depends on| Headless
    AI -->|Depends on| Headless
    Browser <-->|Shared content| Server
    AI -.->|Collaboration and review| Server
    Server --> Storage[Application identity, permissions, and storage]
    classDef default rx:8,ry:8,stroke-width:1px
```

- **Embedded editor**: use Web SDK to display and edit content in your application.
- **Collaborative application**: combine Web SDK and Server SDK so users can edit shared documents together.
- **Human and agent collaboration**: agents edit and verify drafts through AI SDK, Server SDK manages versions and merging, and the Web application provides viewing, comparison, and human review interactions. All three contribute to the complete review workflow.

> [!NOTE: Shared foundation and dependencies]
> Web SDK can run independently as an editor or headless content engine. Server SDK and AI SDK both depend on Web SDK's headless capabilities: running without a browser UI still requires the content runtime. Enable the services and modules your application needs; a complete human–agent workflow typically combines all three SDKs.

## Start with your use case

| I want to | Start here |
| --- | --- |
| Embed a spreadsheet editor | [Web SDK quick start](https://docs.univer.ai/guides/sheets/getting-started/quickstart.md) |
| Choose another Office editor | [Documents](https://docs.univer.ai/guides/docs.md), [slides](https://docs.univer.ai/guides/slides.md), [relational tables](https://docs.univer.ai/guides/bases.md), [boards](https://docs.univer.ai/guides/boards.md), or [PDFs](https://docs.univer.ai/guides/pdfs.md) |
| Let multiple users edit the same document | [Collaboration quick start](https://docs.univer.ai/server/collaboration/quick-start.md) |
| Import or export Office files | [Server SDK file conversion](https://docs.univer.ai/server/import-export.md) |
| Operate documents through an agent or CLI | [AI SDK overview](https://docs.univer.ai/ai.md) |
| Review an agent's changes to a shared document | [AI SDK Worktree](https://docs.univer.ai/ai/worktree.md) |

## Next steps

Start with Web SDK's content runtime, combine it with Server SDK and AI SDK to complete your first application workflow, then add modules incrementally. Keep related Univer package versions aligned and prepare the environment, authorization, and licenses required by your chosen modules.

Explore [Examples](https://docs.univer.ai/overview/examples.md) for open-source applications, agent integrations, demos, and repositories. You can also try the Office products in the [online showcase](https://office.univer.ai).