Univer, an Apache office SDK for people and agents, ranks third on Trendshift

The dream-num/univer repository ranked third on Trendshift's daily list on 14 November. The repository had 24,148 stars that day. Its README calls it an office SDK for building office applications inside your own product. It offers editing, calculation, collaboration and file processing, by the project's own account, for people and existing agents.

Univer is Apache-2.0 for the open parts. Commercial modules are sold separately. The owner on GitHub is Dream Co., Ltd. The project has been under active development since September 2022. The current release is 1.4.0, published on 7 November. The project says 1.4.0 adds office attachment workflows, shape merging in slides and richer pivot support.

What it covers

The README names six document types in one list. They are spreadsheets, documents, presentations, relational tables, canvases and PDFs. Each has a content model, a plugin system, and formula and rendering engines. The project also exposes what it calls a Facade API, so third-party code can call into the editors.

A reader can see the full stack running in Univer Workspace, the project's own reference app. The README calls it a self-hostable workspace. It shows how to tie editors, collaboration, file capabilities, agents and human review into one product. It is at space.univer.ai.

How an agent plugs in

The AI SDK is built on the same content models the editors use. It exposes structured TypeScript APIs. An agent calls a command on the document model. It does not type cell by cell or rewrite a CSV. The README lists three agent capabilities: content inspection tools, structured editing commands and visual inspection through rendered screenshots.

The isolated draft mode is called Worktree. An agent edits a copy of the document. A person reads the changes before they land on the shared copy. That is the pattern a buyer in a regulated setting is most likely to ask for. Let the agent write. Gate the write on a human read.

What the rank is, and what it is not

Trendshift measures what repositories a daily audience is clicking on. Third place is a real number and today's news. It is not a measure of quality, and it is not a release. The underlying repository is three years old, and the last release was 7 November. The rank says more people are reading about Univer this week than most of the other public AI office tools.

For a product team that already owns a document surface, the Univer bet is smaller than it looks. The editors are a library the team can embed. The agents call the same commands a person's menu does. The code is Apache-2.0. For a team moving an existing product in the office direction, the real question is one of fit. Does the SDK match the documents its customers already have? The six document types in the README are a start on that answer.

Source

Primary source: dream-num/univer on GitHub.