Analysis updated 2026-05-18
Let a coding agent make a targeted edit to one paragraph or table cell in a Word document without touching the rest.
Have an AI agent generate a new PowerPoint presentation from an existing file used as a template.
Render slides or pages to images so an agent or reviewer can visually check a document before finalizing it.
Benchmark how well different AI models handle real Word and PowerPoint editing tasks.
| bobaba99/opencode-office | 0xradioac7iv/tempfs | 52191314/web-agent-proxy-sdk | |
|---|---|---|---|
| Stars | 0 | 0 | 0 |
| Language | TypeScript | TypeScript | TypeScript |
| Setup difficulty | moderate | moderate | moderate |
| Complexity | 3/5 | 3/5 | 4/5 |
| Audience | developer | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Windows is untested, the plugin's skill file must be copied into opencode's skills folder manually.
opencode-office is a plugin for opencode, a coding agent platform, that gives an AI agent the ability to read and edit Word and PowerPoint files. It adds five tools plus a written skill document that teaches the agent how to use them well. One tool lets the agent read a document's outline first before deciding what to look at in full, so it understands the structure before touching anything. Another makes anchored, targeted edits, such as replacing one paragraph's text or fixing a single table cell, without disturbing anything else in the file. A third tool creates a brand new Word or PowerPoint file from scratch, or a new PowerPoint file based on an existing one used as a layout template. A fourth renders pages or slides as images using LibreOffice so a person or agent can visually check the result, and a fifth gives direct access to the underlying Python code for tasks the other tools do not cover. Under the hood, the plugin relies on a companion package called opencode-office-core, which uses the Python libraries python-docx, python-pptx, pillow, and pymupdf inside a Python virtual environment that gets created automatically the first time it is needed, so there is nothing to install by hand. LibreOffice is only needed for the image rendering tool, everything else still works without it. The project is developed and tested on macOS and Linux, and Windows is not tested. Installing the plugin means adding it to an opencode configuration file, and separately copying its skill document into opencode's skills folder by hand, since automatic registration is not currently supported. The project includes a small benchmark comparing how well different AI models perform at these Word and PowerPoint editing tasks, scored against a known correct answer rather than the model's own claims, with results across several cloud-hosted models included in the README as a table.
An opencode plugin that gives AI coding agents tools to read, edit, create, and render Word and PowerPoint files.
Mainly TypeScript. The stack also includes TypeScript, Python, python-docx.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.