# Word GPT Plus: an AI and agent add-in for Microsoft Word > Word GPT Plus is a Vue-based Office add-in that puts OpenAI, Azure, Gemini, Groq and Ollama models inside Word, with a LangChain agent that can edit the document itself. The interesting part is also the part to think hard about: an agent with write access to your file. **Kuingsmile/word-GPT-Plus** — Word GPT Plus is a word add-in which integrates the AI&Agent into Microsoft Word. - Repository: https://github.com/Kuingsmile/word-GPT-Plus - Stars: 1,363 · Forks: 231 - Language: Vue - License: MIT - Published: 2026-09-10 · Updated: 2026-09-10 · Language: en - Canonical page: https://hysenlabs.com/projects/kuingsmile-word-gpt-plus ## The problem: model choice inside the document you are already editing Most writing assistants force a workflow: copy text out of Word, paste it into a browser tab, get an answer, paste it back, then spend time re-applying styles. Word GPT Plus takes the opposite position. It is an Office add-in, so it runs in a task pane next to the document, and the README describes its purpose as generating, translating, summarising and polishing text without leaving the Word environment. The target reader is someone who already pays for model access somewhere and wants that access available where the writing happens. That is a narrower audience than "everyone who writes", and the requirements section makes it narrower still: the add-in needs an API key from OpenAI, Azure OpenAI, Google Gemini or Groq, or a local Ollama instance. There is no bundled free tier in the repository. The README lists Ollama as available for local deployments only, which is the closest thing to a no-cost path, and it still assumes you can run a model locally. The project also carries a Chinese README, README_cn.md, and the interface is documented as available in English and Simplified Chinese, so it is clearly built with a bilingual audience in mind. ## Chat mode versus agent mode: two different trust models in one pane The add-in exposes two modes, and the distinction matters more than the feature list suggests. Chat mode is the familiar pattern: you ask, the model answers, and you decide what to do with the response. The README describes per-message actions for replace, append or copy, so the model never touches the document on its own. Agent mode is a different contract. It is described as powered by LangChain and able to read, write and modify the Word document through built-in tools: web search, insert text, format content, create tables, manage bookmarks, and search and replace. The package manifest confirms the stack: @langchain/core, @langchain/langgraph, @langchain/openai, @langchain/google-genai, @langchain/groq and @langchain/ollama sit alongside zod for schema validation. LangGraph implies a graph of steps rather than a single request, which matches the README's mention of multi-step reasoning with conversation memory and a visible thought process. Streaming responses arrive in real time, and the README states you can abort generation at any time. My reading of this design is that agent mode is the actual product and chat mode is the safe default. Anyone evaluating the add-in should treat them as separate decisions, because granting an agent insert, format and table-creation tools over a live document is a materially larger permission than asking a model for a paragraph. ## Installing Word GPT Plus: manifest, sideload, first prompt The README offers two installation routes and recommends the first. Method 1, instant use, needs no build step: download release/instant-use/manifest.xml, save it to a dedicated folder such as C:\Users\username\Documents\WordGPT, and sideload it into Word following Microsoft's own guide on creating a network shared folder catalog for task pane add-ins. The README notes that users in China may need to add msq.pub to their proxy rules or self-host instead. Method 2 is for developers or anyone who wants a private backend, and the Docker path is the shortest version of it. The Dockerfile builds the Vue app on node:22.21.1-alpine3.22, then serves the dist directory from nginx:alpine on port 80. The README's command maps that to host port 3000. ```bash docker pull kuingsmile/word-gpt-plus docker run -d -p 3000:80 kuingsmile/word-gpt-plus ``` After that you download release/self-hosted/manifest.xml and edit it, replacing every instance of http://localhost:3000 with your server's address. Then sideload it the same way. If you prefer to build from source, the README gives this sequence, which needs Node.js 20 or newer: ```bash git clone https://github.com/Kuingsmile/Word-GPT-Plus.git cd Word-GPT-Plus yarn yarn build yarn run serve ``` The serve script is vite preview --port 3000, so the built app answers on port 3000 and the self-hosted manifest already points there. The repository also carries a wordgpt.bat at the top level and a Deploy to Tencent EdgeOne button in the README, which uses npm run build with ./dist as the output directory and yarn install as the install command. Once the pane is loaded, the first real use is the least risky one: select a paragraph, pick a provider and model in the pane, and use a quick action such as translation or polishing rather than agent mode. The README states that responses are formatted with Word styles and that Markdown is parsed and converted, so the first thing to check is whether the output looks like your document's own formatting or like pasted plain text. ## Where the add-in stops: file formats, platforms and the agent's blast radius Three constraints are stated plainly and are worth taking at face value. First, the README says the add-in works only with .docx files and is not compatible with the older .doc format. If your archive is full of legacy documents, this is not the tool for them. Second, the software requirements are specific: Word 2016 or 2019 in the retail version, Word 2021, or Microsoft 365, plus the Edge WebView2 Runtime. Volume-licensed or older perpetual builds are not in that list, and WebView2 is a separate install on some machines. Third, the add-in has to be sideloaded, which in many corporate environments means either a central deployment or nothing at all. The larger risk is not compatibility but agent mode. The README gives the agent tools to insert text, create tables, manage bookmarks and run search and replace across the document. An agent that misreads a request can make many edits before you notice, and the README does not document an undo, a diff view or a rollback path for agent actions. That is a real gap, not a nitpick. The practical mitigation is procedural: work on a copy, keep Word's own undo history in mind, and prefer chat mode when the instruction is ambiguous. ## How it compares with a plain OpenAI-compatible Word plugin The obvious alternative is a thin Word add-in that points at one provider and offers a chat box. Those exist, and the trade-off is legibility: one endpoint, one key, one behaviour to reason about. Word GPT Plus goes the other way. Its provider list spans OpenAI-compatible endpoints, Azure OpenAI with custom deployment names, Google Gemini, Ollama for local deployments, and Groq, and the README says custom model names are supported for all platforms, with adjustable temperature and max tokens per provider and custom base URLs and proxies. That breadth is the selling point and the maintenance surface at the same time: each provider is a separate LangChain integration in package.json, and each can drift. A second alternative is doing the work through a general chat client and pasting results into Word. That keeps the document out of the agent's reach entirely, at the cost of losing Word-aware tools and the automatic style formatting. Neither alternative is worse in the abstract. If your only need is summarising a selection with one model, the simpler plugin is easier to justify. Word GPT Plus earns its complexity when you genuinely switch between providers or want the agent to manipulate structure, such as building a table from a set of notes. ## Maintenance, storage and what the MIT licence does and does not cover The repository is not archived and the last push was on 2026-09-09, so the project is being touched. The release cadence visible in the repository is modest: v2.0.1 on 2026-01-04, v2.0.0 on 2025-12-28, v1.1.2 on 2025-11-13, while package.json already declares version 2.0.2, which suggests unreleased work in the tree. The dependency list is broad and moves fast: LangChain packages at major version 1.x, Vue 3.5, Vite 6, Tailwind 4, Dexie 4, zod 4. There is a renovate.json at the top level, so dependency updates appear to be automated, and that also means the build can break when an upstream package changes. The README states that data is stored in local storage, which is a privacy argument in the add-in's favour because prompts and keys are not routed through a project-operated server. It is also a cost: settings live per browser profile, and clearing site data means re-entering keys. The licence is MIT, which is permissive and short, but it governs the code, not the model providers you connect to. Your API usage is bound by whichever provider's terms you accepted, and the README's own links to OpenAI, Azure, Google AI Studio and Groq are the places to check those. This is not legal advice; read LICENSE and the provider terms yourself. ## Conclusion Adopt it if you already hold an API key for a supported provider and want model choice plus document-aware tooling inside Word, and you are comfortable sideloading a manifest and reviewing an agent's edits before saving. Do not adopt it if you need the older .doc format, if your organisation blocks sideloaded add-ins, or if you expect a managed service with a support contract. Before you roll it out, verify which manifest path fits your setup (the instant-use one or the self-hosted one pointing at your own server), confirm that your Word build is 2016/2019 retail, 2021 or Microsoft 365, and check the licence text in the repository for anything your legal team cares about. ## FAQ ### Is there a ChatGPT add-in for Word? Word GPT Plus is one: it is an Office add-in that runs in a task pane and connects to OpenAI as well as Azure OpenAI, Google Gemini, Groq and Ollama. It is installed by sideloading a manifest rather than through a store listing. ### What does Word GPT Plus do? It generates, translates, summarises and polishes text inside Word, and its agent mode, built on LangChain, can read, write and modify the document through tools for inserting text, formatting, creating tables, managing bookmarks and search and replace. ### How can I use Word GPT Plus for free? The add-in itself is MIT-licensed and the repository does not charge for it, but you still need model access. The README lists Ollama as available for local deployments only, which is the option that does not require a paid API key. ### Does Word GPT Plus need a paid Microsoft Word subscription? The README's requirements list Word 2016 or 2019 in the retail version, Word 2021, or Microsoft 365, so a perpetual retail copy can work. It also requires the Edge WebView2 Runtime, and the add-in only handles .docx files. ## Sources - [Issues](https://github.com/Kuingsmile/word-GPT-Plus/issues) - [Kuingsmile/word-GPT-Plus on GitHub](https://github.com/Kuingsmile/word-GPT-Plus) - [License: MIT](https://github.com/Kuingsmile/word-GPT-Plus/blob/master/LICENSE) - [README](https://github.com/Kuingsmile/word-GPT-Plus/blob/master/README.md) - [Releases](https://github.com/Kuingsmile/word-GPT-Plus/releases) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/kuingsmile-word-gpt-plus