AI Agent Editor
The AI agent is a conversational editor that can modify any part of your documentation site. Tell it what you want to change in plain language, and it reads your files, makes the edits, and writes them back -- all reflected instantly in the live preview.
What the Agent Can Do
The agent has full read/write access to your docs site. Common requests include:
- Content -- rewrite a page, add a new section, improve wording, add code examples
- Branding -- change primary color, update logo, adjust fonts
- Navigation -- add or reorder navbar links, update footer links
- Pages -- create new documentation pages, delete pages, move pages between categories
- Styling -- customize CSS, adjust spacing, change border radius, modify dark mode styles
- Layout -- edit the homepage hero section, update feature cards, restructure the sidebar
- Media -- reference uploaded images, embed videos
- Configuration -- change site title, tagline, favicon, color mode
There are no off-limits areas. If it exists in the docs site, the agent can change it.
Agent Tools
The agent has 7 tools it uses to interact with your site:
| Tool | What It Does |
|---|---|
readFile | Reads a page or file from the site |
writeFile | Creates or updates a file (changes appear instantly) |
deleteFile | Removes a page or file from the site |
listFiles | Lists the contents of a directory |
searchFiles | Searches for text across all pages and files |
runCommand | Runs a shell command (install packages, run scripts, etc.) |
listProjectStructure | Shows the complete file tree of the site |
The agent always reads a file before editing it and writes the complete file content to avoid partial updates that could break the page.
How to Use the Agent
The agent panel is on the left side of the editor. Type what you want in the chat input and press Enter.
Be specific when you can. Instead of "make it look better," try "increase the font size of sidebar items and add more vertical spacing between them."
You can reference pages by name. The agent knows which page you are currently viewing and has a list of all your documentation pages.
Example Prompts
Here are some practical examples:
| Goal | Prompt |
|---|---|
| Change brand colors | "Change the primary color to #1a56db" |
| Add a page | "Add a new page about webhooks in the guides section" |
| Rewrite content | "Rewrite this page to be shorter and more direct" |
| Update the navbar | "Add a link to our GitHub repo in the navbar" |
| Custom CSS | "Make the sidebar font larger and add more spacing" |
| Add code examples | "Add a curl example to the authentication page" |
| Restructure navigation | "Move the FAQ page from guides into the getting-started section" |
| Embed a video | "Add a YouTube embed at the top of the quickstart page" |
| Install a package | "Install the docusaurus-plugin-image-zoom package" |
If you are not sure what to ask, the agent panel shows suggestion prompts when the conversation is empty. These cover common tasks like changing brand colors, adding pages, rewriting content, and updating navigation.
Uploading Files
You can upload images and documents through the attachment button in the agent panel.
Images
Uploaded images are placed in the /img/ directory of your site. Reference them in any page with standard Markdown:
The agent is aware of uploaded images and can insert them into pages for you.
Documents
Uploaded documents (PDF, DOCX, TXT, Markdown) are processed for text extraction. The extracted text is provided to the agent as reference material, so you can say things like "Use the API spec I uploaded to write the endpoints page."
How the Agent Communicates
The agent keeps its responses short -- typically 1-2 sentences describing what it changed. It never mentions underlying technologies like Docusaurus, React, MDX, or TypeScript. It refers to things as "your site," "your navbar," "your styles," and so on.
For content and style changes, the agent will note that you can check the live preview. For configuration changes (navbar, footer, site title), it may mention that the update can take a moment to appear as the development server refreshes.
In long editing sessions, older tool outputs are automatically trimmed to keep the conversation responsive. The agent retains full context for recent messages (the last 6 exchanges) and summarized context for older ones.