Ask agent
Use the editor's built-in AI agent to write content, edit pages, leave comments and suggestions, search your site, upload files, and configure settings.
The editor includes a built-in agent that can edit pages, reorganize navigation, update docs.json, search across your entire repository, and manage dashboard settings. The agent is also available on dashboard settings pages.
The editor agent makes changes directly in your current branch. Like your own edits, the agent’s changes save automatically but don’t reach your live site until you publish them.
Open the agent
Only editors and admins can open the agent.
Click Ask agent in the editor toolbar, or press Cmd + I (macOS) or Ctrl + I (Windows) when no text is selected. You can also open the agent on dashboard settings pages. Your chat session stays open as you move between the editor and settings.
Type @ to mention a specific page in a prompt for the agent to focus on. Without a mention, the agent uses the page you currently have open as its initial context. On your dashboard, the agent knows which page you’re viewing so you can ask about the current page without naming it.
Example prompts:
simplify the introduction pagefix all grammar errors across my contentadd a new page that explains authenticationrename every mention of "Acme Pro" to "Acme Team" across all pagesupdate docs.json to add a new group called "Guides"
Attach files
Click the paperclip icon in the chat box, or drag and drop files onto the panel. The agent reads attached files as context for your request.
Supported file types:
- Images: JPG, PNG, GIF, WebP, SVG
- Documents: PDF
- Code and text:
.js,.ts,.jsx,.tsx,.mdx,.md,.json,.yaml,.html,.css, Python, Go, Rust, Ruby, Java, Swift, C, C++, SQL, shell scripts, and more
Maximum size: 5 MB per file, except SVG files, which Mintlify limits to 256 KB. Maximum: 10 files per message.
Mintlify strips scripts and other active content from SVG files before the agent reads them.
Add a selection to the agent
In visual mode, select text and click Add to agent in the floating toolbar to send it to the agent as context.
Ask for comments and suggestions
The agent edits content directly by default. To get feedback without immediately changing the page, explicitly ask the agent to review the page, leave comments, or make suggestions.
- Suggestions propose a specific replacement that you can accept or reject. The original text remains visible with a strikethrough until you decide, and the agent can include an explanation in the suggestion thread.
- Comments leave feedback or a question anchored to specific text without changing the page.
Example prompts:
review this page and leave suggestionscomment on anything that needs more contextsuggest ways to make the introduction more concise
The agent uses the page you currently have open unless you name another page. It can annotate rendered page text, including prose inside components. It cannot annotate frontmatter, page metadata, configuration, or raw MDX component tags and attributes.
Review what the agent changed
A file changes list appears in the chat as the agent works. Expand it to see every file modified in the session, and click any file to compare the changes against the original in diff view.
To revert the agent’s edits, click Discard changes next to a file or click Discard all to discard every change from the session. To revert the changes from a single response, click Undo changes on that message. Click Redo changes to restore it.
What the agent can do
Edit pages
The agent can write, rewrite, expand, and reorganize content on any page. It reads your existing content to match your style and structure.
Search and navigate your content
The agent can search across your entire repository, not just the page you have open. Use it to find information, check for inconsistencies, or verify that content exists before adding it.
Update navigation and docs.json
The agent can add, rename, reorder, and delete navigation elements. These are the same changes you’d make manually in the navigation panel. It can also update docs.json configuration directly, including adding new groups, adjusting settings, and configuring redirects.
Example: add a "Quickstart" group under the Getting Started tab and move the quickstart page into it
Run bash commands
The agent can run grep, rg, and other bash commands against your repository. Use this for bulk operations that span many files.
Example: find every page that mentions the deprecated /v1/auth endpoint
Configure your site (code mode)
For requests that go beyond editing pages (setting up authentication, managing workflows, or changing deployment settings), the agent switches into code mode. It writes and runs a script against your Mintlify dashboard on your behalf.
Code mode respects your dashboard permissions. If you don’t have access to a setting in the dashboard, the agent can’t change it either.
Example prompts that use code mode:
enable JWT authentication for my sitecreate a workflow that updates my site when I merge a PRadd a custom domain
Use connected integrations
Integrations require an Enterprise plan.
The editor agent can use third-party apps connected through Integrations as read-only tools while it researches, edits pages, or answers your questions.
- Shared integrations are available to everyone in your organization.
- Personal integrations use the account connected by the teammate chatting with the agent. For example, a Google Drive search uses your own Drive connection.
Ask the agent to use a connected app naturally in your request, such as:
update the migration guide based on the latest Jira issuecheck what the Notion launch brief says about availability and add it to the release notes page
If the integration you need isn’t connected yet, ask the agent to connect it. The agent returns an authorization link. Complete the flow, then send another message so the agent can confirm the connection and continue.
Ask about your analytics
The agent can answer questions about your documentation’s traffic, search, assistant, and feedback data. The agent returns information in writing, as a chart, or as a ranked list. Use the agent to request specific explanations, compare page performance, or investigate changes in traffic.
Example prompts:
what are the top 5 user agents hitting my site?which pages do AI agents visit more often than humans?what are people searching for and not finding results for?graph traffic for the REST API introduction page over the last month
Read your Slack workspace
If you install the Mintlify Slack app, the editor agent can read your Slack workspace for context while it researches or answers your questions. The editor agent has read-only access. It cannot post messages, add reactions, or perform other actions in Slack.
Ask the agent to use Slack in a request. For example:
check the #engineering channel for context on the new auth flowsummarize what the team decided about rate limiting yesterday
If you haven’t installed the Slack app, ask the agent to connect it for you. The agent returns an authorization link and guides you through the installation process.
Continue an automation run
When you open the editor from an automation’s Slack or email notification, the agent panel opens automatically with context on what the automation did. An Updated pages card at the top of the chat lists every page the automation modified. Click any page to open a diff view.
The agent has context on the automation’s prompt, a summary of the changes it made, and which pages it modified. You can ask it to refine or extend the automation’s work without re-explaining the background.
For example: The new section on rate limits is too long. Trim it to three sentences.
Session history
Click the clock icon in the panel header to see your previous chat sessions. Click any session to reopen it and see what the agent changed. To start over, click New chat.
AI instructions
To give the agent persistent guidance like voice rules, terminology, or formatting conventions, configure AI instructions in Editor settings. The agent follows these instructions on every request without you needing to repeat them.