One agent, with your repository and your entire docs tree in context. It drafts the missing pages, rewrites the stale ones, repoints the links and writes the metadata. Then it stops and waits for you to read the diff.
Jump to the AI Agent →, the writing tools → or what readers get →
Generic assistants guess. This one has the code, the docs and the reader data in front of it, which is the only reason its suggestions are worth reviewing.
Search analytics → Reader feedback →Connect GitHub and it reads the code behind the docs: routes, types, the spec, the commits. It watches the pull requests you open, and starts the changeset before the branch has even merged. Disconnect it and the copy we hold is deleted.
Every page, every version, every synced block and the API reference, plus the docs repo itself, where it checks each pull request for links and references that no longer resolve. Not just the page you happen to have open.
Searches that returned nothing. Pages rated unhelpful. That list is your docs backlog, written by the people using them.
The agent is the full-sized one. It works across your whole docs tree rather than a paragraph, it watches your code repository, and the work arrives as a changeset you review page by page. This is what keeps documentation updating itself.
One line of instruction becomes headings, steps, code blocks and a table where the content wants one, written in the shape your other pages already use.
“Rework API Basics for a first-time reader” covers all nine pages, and the internal links and page syntax are corrected as it goes.
Stale snippets, endpoints that no longer exist, two guides that contradict each other. Surfaced with a fix attached, not just a warning.
Open a pull request on the repository behind your docs and the agent begins the changeset there and then, so the writing is waiting when the code lands.
It watches the docs repository too, validating that links and references still resolve on each pull request instead of after a reader finds the 404.
Everything a run touched lands in a single changeset. Approve it, request changes, or take three pages of it and leave the rest.
The writing tools are the lightweight half, and they never leave the paragraph you are in. Select any text in the editor, press the AI Tools button, and pick what you want done to it. No prompt to compose, no run to wait for, no changeset to review: the sentence changes and you carry on.
Both of them propose, neither publishes. Agent work arrives as a changeset against your current pages, with the old text beside the new. Take the parts you agree with and leave the rest.
The same content, indexed. A reader types a question into one bar and gets an answer with the page it came from, across guides, API reference and code steps. Every query is logged, so the searches that come back empty become the next thing the agent drafts.
Metadata, commit messages, reading levels, review states. None of it is why you got into this, and all of it is why docs rot.
Connect DeveloperHub to any MCP client and your documentation becomes a tool the rest of your AI stack can read and write, with your project's context attached.
Publish a set of changes and the message is written for you, describing what was added, updated and removed. Edit it before it saves.
Meta titles and descriptions generated page by page. Reading grade, estimated read time and Flesch score on every page, on every plan, including Free.
AI output lands in a draft. Share it, comment on it, request changes, publish when it is right. Admin, Publisher, Writer and Reviewer are real roles with real limits.
The AI terms are short and specific, which is the point. Three of them are worth reading before you connect anything.
No AI feature publishes to your live documentation without a person reviewing and approving the change first. The output is a proposal, and we say so in the terms.
AI requests go only to providers who do not retain your content or use it to train their models. The providers are listed publicly, not buried.
Where there is a choice of model, you make it, and the hosting region of each one is shown at the point you choose it.
No submit button, no page reload, no spinner to watch.
Every localised version searched at once, best match wins.
Behind SSO, JWT or a password, with results filtered per reader.
Generated for you, so AI tools index your docs properly.
Placeholders swapped for the reader's real key, URL and org.
Top queries, no-hit rates and click positions, six months back.
Personal data stripped out of reader messages automatically.
The bots that find your feedback box do not reach your inbox.
Suggested from your own content, with hover definitions inline.
Numbered, multi-language walkthroughs from a description.
Tags and categories proposed as you write, no taxonomy meeting.
The agent, the writing tools, commit messages and metadata come with every paid plan, from Plus at $99 a month, which includes 1,025 AI credits a month free. Pro includes 1,800 and adds the AI Assistant. You can buy more credits whenever you need them.
Fourteen days free, no card. The agent proposes, you approve, and your docs stop being three sprints behind the product.
