The documentation platform that started as an open-source tool for developers now wants to keep the world's product docs honest - for humans and the AI agents reading over their shoulder.
Nobody reads the documentation until it is wrong. Every developer knows the ritual: the code changes, the docs do not, and three weeks later somebody files a support ticket that starts with the words "the docs say." For most of software's history this was treated as a chore, a thing you did at the end of a sprint if there was time, which there never was. GitBook, a company that has spent a decade building tools for exactly this unloved corner of the industry, arrived at a stranger conclusion. The docs were not just being ignored by humans. They were increasingly being read by something else.
GitBook makes a platform for writing, publishing, and maintaining documentation - the product manuals, API references, and internal wikis that sit behind almost every piece of software you use. It began in 2014 as an open-source project out of France, built by Samy Pessé and Aaron O'Mullan, aimed at turning Markdown files and Git repositories into readable books. That heritage still shows. Today the company reports more than two million people use it, and its customer list reads like a directory of companies whose products you already have open in another tab: Google, Nvidia, Zoom, FedEx, Snyk.
The pitch is deceptively plain. A team writes its documentation in GitBook's editor, hosts it on a custom domain, and readers get a clean, searchable site. What makes it more than a text box is the plumbing underneath. GitBook's Git Sync keeps the published docs in step with a GitHub or GitLab repository, so an engineer can edit docs in a pull request while a writer edits the same content in a friendlier editor, and the two stay reconciled. For API documentation, the platform reads an OpenAPI file and generates an interactive reference. It is, in the jargon of the field, "docs-as-code" - and GitBook's contribution was making that idea usable by people who do not live inside a terminal.
The customers fall into a few clean buckets. There are the software companies publishing public product docs and developer references, the kind a user hits when an integration misbehaves. There are teams running internal wikis and knowledge bases, keeping the institutional memory that otherwise lives in one senior engineer's head and leaves when they do. And there are the open-source projects taking GitBook up on free hosting, the same audience that gave the tool its first users a decade ago. It is a rare product whose free tier and enterprise contracts are aimed at recognisably the same person, a few years apart in their career.
"AI made docs easy to write. Not easy to trust."
GitBook homepage, 2026That distinction - easy to write versus easy to trust - is the hinge the whole company now turns on. Generative AI made producing a first draft of anything nearly free. The scarce thing became knowing whether the draft was correct, current, and safe to ship. Documentation is where that scarcity bites hardest, because a wrong sentence in the docs is not just embarrassing; it breaks someone's integration at two in the morning.
Then came the plot twist GitBook is best positioned to notice. Because it hosts so many docs sites, it can see who - and what - is reading them. At the start of 2025, the company says, AI systems made up under 10% of the readers on documentation it published. By the end of the year, they were over 40%. Chatbots, coding assistants, and autonomous agents had quietly become a large share of the audience, crawling docs to answer questions their users typed somewhere else entirely.
If a growing share of your readers are language models, the design brief changes. A human skims headings and screenshots; a model wants clean, structured text it can ingest without tripping over your navigation menu. GitBook's answer in 2025 was to make its docs natively legible to machines: every page can be served as plain markdown, sites publish llms.txt and llms-full.txt files that hand an AI a map of the content, and each docs site now auto-generates an MCP server - effectively turning a set of documentation into a queryable interface that agents can call. Documentation stopped being only a web page and started being an API.
Knowledge should move as fast as your product.
GitBook, company missionMaking docs readable is only half of "trust." The other half is keeping them right. Here GitBook has pointed AI at its oldest enemy: rot. GitBook Agent scans content and flags pages that have gone stale - the paragraph describing a feature that shipped differently, the endpoint that quietly changed. AI Insights watches where readers get stuck and points writers at the pages that need work. And GitBook Assistant sits inside a published docs site answering reader questions directly, which, if it works, means a question answered is a support ticket never filed.
| Layer | What it does | Who it serves |
|---|---|---|
| Editor + Git Sync | Write docs; keep them in sync with the code repo | Writers & engineers |
| API / OpenAPI | Generate interactive API references from a spec | Developers |
| GitBook Assistant | Answer reader questions from the docs | End users |
| GitBook Agent | Detect stale, outdated content | Docs owners |
| MCP / llms.txt | Make docs consumable by AI agents | Machines |
The business runs on the familiar SaaS ladder. There is a free tier - including free hosting for open-source and non-profit projects, a nod to where the company came from - then paid plans for teams, and Business and Enterprise tiers that unlock single sign-on, tighter security, and the heavier AI features, some of which are sold as an add-on. Third-party trackers peg GitBook's revenue in the low single-digit millions and its valuation around the same, though the company has kept its finances private and those figures should be read as estimates. Its disclosed funding history is modest: early backing from Point Nine Capital and Fly Ventures, and none of the mega-rounds that define the category's louder players.
That restraint is part of the story. GitBook competes in a crowded room - Notion and Confluence from the general-workspace side, ReadMe, Mintlify, Docusaurus, and Read the Docs from the developer-docs side. The general tools are more flexible but not purpose-built; the developer tools are precise but often demand you live in a repo. GitBook's position is the seam between them: structured enough for engineers, approachable enough for everyone else, and now further out on the AI-readability edge than most.
The expertise it draws on is narrow by design. GitBook has never tried to be a whiteboard, a project tracker, or a second inbox. For ten years it has done one thing - the structured, versioned, publishable document - and the depth shows in the details: the way a change in a repo threads through to a live page, the way an OpenAPI spec becomes a browsable reference, the way a docs site can now be locked behind authentication for internal-only content. That focus is also the risk. A single-purpose tool lives or dies on whether its one purpose stays valuable, and GitBook is now betting that documentation is about to matter more, not less.
A wrong sentence in the docs is not embarrassing. It breaks someone's integration at two in the morning.
In 2025 the company changed at the top. Co-founder Samy Pessé stepped back from the CEO role to become CTO, focusing on product and engineering, while Sara Tandowsky took over as chief executive. It is the kind of move that reads as a company deciding what it wants to be when it grows up: a builder handing the wheel so he can keep building. GitBook still describes its culture in the vocabulary of its origins - open source, transparency, respect for the community that used the free tool long before the enterprise contracts arrived.
The bet underneath all of it is simple to state and hard to pull off. Documentation has always been treated as overhead. GitBook is arguing it is becoming infrastructure - the layer where a company's product knowledge is kept accurate and served to whoever, or whatever, needs it. If AI keeps eating docs at the rate the company is seeing, the unglamorous business of keeping the manual correct may turn out to be one of the more consequential jobs in software. GitBook has been doing it for ten years, which is either good timing or a lot of patience. Possibly both.