Turn your documentation into an AI assistant
Documentation search rewards people who already know the vocabulary. TeklDesk answers the question as it was asked, and links to the page the answer came from.
The problem
Docs search is keyword search. It works for the reader who knows your product's word for the thing they want, and fails for everyone else — which is precisely the group reading the documentation. Someone looking for "how do I stop it retrying" will not find the page titled "Backoff configuration".
So the reader either scrolls the sidebar hoping a heading looks right, or gives up and asks a human. Both outcomes are the documentation failing at the only job it has.
What visitors actually ask
- How do I authenticate a request?
- What's the rate limit?
- Why am I getting a 403 on this endpoint?
- How do I stop it retrying failed jobs?
- What's the difference between a workspace and a project?
- Which environment variables are required?
Docs Assistant
Answers from this site
Hi! Ask me anything about the documentation.
Natural language in, a cited page out
It understands what's being asked rather than just matching keywords, so a reader can describe the problem in their own words and still land on the right page. The answer comes back with links to the pages behind it, which is what turns a chat reply into a route into your docs rather than a dead end.
Because it understands meaning rather than exact wording, a reader can also ask in one language about a page written in another.
Handles modern, JavaScript-built docs
A large share of documentation sites are built with JavaScript and don't show their real content on the first load — simpler tools miss this and report success on an empty page.
TeklDesk checks what it actually gets back, and if a page looks too thin to be real, it loads it properly and reads what a visitor would actually see. You don't have to know in advance which of your pages need it.
A backlog written by your readers
Every question the bot could not answer is recorded, with the closest match it found. For a docs team that is the highest-signal list available: real questions, from real readers, that your documentation does not currently answer.
Alongside it, the pages cited most often — which tells you which docs are load-bearing before you restructure them.
Common questions
Does it work with Docusaurus, MkDocs, GitBook, or Mintlify?
Yes. TeklDesk reads your published pages rather than your source files, so the generator behind the site doesn't matter — including sites that build their content with JavaScript.
Can I index a private or staging docs site?
Only pages that are publicly reachable can be crawled. For content behind a login you can paste the text in directly or upload it as a file — plain text, Markdown, Word, and PDF are all accepted.
How does it handle code samples?
Code blocks are indexed as part of the page text and can be quoted back in an answer. The answer still links to the page, which is where a reader should go for the full example in context.
What if the docs are wrong?
The bot will repeat them, because it answers from your content rather than from general knowledge. That is the trade: it cannot invent something your docs do not say, and it also cannot correct something they say wrongly. The citation on every answer is what makes a mistake traceable back to the page that caused it.
Point TeklDesk at your site and see what it answers
Index your pages, paste one line of script, and let visitors ask. Free to start, and every answer links back to the page it came from.
Free to start. No credit card required.