AI & search

Answers, with receipts

Your readers ask in their own words and get a clear answer that cites its source — drawn from your documentation, never the open web. What they can’t find becomes your backlog.

How do I rotate an API key?
Grounded in your docs
Searching your documentation…

Simulated assistant. Every answer is drawn from your published docs and links back to the source — never the open web.

Grounded, not guessed

Answers you can actually trust

A reader assistant is only worth shipping if its answers are right. Three guarantees keep it honest.

Reads only your docs

Every answer is retrieved from your published, reviewed content — never the open web, and never invented.

Always cited

Each claim links back to the exact source block, so a reader can verify it in one click and trust what they read.

Says when it can’t answer

If the docs don’t cover it, the assistant says so plainly — and that question becomes a gap you can fill.

Gap detection

The questions you’re not answering yet

Every question the assistant can’t answer, and every search that comes up empty, is a signal. Gravity turns those dead ends into a ranked list of the docs worth writing next.

Searched, not found

Docs worth writing next

Promoting gaps to backlog…

MCP

Memory for your coding agents

The same reviewed docs that answer your readers can answer your tools.

Model Context Protocol

Your docs become memory your agents can read

Expose your published, reviewed reference over MCP and every coding agent on your team can query it directly. An assistant answering "how does the billing service work?" pulls from your real documentation — not a stale copy, and not a guess.

  • One connection, available to every agent and tool
  • Always the reviewed, current version — never a snapshot
  • Scoped to what each agent is allowed to see
MCP serverYour docs

Search that understands meaning

Readers shouldn’t have to know your words

Keyword search only finds the exact terms you happened to write. Semantic search reads what the reader actually means — so they land on the right page whether or not they know your terminology. Toggle between the two on the same query.

  • Matches intent, not just literal words
  • Ranks the page that truly answers first
  • Keyword search still available where you want it
why is my deploy stuck on pending

Troubleshooting a build that never finishes

A deployment that stays in “pending” is usually waiting on a queued build slot…

Deployment statuses explained

Pending, building, live and failed — what each state means and when to act…

Pending invitations

How to resend or revoke a teammate invitation that is still pending…

Semantic search reads the intent, so the page that actually answers the question comes first — even with none of the same words.

0%
of reader questions answered from your own docs
0%
of answers carry a source citation
more dead-end searches turned into a backlog
0
connection exposes your docs to every coding agent

In your docs

Where it lives

The assistant ships inside your own branded documentation site, on your domain.

The assistant, in your published docs

A live product screenshot will sit here — the ask box, the cited answer and the source panel, rendered inside your own branded documentation site.

Placeholder — representative of the in-product assistant.

Turn your docs into answers.

Publish your reviewed documentation and give your readers — and your agents — a cited assistant on day one.