MIDNIGHT
ProductPricingDocsSign in
Start free
FOR SAAS TEAMS

Your docs already have answers nobody can find

Search gives your users ten links. An agent trained on the same pages gives them the answer, with a link to the page it came from — and tells you which questions your documentation cannot answer at all.

Train one on your site, free
Paste your URL. It reads your pages and starts answering in minutes.
WHAT IT HANDLES

The questions you answer again and again

“Why am I getting a 401 from the API?”
“What is the rate limit on the free plan?”
“Do you support SAML SSO?”
“How do I rotate an API key without downtime?”
“Is there a webhook for subscription changes?”
WITH RECEIPTS

Every answer says where it came from

The agent answers from your pages and cites them. When something is wrong you can see which page taught it that, which turns an argument about the bot into a five-minute content fix.

And when retrieval is too weak to be sure, it says so instead of filling the gap with something that sounds right.

VISITOR
I keep getting 401 on /v1/chat even though my key is right
AGENT
A 401 on that endpoint usually means the session token is bound to a different origin than the one calling it. The token carries the origin it was issued to, and every later call re-checks it — so a token taken from one site will not work on another.
1. API reference — Authentication2. Embed guide — Troubleshooting
WHAT TO TRAIN IT ON

You already wrote all of this

1
Your documentation site
The whole thing. Crawled, chunked, and cited back with a link to the exact page.
2
API reference
Parameter names and error codes are precisely what people search for and precisely what search engines handle badly.
3
Changelog and release notes
So the agent answers "is this shipped yet" from the record rather than from vibes.
4
Support macros and past tickets
Paste the answers your team already types. Each becomes retrievable.
It is a mirror, and mirrors are unflattering
The agent can only be as good as the docs behind it. If a page is out of date, the agent repeats it confidently — with a citation, which at least tells you exactly which page to fix. Most teams find the first fortnight is less about the bot and more about the four pages it keeps having to decline to answer from.

Questions we get

How is this different from docs search?
Search ranks pages; this reads them. It also refuses when retrieval is weak, so a user gets "I don't know" instead of a confidently wrong paragraph assembled from the wrong section.
Can I see why it answered the way it did?
Yes — the retrieval inspector shows every chunk considered, what it scored, and what the reranker did with it. That is how you tell a bad answer from a documentation gap.
Does it work on a private docs site?
You can upload the content directly as files or text rather than pointing at a URL, which covers docs behind a login.
Built for a different kind of business?
E-commerceLaw firmsClinics & practicesAgencies