Knowledge Base Support
A customer asks the same question you answered on Tuesday. This agent reads it, finds the paragraph in your own help content that covers it, and drafts the answer with a link back to where it came from. When nothing you have published covers the question, it says so and routes it to a person, then logs it. That log becomes your list of docs worth writing.
Yours outright — no required subscription.
Every agent is built for your business — your systems, your approval chain, your way of working. We scope it on the call.
If this sounds too technical, don't worry. We take care of everything for you.
- Private VPS deployment. This agent runs on a server that belongs to you, not a shared cloud tenant. Your help content and your customers' questions stay on your box.
- A technician agent alongside it. Every hire ships as a pair: the Knowledge Base Support Agent answering from your documentation, and a VPS-technician agent on the same box that keeps it patched, backed up, and monitored. You're not hiring one thing, you're hiring a small team of two.
- Communication channel connection. Wired into wherever questions actually arrive — a shared inbox, a chat widget, a client or patient portal, a web form — so nothing has to be forwarded to it by hand.
- Answers tied to your published content. It answers from your help center, FAQ, or internal knowledge base and nothing else. No general knowledge, no web, no plausible-sounding paragraph filling a gap. Every answer names the article and section it came from, with a link.
- A refusal log that is your gap report. Every question it declined is timestamped with the reason. That log is the literal list of things customers asked that your documentation does not answer — not an inferred trending topic. It can draft a first version of a missing article; whether that gets published, and what it says, stays yours.
- Security hardening. Private VPN, firewall, and encryption configured on your server before it answers a single question.
- 14 days of priority support after launch, to tune it against your real question mix — including how readily it should refuse. You own everything when it's over: the server, the agent, the code it runs on.
Not sure it fits? Check fit in 90 seconds in the free assessment chat.
Who this is for
- You're answering the same fifteen questions every day, and the answer usually lives in a document somebody wrote a year ago. It's still faster to scroll your Sent folder than to go find it.
- You have a help center. It was populated in one big push, it hasn't been touched since, and your team quietly answers around the parts that are now out of date.
- You know you should be writing more documentation, and you don't know which pages to write, because nobody in the company owns the report that would tell you.
- You won't put anything in front of customers that might invent an answer, and every tool you've looked at has asked you to take that on faith.
How it earns trust
The failure mode of AI on support questions isn't that it gets things wrong. It's that a wrong answer reads exactly like a right one, and you hear about it from the customer. This agent makes the opposite bet: every answer, and every refusal, leaves something you can check.
Every answer names its source
The article and section it drew from, with a link. A wrong answer traces to a specific paragraph instead of to a black box — so you can tell whether the agent misread the document or the document itself is wrong. Those are different fixes.
The passages it retrieved are kept
Recorded against the question, so a near miss is visible as a near miss. When someone asks about returning a sale item and the article only covers returns in general, you can read what the agent thought it was matching rather than guessing at it.
A weekly digest of answered, cited, refused, escalated
A spot-check is a five-minute read instead of a data export. Pull any ten answers, open the articles they cite, and you'll know whether the problem is the agent or the docs.
Nothing here depends on taking our word for it — you can open the citations yourself, any week you feel like checking.
Pairs well with
These share a workflow with this role. Tick any to add them to your setup.
We have a help center. Nobody reads it. How does this help?
It doesn't ask anyone to read it. The customer asks their question the way they'd ask a person, and the agent goes and finds the paragraph that answers it. Your documentation stops being a page people are supposed to visit and starts being the source your answers come from. Setup is scoped on a free 15-minute call, and delivery is typically about seven days from kickoff.
I'm not putting a bot in front of customers if it might make something up.
It can only answer from what you've published. If nothing in your content covers the question, it says so and routes to a person — that refusal is the product, not a failure of it. It also never auto-sends outside a scope you define, and it never removes an escalation path you have today. The people who handle the hard questions keep handling them, in the same order.
Our FAQ page hasn't been touched since we changed prices. What happens then?
It repeats the old price, cites the article, and sounds confident doing it. Grounding protects against invention, not against your own outdated content. A person answering by hand quietly routes around a wrong document; this does not. The upside is that the citation points straight at the paragraph, so one edit fixes the answer for everyone who asks next.
I know we should write more docs. I don't know which ones.
The refusal log tells you, in the customer's own words, with dates. Read it with judgment: thirty "where's my order" messages are a status-lookup problem, not a missing article, and one frustrated person asking the same thing five ways can show up as five gaps. It gives you the raw list; you decide what deserves a page.
What is it never allowed to do?
It never answers from anything outside your published content. It never publishes, edits, deletes, or overwrites a help document on its own — it drafts a candidate and flags the gap. It never acts on an account: no refunds, no order changes, no rescheduling or cancelling appointments, no discount codes, no credentials. It answers questions about your policies; it does not execute against them.
What's covered under the $1,997 setup, and what isn't?
One company, one system of record for your content, one approval chain. Multiple products or locations with genuinely different policies, or content split across sources that contradict each other, get scoped on the call rather than priced differently. You own the server, the code, and the keys; we keep the playbook we used to build it. After the 14-day support window, we remove our own access.
Hiring more than one? A department on tap — the subscription puts a build team behind every request, agent after agent.