base9 keeps that knowledge complete, consistent and current. It builds it from what you already have: your articles, your resolved tickets and your internal documents, and writes the fix for your team to approve.
A work email, approved by a person within a day or two. You get a made-up help center with 30 solved tickets to try, and can add your own. See how it works
Your help center could have answered 122 of your last 200 solved tickets.
Made-up example. The numbers, tickets and article are illustrations, not results.
The problem
An agent solves something new in a reply or an internal note, the ticket closes, and the answer stays in that ticket. The next agent starts from zero, and so does any AI agent you put in front of your customers.
Fixes and workarounds stay in tickets, internal notes and chat threads.
Releases ship and policies change. Articles that were right become partly wrong.
Two articles give different steps or limits. Agents and AI pick one at random.
Writing articles competes with answering tickets, and loses. Knowledge programs fade.
How it works
The idea behind Knowledge-Centered Service (KCS), without the manual work.
Your helpdesk and your knowledge base, plus internal runbooks and saved replies if you want them in.
Each one is summarised as it's read: what was asked and what fixed it. The conversation, and who the customer is, aren't stored.
Compared with your articles, with quoted evidence from both the ticket and the article.
A new article or an edit, written before you ask and only from what your agents wrote. New articles are checked sentence by sentence.
One change at a time. Each sitting ends with your coverage before and after.
For every resolved ticket
Each result comes with the article and the passage it rests on, checked word for word. Tickets that ask the same thing are grouped into one topic, with the count: "asked in 14 tickets".
Counted as answered, and the article is credited as doing the work.
The step or condition that fixed it is missing. An addition is drafted from the ticket.
A new article, written from how your team resolved the tickets that asked.
Both passages side by side. Which one is right is your call.
Closed with no answer, a duplicate, or not a question. Skipped, with the reason recorded.
What needs you
What base9 finds is turned into the fix that settles it, and each fix is one decision. One number says how many are waiting, and each kind says about how long it will take.
A missing step or an outdated value can't be worked out from the text. base9 lists these and asks. It doesn't guess.
9 articles need something only you know before a fix can be written.
Made-up example numbers.
A few minutes a week
Nobody has to open a finding, ask for a fix and wait. The most valuable fixes are prepared in the background, and you go through them one at a time.
The answer your team gave on SUP-3120 disagrees with this article.
Password reset links are valid for 24 hours30 minutes.
Approving replaces this one passage, exactly where it stands.
Done. "Account security FAQ" now says: "Password reset links are valid for 30 minutes."
Was 122 of 200. Now 132 of 200. 8 fixes approved.
Made-up example.
When things disagree
Two articles give different steps or limits, or an article says one thing and your team told a customer another. base9 can't check your product, so it doesn't decide. It puts the two passages side by side and asks.
Two articles give different expiry times for a password reset link.
Both can be true · Not sure, skip
Made-up example.
The number to watch
The share of resolved tickets whose answer was already in your knowledge base. Few teams know it. It shows where your help center is working, and it goes up as approved fixes go in.
| Article | Answered | Partly | Contradicted |
|---|---|---|---|
| Resetting a device to factory settings | 64 | 2 | – |
| Apps not appearing after install | 8 | 13 ◐ | – |
| Account security FAQ | 31 | 4 | 2 ⚠ |
| API rate limits | 22 | – | – |
⚠ Contradicted: your team told a customer something else · 3 articles
◐ Close but not enough: comes close more often than it answers · 5 articles
○ Not needed by any ticket in this period · 16 articles
"Answered" means the article holds the answer and the customer wrote in all the same.
Made-up example numbers.
How do you know it's right?
Nothing here asks you to take the AI's word. Three ways to see for yourself.
Accuracy
Scored against your team's answersSay whether base9's results on your own tickets are right. It is scored against what you said, and it tells you when there are too few to rely on.
Try a question
Ask what a customer wouldType a question and see what your articles would answer. Each part of the answer is shown with the passage it came from. Nothing is sent to anyone.
Guardrails
Checks the AI can't skipThey run in code on every result: every quote is looked up in its source, nothing changes without approval, and words you never want in an article are caught before a fix can be approved.
Is it working?
Counted from your own tickets, for the last 4, 12 or 26 weeks. Every row leads to the fix or the tickets behind it.
| By product | Tickets | Answered |
|---|---|---|
| Single sign-on Weakest | 22 | 18% |
| Integrations | 61 | 56% |
| Billing | 84 | 78% |
| Fix | Tickets since | Answered |
|---|---|---|
| Correct: reset link expiry | 9 | 9 ✓ |
| New: sync stops when a token expires | 7 | 6 ✓ |
| New: re-linking SSO | 8 | 3 ⚠ |
⚠ Re-linking SSO: 5 of 8 tickets since still weren't answered. The article may be missing something.
Made-up example numbers.
It reads every article
That's how it notices a step written for an older version, or an answer meant for a different product.
SAML response invalidSettings → Security → Single sign-on1. Open Settings → Security → Single sign-on and choose SAML.
2. Paste your identity provider's metadata URL and save.
3. Test with one admin account before turning it on for everyone.
Made-up example. Highlighted: the passage that answered a ticket, checked word for word.
Safe by design
Every new article and every edit waits for you. Sending approved fixes to your knowledge base is a separate step, and stays your call.
A new article uses only what your tickets say, and is checked sentence by sentence. An edit replaces one passage, exactly where it stands, or it isn't applied.
Each ticket is read once and summarised: what was asked and what fixed it. The conversation, and who the customer is, aren't stored.
Missing steps and outdated values are listed for you, not invented. Where two articles disagree and the evidence doesn't settle it, base9 says so.
base9 reads tickets and articles; writing to your knowledge base is a separate permission you grant. Invite colleagues at your company: owners decide what is connected and checked, everyone can review.
Kept in your own workspace, separate from every other customer's. Security & privacy
Works with what you have
No migration and no new place to write. base9 reads from where your knowledge already lives.
Makes every AI better
Better knowledge, better answersWhoever answers your tickets, a new hire, an outsourced team, your helpdesk's built-in AI or a chatbot, they can only be as good as the knowledge they work from.
Start here
Read-only, with no card. You see your coverage and the fixes base9 would make, before deciding on anything ongoing.
You get
Coverage and the top gapsYour coverage, the topics costing the most tickets, and drafted articles for them.
Plus
Conflicts and article healthArticles that contradict each other or your agents' answers, articles with problems, and articles no ticket has needed.
We need
Your help center and solved ticketsThe address of your help center and up to 500 solved tickets. Book a demo if you would rather set it up with us and walk through what it finds.
What's built, and what's rolling out with pilots
Built
Rolling out with pilots
Everything under "Built" runs today. The connections have been run against test accounts; pilots are where they meet a customer's own tickets.
After the month
Ongoing
The loop on every resolved ticket, with changes ready for a few minutes of review a week. From $500 a month, set by the tickets you solve, not per user. Pricing
No commitment
The free month stands on its own: nothing to cancel. Continue only if the numbers are worth it.
Early access
We're working with a small number of teams first, closely. Their feedback shapes what we build next.
Questions
No. base9 reads your existing knowledge base, and your team keeps writing and publishing where it already does. Approved changes are applied in base9 first; sending them back is a separate step you choose.
No. Every new article and every edit waits for approval, and what reaches your knowledge base stays your decision.
Built-in AI uses your knowledge as it is, or helps a person write. base9 finds what's missing or wrong, from your resolved tickets, with evidence, and drafts the fix from what your agents actually wrote. It works across tools, and it makes the AI you already have better.
A new article is built only from your tickets, never from the AI's general knowledge, and checked sentence by sentence against them. An edit replaces one passage and is refused if the article has changed since it was written. Where base9 can't know (a missing step, which of two articles is right), it asks. And a person approves everything. And you can check it yourself: Accuracy scores base9 against results your own team confirmed.
base9 doesn't keep your tickets. Each one is read once and summarised: what was asked and what fixed it, with the customer left out. The summary is then checked for leftover emails, phone numbers and names. To write it, the ticket is sent to an AI model over a business API; see Security & privacy.
A few minutes a week. The changes are written ahead of time and reviewed one at a time, in sittings of about ten. Disagreements between articles are asked as questions with a suggested answer.
Read-only access to your helpdesk and knowledge base; to start, your help center's address and some solved tickets are enough. Ongoing, someone who owns the knowledge base and reviews the changes.
No. base9 Knowledge keeps your knowledge complete and current, and is where every team starts. base9 Triage uses that knowledge to route and draft replies for new tickets; it is optional and available on request, configured for one team at a time and tested on that team's own tickets.
Next step
A work email and your website is all we need. Before we talk, we'll look at your public help center.
Got it, thank you. We'll be in touch within a day or two. If it's urgent, email contact@base9.io.