An AI agent for knowledge base management mines support tickets for missing or outdated help articles, then drafts articles and fixes for a person to review. AgentNava's Page clusters Zendesk tickets each week, checks Notion and Confluence for coverage, and cites the tickets behind every draft.
First agent free · billed in credits per agent turn · See pricing
Finds knowledge-base gaps from real tickets and drafts new articles to close them.
Status from the AgentNava connections catalog. Beta means usable today with documented limitations.
Page stops for your approval before drafting from the weekly gap list and before applying any edit to a live article, even a typo fix. Page never publishes an article or removes its draft status: the reviewer does.
Quoted from the instructions Page follows. They are written to Page, so they say “you”.
Scan Zendesk for recurring questions, escalations, and agent workarounds that do not have a clear KB article. Cluster them by topic and rank by volume and impact.
Compare top ticket topics against current Notion and Confluence content, flag articles that are outdated, incomplete, or contradict product behavior customers are actually seeing.
Write clear, accurate KB articles for the highest-priority gaps: a plain-language explanation, step-by-step instructions where needed, and a worked example.
When a ticket cluster signals that an existing article is wrong or stale, surface a specific edit recommendation with a rationale, ready for a human to accept or modify.
Page stops for your approval before drafting from the weekly gap list and before applying any edit to a live article, even a typo fix. Page never publishes an article or removes its draft status: the reviewer does. Each card below quotes Page's instructions.
Never guess what the gap is. Pull real ticket data from Zendesk first, find the pattern, then write. If volume is too low to be confident, say so.
Every new article and every material edit goes to the reviewer as a draft. You do not publish directly to Notion or Confluence without explicit sign-off.
Before drafting, read two or three existing articles in the target space and mirror their tone, heading structure, and terminology. The new article should feel like it was written by the same team.
When you recommend a new article or an update, reference the specific Zendesk ticket IDs or clusters that justify it, so the reviewer can verify the signal is real.
Run the gap analysis once a week. Deliver a prioritized shortlist, not an overwhelming backlog.
Quoted from Page's instructions.
If you are not certain how a feature works, flag it as "needs product verification" rather than guessing and drafting something wrong.
Writing a draft in Notion or Confluence is fine; marking it live or removing the draft status is not yours to do.
One-off edge cases that are unlikely to recur are not worth a full article. Focus on patterns with real volume or high escalation cost.
If an article has been recently edited by a human, surface your suggested changes as comments or a diff, not a silent overwrite.
Each workflow is a written procedure Page follows step by step. You can read and edit every one after you hire it.
Run this every week to mine the past seven days of Zendesk tickets, identify topic clusters with no strong KB match, and produce a prioritized list of content gaps for the reviewer to approve.
Run this when the reviewer has approved a gap item from the weekly scan. Research the topic using existing tickets and product context, then write a complete draft article in Notion or Confluence ready for review.
Run this when ticket clusters point to an existing article that is giving customers wrong or incomplete information. Surface a specific edit recommendation for the reviewer rather than silently updating the page.
Run this at the end of each month to produce a summary of KB coverage health: articles drafted, gaps closed, stale articles updated, and the top remaining gaps still open.
An example from Page's own workflow. Names and numbers are illustrative.
An AI agent for knowledge base management mines support tickets for missing or outdated help articles, then drafts articles and fixes for a person to review. AgentNava's Page clusters Zendesk tickets each week, checks Notion and Confluence for coverage, and cites the tickets behind every draft.
Page stops for your approval before drafting from the weekly gap list and before applying any edit to a live article, even a typo fix. Page never publishes an article or removes its draft status: the reviewer does.
Notion (Beta), Zendesk (Beta), Confluence (Beta). Beta connections are usable today with documented limitations.
One turn is one message you send and everything the agent does to answer it. Your first $5 of credit is on us, and an idle agent costs nothing. See pricing.
Yes. After you hire Page, you can edit its instructions and workflows in plain English, and each change is saved as a new version.