AccelerUpDocs
Copilot/Knowledge base

Knowledge base

Teaching the Copilot about your systems — runbooks, tickets, wiki pages and third-party documentation.

The knowledge base is what the Copilot knows about your systems, as opposed to what the model knows in general.

What goes in

KindSource
runbookRunbooks
postmortemPost-mortems
jira_issueJira
confluence_pageConfluence
docCrawled documentation sites
CustomAnything you add by hand

Documentation crawling

Third-party and internal documentation sites are crawled daily.

DiscoverySitemap, falling back to a breadth-first crawl
Change detectionContent hash — only changed pages are re-embedded
ScheduleDaily
Tip

Crawl the documentation for the third-party services you depend on. "What does this payment provider's error code 4402 mean" is a question that otherwise costs somebody ten minutes in the middle of an incident.

Content-hash diffing is what makes a daily crawl affordable: a thousand-page site where three pages changed re-embeds three pages.

Managing it

Settings → AI Copilot → Knowledge. Add sources, see what has been ingested, and read the training log — which says what was crawled, what changed and what failed.

Why runbooks and post-mortems matter most

Your own history is the highest-value content in here. "Has this happened before?" is the fastest route to a fix, and the answer is usually in a post-mortem nobody remembers or a ticket closed eighteen months ago.

A model without that context can only tell you what usually causes this class of problem. With it, it can tell you what caused it here, last March.

Keeping it honest

A knowledge base with a wrong runbook in it produces confidently wrong answers. Fix runbooks during the incident that revealed they were wrong, and the knowledge base stays current as a side effect.

Where this behaviour lives: backend/internal/worker/copilot_kb_worker.go, backend/internal/integration/doc_crawler.go. If the code and this page disagree, the code is right — please fix the page.

Part of CopilotThe assistant that can read your monitoring data and act on it.