Skip to main content

Knowledge Sources Overview

Knowledge Sources let you supply OpsWorker with two kinds of organization-specific knowledge that the AI couldn't otherwise infer from your cluster:

  1. Internal knowledge — documentation you already write: runbooks, postmortems, architecture notes, internal wikis. OpsWorker references this when answering chat questions and during investigation reasoning.
  2. Best-practice rules — explicit constraints on how OpsWorker should think about your environment. Rules can be hard ("never recommend deleting PVCs") or soft ("prefer HPA changes over manual scaling").

Both kinds load into the AI's context the same way as memory: automatically, when relevant.

Knowledge Sources vs. AI Memory

There's overlap between Knowledge Sources and AI Memory. Use this guide:

ConcernUse AI MemoryUse Knowledge Sources
FormatOne-line facts and preferencesMulti-paragraph docs and structured rules
Update cadenceAdd as you learnBulk-load from existing docs
SourceManually entered factsExisting internal documents
Best for"Team B owns pay-*""How to recover from a database failover"
Best for"Keep answers concise""These are our hard guardrails for prod changes"

In short: memory captures discrete knowledge atoms; knowledge sources capture the documents and rule sets you already maintain.

Where Knowledge Sources Live

In the OpsWorker portal, Knowledge Sources appears in the left navigation with two subviews:

  • Feed internal knowledge — connect, upload, or paste internal documentation
  • Best-practice rules — define rule sets that constrain OpsWorker's behavior

How OpsWorker Uses Knowledge Sources

When an investigation runs or a chat session starts, OpsWorker retrieves relevant knowledge source entries and includes them in the AI's context window. The AI:

  • References internal knowledge when a topic in the conversation matches a connected document — for example, if you ask about a recovery procedure documented in a runbook, the AI quotes or summarizes the runbook.
  • Respects best-practice rules as constraints — they shape what the AI will and won't suggest. A "never auto-remediate" rule, for instance, means the AI never recommends an action without explicit human approval.

Best Practices

  • Start with rules first. Even a small ruleset shapes AI behavior immediately. Internal docs take longer to curate but provide deeper context.
  • Keep rules unambiguous. If a rule needs interpretation, it'll be interpreted inconsistently. "Don't restart pods in prod between 09:00 and 17:00 UTC" is better than "be careful with prod restarts".
  • Refresh internal knowledge as it changes. Stale runbooks generate stale recommendations.
  • Don't put secrets in knowledge sources. Knowledge sources are not a credentials vault.

Next Steps