AI Memory Overview
What AI Memory Does
AI Memory is a persistent context store that travels with OpsWorker between sessions. When an investigation runs or you ask a question in chat, the AI loads relevant memory entries and uses them as part of its reasoning. This means:
- The AI knows the same things you know about your environment.
- Knowledge that lived in one engineer's head — "the payments DB is owned by Team B", "we never restart the ingress-nginx pods during business hours" — becomes available to every investigation.
- Personal preferences ("keep answers concise", "always include the kubectl command") stay attached to your sessions without needing to repeat them.
Memory is opt-in to capture but always-on once stored. There's no need to explicitly load it — the AI reads from it whenever it's relevant.
Memory Scopes
Memory entries are scoped to one of three levels:
| Scope | Visibility | Use For |
|---|---|---|
| Personal (My Memory) | Only you | Your preferences, your personal notes about a cluster |
| Cluster | Everyone working on this cluster | Cluster-specific facts (ownership, on-call, deployment rules) |
| Organization | Everyone in the org, across all clusters | Company-wide standards, escalation policies |
Personal entries never become visible to teammates unless you explicitly promote them. See Promotion.
Memory Types
Two kinds of memory entries exist:
| Type | Description | Example |
|---|---|---|
| Fact | Something objectively true about your environment | "The checkout service is deployed in the payments namespace" |
| Preference | How you want the AI to behave | "Don't suggest restarting pods unless logs clearly indicate a deadlock" |
Facts can be promoted up the scope hierarchy (personal → cluster → organization). Preferences are personal only — they don't get shared.
How OpsWorker Uses Memory
Memory is consulted at two key points:
- Investigation start — When an alert triggers an investigation, OpsWorker loads cluster and organization facts relevant to the affected resources. The agents use these as ground truth when interpreting logs, events, and configurations.
- Chat session start — When you open AI Chat, OpsWorker loads your personal preferences (so the AI matches your style) and any cluster or organization facts relevant to the selected cluster.
The AI then references memory the same way it references any other data source — when a memory entry contradicts what it sees in the cluster, it surfaces the discrepancy rather than silently picking one.
Where to Find Memory in the Portal
Memory lives under AI Memory in the left navigation:
- My Memory — your personal preferences and per-cluster notes
- Cluster Memory — shared facts for the currently selected cluster
- Organization Memory — facts shared across all clusters in your org
You can create, edit, and delete entries from these views.
When to Use Memory vs. Knowledge Sources
Memory holds short, structured facts and preferences. For longer-form documentation — runbooks, postmortems, internal architecture docs — use Knowledge Sources instead. Memory is the right place for "the AI should always know X". Knowledge Sources are the right place for "the AI should be able to look up Y when relevant".
Next Steps
- Memory Types — Facts vs. preferences
- My Memory — Personal scope
- Cluster Memory — Team scope
- Organization Memory — Org-wide scope
- Promotion — Sharing personal facts upward