Memories

A KomputerMemory is a named, persistent knowledge resource that can be shared across agents. When an agent has memories attached, their content is injected into Claude's system prompt before each task — giving the agent persistent context without repeating it in every task prompt.

When to use it

• Runbooks — Standard operating procedures, escalation paths, cluster-specific notes
• Domain knowledge — Background information an agent always needs (API docs, schemas, team conventions)
• Shared context — Facts that should be consistent across multiple agents (company policies, product details)

How it works

1. Create a KomputerMemory CR with a content field (markdown supported) and an optional description
2. Reference it by name in spec.memories on a KomputerAgent
3. When the agent wakes up, the API resolves all attached memory names, fetches their content, and prepends it to the agent's system prompt
4. The .status.attachedAgents field tracks how many agents reference each memory

Example

apiVersion: komputer.komputer.ai/v1alpha1
kind: KomputerMemory
metadata:
  name: deployment-policy
  namespace: platform
spec:
  description: "Production deployment rules"
  content: |
    ## Deployment Policy
    - No direct pushes to main — all changes via PR
    - Run `make test` before every deploy
    - Rolling update strategy only (never Recreate in prod)

Attach to an agent:

spec:
  memories:
    - deployment-policy

Agents can also create and attach memories dynamically at runtime using the create_memory and attach_memory manager tools.