Skip to main content
Memories allow agents to store and retrieve information across multiple executions. This enables agents to learn from past runs, track state over time, and maintain context that persists beyond a single invocation.

​
What Are Memories?

Memories are key-value pairs that agents can:
  • Write during execution to save information
  • Read in future runs to retrieve stored context
  • Update when information changes
  • Delete when no longer needed
Think of memories as the agent’s long-term storage - a way to remember important facts between runs.

​
Use Cases for Memories

Tracking State

Remember progress on multi-run workflowsExample: User offboarding agent tracks which systems have been processed

Learning from History

Store patterns discovered over timeExample: Alert triage agent remembers known false positives

Caching Data

Store expensive lookups to avoid repeated API callsExample: Cache org structure from Active Directory (refresh weekly)

Maintaining Context

Preserve org-specific knowledgeExample: Remember team preferences, contact lists, custom procedures

​
How Memories Work

1

Agent Reads Memories

At the start of execution, agent automatically loads relevant memories based on context
2

Agent Uses Information

Memories are part of the agent’s context, informing decisions and actions
3

Agent Writes Updates

During or after execution, agent can create/update memories using the save_memory tool
4

Memories Persist

Stored memories are available in all future runs (until deleted or expired)