Using Memories
Memories are persistent key-value pairs stored at the workspace level. They allow agents to save and retrieve information across separate executions, making it possible to build workflows that learn and adapt over time.
How Memories Work
Each memory has three parts:
- Namespace groups related memories together. Think of it as a folder name.
- Key identifies a specific piece of data within a namespace.
- Value is the stored information.
The combination of namespace and key must be unique within a workspace. If an agent writes to a memory with the same namespace and key, it updates the existing value.
Viewing Memories
- Open Memories from the sidebar.
- The page displays all stored memories in the format
namespace:key = value. - Each entry shows when it was last updated.
Long values are truncated in the list view. You can expand them to see the full content.
If no memories have been created yet, the page shows an empty state.
Using Memories in Agents
Agents can read and write memories during execution through the memory variable type. When you define a variable with the "memory" type on an agent, the agent gains access to the workspace memory store.
This is useful for scenarios such as:
- Keeping a running summary that updates with each execution.
- Storing preferences or configurations that persist across runs.
- Tracking state for multi-step processes that span several executions.
Managing Memories
You can create, update, and delete memories directly from the Memories page or let agents manage them during execution.
Each memory entry tracks its last update timestamp, making it easy to see which data is current and which may be stale.