Skip to main content

Configuring Memory

The Memory tab controls what an agent can remember across runs, and lets you inspect what it has actually stored once it's deployed.

Turn on memory​

  1. Open the Memory tab.
  2. Turn on the switch next to Memory.

Turning memory on reveals Advanced settings and gives the agent a memory space.

Configure memory settings​

  1. Select Advanced settings.
  2. Optionally change the Memory space — the identifier this agent's memory is stored under.
  3. Set Visibility to Private to this agent or Shared with organization.
  4. Under Memory types, choose which kinds of memory the agent keeps: Episodic (conversation history), Semantic (facts/preferences), and Procedural (learned steps).
  5. Set Recall top K — how many memories the agent retrieves per turn.
  6. Set Similarity threshold — how closely a memory must match to be recalled.
  7. Optionally turn on Entity-keyed memory to thread memory by a business entity, such as a customer ID, instead of by session. When it's on, set an Entity key field and, optionally, an Entity lookup resource.

Changes here save to the agent's draft as you make them — select Save draft in the top bar to keep them, same as any other tab.

Inspect what's stored​

Once memory is enabled, the Memory inspector section shows a read-only view of what the engine has actually written — it does not let you edit entries.

  1. Select a Deployment (runtime workspace) to inspect.
  2. Optionally enter a value in Filter by entity.
  3. Select Refresh to reload.

If nothing appears, the agent has not yet run in a deployed workflow that wrote to memory.

How you know it worked​

The Memory inspector table lists entries once the agent has run somewhere memory gets written to — typically a workflow using this agent's published version, since a run needs a real deployment to write memory against.

Next steps​