Memory Read

Overview

The Memory Read node retrieves previously stored data from the platform's memory system. Memory is scoped to a named namespace (memory_id) and can be either session-scoped (tied to a single workflow run and discarded when the run ends) or long-term (persisted across runs for the same user or context). Use this node at the start of a workflow to restore state from a prior run, or mid-workflow to look up facts written by an earlier memory_write node in the same session.

Configuration

Field	Type	Required	Default	Description
memory_id	string	Yes	—	Namespace that groups related memory entries. Typically matches a user ID, session ID, or logical context name (e.g. "user:{{userId}}:prefs"). Supports template expressions.
key	string	No	"*"	Specific memory key to retrieve. Use "*" to return all keys within the namespace up to limit.
type	string	No	"session"	Memory tier to read from. One of "session" (run-scoped) or "long_term" (persistent across runs).
limit	integer	No	10	Maximum number of entries to return when key is "*". Range: 1–100. Ignored when a specific key is requested.
default_value	any	No	null	Value emitted on the output port when the key is not found, instead of routing to not_found.

Inputs

Port	Type	Description
memory_id	string	Optional runtime override for the memory_id config field.
key	string	Optional runtime override for the lookup key.

Outputs

Port	Type	Description
output	any	The retrieved value (or default_value if the key was absent and a default is configured). When key is "*", emits an array of { key, value, updated_at } objects.
not_found	null	Emitted when the key does not exist and no default_value is set. Use to branch workflows on a cache-miss.

Example

{
  "nodeType": "memory_read",
  "config": {
::: v-pre
    "memory_id": "user:{{run.userId}}:conversation",
:::
    "key": "history",
    "type": "long_term",
    "limit": 10,
    "default_value": []
  }
}

Notes

• Session memory is automatically purged when the workflow run completes; long-term memory persists until explicitly deleted or overwritten by a memory_write node.
• Reading all keys ("*") is useful for debugging but should be avoided in production for namespaces with many entries; use a specific key and paginate via limit instead.

• memory_id supports {{expression}} template syntax, allowing the namespace to be parameterised by run context variables such as run.userId or run.sessionId.

• The not_found output port allows clean cache-miss handling; connect it to an if_condition node or directly to a node that generates the missing value.