Skip to content

Memory Write

The Memory Write component provides a remember tool that stores facts in the global memory system. When connected to an agent, the agent can persist key-value pairs that survive across executions and are accessible to any workflow on the platform.

Property Value
Component Type memory_write
Category Sub-component (Memory)
Display Name Memory Write

Ports

Inputs

This component has no wired inputs. It operates as a LangChain tool that the agent invokes with arguments during its reasoning loop.

Outputs

Port Data Type Description
result STRING Confirmation of what was stored

Configuration

The Memory Write component accepts the following configuration in extra_config:

Field Type Default Options Description
fact_type string world_knowledge user_preference, world_knowledge, self_knowledge, correction, relationship Type classification for the stored fact
overwrite boolean true -- Whether to overwrite an existing fact with the same key

Fact types

Fact Type Description Example
user_preference User-specific preferences and settings Preferred language, notification settings
world_knowledge General facts about the world API endpoints, project names
self_knowledge Facts the agent knows about itself Its own capabilities, assigned roles
correction Corrections to previously stored facts Updated values after user feedback
relationship Relationships between entities "Alice manages the backend team"

Usage

  1. Add a Memory Write node from the Node Palette (Memory category)
  2. Connect it to an agent node via the amber diamond memory handle
  3. Optionally configure the default fact type and overwrite behavior in the node's Extra Config

The agent's LLM will then have a remember tool available. The tool accepts three parameters:

Parameter Required Description
key Yes The identifier for the fact
value Yes The content to store
fact_type No Override the configured default fact type

Write behavior

When the remember tool is called:

  1. If a fact with the same key already exists and overwrite is true, the value is updated and the confirmation count is incremented
  2. If a fact with the same key exists and overwrite is false, the write is skipped
  3. If no fact with that key exists, a new fact is created

Return format

The tool returns a confirmation string:

Remembered: user_name = Alice (created)
Remembered: user_name = Alice Smith (updated)
Remembered: user_name = Alice (skipped)

Example

An agent with Memory Write connected can store information for later retrieval:

User: My name is Alice and I prefer dark mode.
Agent: [calls remember(key="user_name", value="Alice", fact_type="user_preference")]
       → Remembered: user_name = Alice (created)
       [calls remember(key="ui_preference", value="dark mode", fact_type="user_preference")]
       → Remembered: ui_preference = dark mode (created)
Agent: Got it! I've noted your name and preference.

flowchart LR
    T[Chat Trigger] --> A[Agent]
    M[AI Model] -.->|model| A
    MW[Memory Write] -.->|memory| A

    style MW fill:#f59e0b,color:white

Pair with Memory Read

Memory Write is most useful when paired with Memory Read on the same agent. The agent can then both store and recall information, building persistent knowledge over time.

Global scope

All facts are stored in the global memory scope. Any workflow on the platform can read facts written by this component. Be mindful of key naming to avoid unintended collisions across workflows.