Memory¶

Memory components give agents persistent knowledge that survives across executions. They connect to agents as sub-component tools via the amber diamond memory handle, allowing an agent to read from, write to, and personalize interactions using a global memory system.

Overview¶

There are three memory component types:

How memory components work¶

Memory Read and Memory Write are sub-component tools -- they connect to agent nodes via the amber diamond memory handle and become available as LangChain tools inside the agent's reasoning loop. The agent decides when to recall or remember information based on the conversation.

Identify User is different -- it is a direct node in the workflow graph that runs before the agent, extracting user identity from trigger payloads and loading personalized context into the workflow state.

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

Memory types¶

The global memory system stores three types of knowledge:

Memory Read can search across one type or all types simultaneously. Memory Write stores facts with configurable type classification.

Memory scope¶

All memory operations use a global scope by default. Facts written by one workflow are readable by any other workflow on the same platform instance. This enables shared knowledge across agents -- for example, one agent can learn a user's preferences while another agent uses that knowledge to personalize responses.

Connecting memory to agents¶

1. Add a Memory Read and/or Memory Write node from the Node Palette
2. Connect them to an agent node via the amber diamond memory handle at the bottom
3. Optionally add an Identify User node before the agent to load user-specific context

The agent's LLM will then have recall and remember tools available for reading and writing memory during its reasoning loop.