A Comprehensive Guide to WilmerAI Workflow Memory Specific Nodes

A Comprehensive Guide to WilmerAI Workflow Memory Specific Nodes

This document provides a comprehensive catalog of memory-related nodes for use within the WilmerAI workflow system. Each entry includes a high-level overview, a complete JSON example with all available fields, and a brief description of each field and the node's function.

QualityMemory

The QualityMemory node is the primary creator node for persistent memory. It analyzes the recent conversation and, if enough new content exists, generates and saves a new memory chunk. It produces no direct output, as its sole purpose is to write to memory in the background. Its behavior (creating a vector memory vs. a file-based memory) is controlled by the discussion's configuration settings.

Complete JSON Example

{
  "title": "Update Memories with Latest Turn",
  "type": "QualityMemory"
}

Field Rundown

• title: (Optional) A descriptive name for the node used in logging.
• type: (Required) Must be "QualityMemory".

Actions & Output

• Action: This node writes to memory. Depending on the useVectorForQualityMemory flag in the discussion's configuration, it will either:
  • Append a new structured memory to the vector database (<id>_vector_memory.db).
  • Append a new summarized memory chunk to the long-term memory file (<id>_memories.jsonl).
• Output: This node produces no output. Its {agent#Output} variable will be empty.

VectorMemorySearch

The VectorMemorySearch node is the primary retriever for Retrieval-Augmented Generation (RAG). It performs a relevance-based keyword search against the discussion's vector memory database (_vector_memory.db) to find specific facts or details from the conversation's history. This node requires an active discussionId to function.

Complete JSON Example

{
  "title": "Search for Relevant Facts",
  "type": "VectorMemorySearch",
  "input": "Project Stardust;mission parameters;Dr. Evelyn Reed",
  "limit": 5
}

Field Rundown

• title: (Optional) A descriptive name for the node.
• type: (Required) Must be "VectorMemorySearch".
• input: (Required) A string of keywords to search for. Keywords must be separated by a semicolon (;). Supports variables.
• limit: (Optional) The maximum number of memory results to return. Defaults to 5.

Actions & Output

• Action: This node reads from the vector database (<id>_vector_memory.db).
• Output: Returns a single string containing the text of the most relevant memories, separated by \n\n---\n\n. If no memories are found, it returns a message stating so.

RecentMemorySummarizerTool

The RecentMemorySummarizerTool is a fast retriever node that fetches the most recent memory chunks from the long-term memory file (_memories.jsonl). It is ideal for giving an LLM a quick summary of recent events without performing a complex search. It can also operate in a stateless mode if no discussionId is present.

Complete JSON Example

{
  "title": "Get a summary of the last 3 major events",
  "type": "RecentMemorySummarizerTool",
  "maxSummaryChunksFromFile": 3,
  "maxTurnsToPull": 5,
  "customDelimiter": "\n\n---\n\n",
  "lookbackStart": 0
}

Field Rundown

• title: (Optional) A descriptive name for the node.
• type: (Required) Must be "RecentMemorySummarizerTool".
• maxSummaryChunksFromFile: (Required) In stateful mode (with discussionId), the number of recent memory chunks to retrieve from the file.
• maxTurnsToPull: (Required) In stateless mode (no discussionId), the number of recent conversation turns to pull from the chat history.
• customDelimiter: (Optional) A string used to separate the retrieved memory chunks. Defaults to "--ChunkBreak--".
• lookbackStart: (Optional) The number of turns to skip from the end of the conversation before pulling content. Defaults to 0.

Actions & Output

• Action: This node reads from the long-term memory file (<id>_memories.jsonl).
• Output: Returns a single string containing the text of the requested recent memories.

GetCurrentSummaryFromFile

The GetCurrentSummaryFromFile node is a simple and extremely fast retriever. Its only job is to read the entire contents of the rolling chat summary file (_summary.jsonl) and return it as a string. It performs no checks and triggers no updates.

Complete JSON Example

{
  "title": "Quickly grab the current conversation summary",
  "type": "GetCurrentSummaryFromFile"
}

Field Rundown

• title: (Optional) A descriptive name for the node.
• type: (Required) Must be "GetCurrentSummaryFromFile".

Actions & Output

• Action: This node reads from the rolling chat summary file (<id>_summary.jsonl).
• Output: Returns a single string containing the full text of the current chat summary.

FullChatSummary

The FullChatSummary is a combined creator and retriever node. By default, it first ensures the file-based memories (_memories.jsonl) are up-to-date, then checks if the rolling summary (_summary.jsonl) is stale and updates it if needed, and finally returns the summary's content. This process can be slow. Setting isManualConfig to true disables the creation/update logic, turning it into a fast "read-only" retriever.

Complete JSON Example

{
  "title": "Update and Get Full Chat Summary",
  "type": "FullChatSummary",
  "isManualConfig": false
}

Field Rundown

• title: (Optional) A descriptive name for the node.
• type: (Required) Must be "FullChatSummary".
• isManualConfig: (Optional) If true, disables the slow update logic and makes the node a fast, direct reader. Defaults to false.

Actions & Output

• Action: When isManualConfig is false, this node can write to both the long-term memory file ( <id>_memories.jsonl) and the rolling summary file (<id>_summary.jsonl). It always reads from the summary file.
• Output: Returns a single string containing the full text of the chat summary.

chatSummarySummarizer

The chatSummarySummarizer is a low-level creator node that generates an updated rolling chat summary. It is designed to take an existing summary and a batch of new memory chunks, and use an LLM to integrate them into a new, cohesive summary. It uses two special placeholders, [CHAT_SUMMARY] and [LATEST_MEMORIES], in its prompts.

Complete JSON Example

{
  "title": "Update the Rolling Conversation Summary",
  "type": "chatSummarySummarizer",
  "minMemoriesPerSummary": 2,
  "loopIfMemoriesExceed": 5,
  "systemPrompt": "You are a summarization AI. Your task is to seamlessly integrate new conversation memories into the existing summary.",
  "prompt": "EXISTING SUMMARY:\n[CHAT_SUMMARY]\n\nNEW MEMORIES TO INTEGRATE:\n[LATEST_MEMORIES]\n\nPRODUCE THE NEW, UPDATED SUMMARY:",
  "endpointName": "Text-Processing-Endpoint",
  "preset": "Summarizer_Preset"
}

Field Rundown

• title: (Optional) A descriptive name for the node.
• type: (Required) Must be "chatSummarySummarizer".
• minMemoriesPerSummary: (Optional) The minimum number of new memories required to trigger an update. Defaults to 3.
• loopIfMemoriesExceed: (Optional) The batch size for processing new memories in a loop. Defaults to 3.
• systemPrompt / prompt: (Required) Prompts for the summarization LLM. Must contain the [CHAT_SUMMARY] and [LATEST_MEMORIES] placeholders.
• endpointName: (Optional) The LLM endpoint to use for summarization. Supports LIMITED variables: only {agent#Input} and static workflow variables, NOT {agent#Output}.
• preset: (Optional) The generation preset to use. Supports LIMITED variables like endpointName.

Actions & Output

• Action: This node generates the text for a new summary. It does not write to a file itself.
• Output: Returns a single string containing the newly generated summary text.

RecentMemory (Legacy)

The RecentMemory node is a legacy, dual-function node that combines creation and retrieval. It first triggers the slow, blocking process of creating new file-based memories and then immediately retrieves the most recent ones. This node is inefficient and not recommended for new workflows. Use QualityMemory and RecentMemorySummarizerTool separately instead.

Complete JSON Example

{
  "title": "Update and Get Recent Memories",
  "type": "RecentMemory"
}

Field Rundown

• title: (Optional) A descriptive name for the node.
• type: (Required) Must be "RecentMemory".

Actions & Output

• Action: This node writes to the long-term memory file (<id>_memories.jsonl) and then reads from it in the same step.
• Output: Returns a single string containing the text of the most recent memories.

ChatSummaryMemoryGatheringTool (Legacy)

The ChatSummaryMemoryGatheringTool is a specialized legacy retriever. Its purpose is to gather all new memory chunks from the long-term memory file (_memories.jsonl) that have been created since the last rolling summary was generated. It's intended to be the first step in a manual summary update workflow.

Complete JSON Example

{
  "title": "Gather all new memory chunks since last summary",
  "type": "ChatSummaryMemoryGatheringTool",
  "maxTurnsToPull": 20
}

Field Rundown

• title: (Optional) A descriptive name for the node.
• type: (Required) Must be "ChatSummaryMemoryGatheringTool".
• maxTurnsToPull: (Required) The number of recent turns to pull if running in stateless mode (no discussionId).

Actions & Output

• Action: This node reads from both the long-term memory file (<id>_memories.jsonl) and the summary file ( <id>_summary.jsonl) to determine which memories are new.
• Output: Returns a single string containing the text of all new, unsummarized memory chunks. Returns an empty string if no new memories exist.

ConversationMemory (Legacy)

The ConversationMemory node is a legacy, dual-function node that runs a hardcoded internal sub-workflow for memory creation and retrieval. It is inflexible, inefficient, and not recommended for new workflows.

Complete JSON Example

{
  "title": "Run legacy conversation memory process",
  "type": "ConversationMemory"
}

Field Rundown

• title: (Optional) A descriptive name for the node.
• type: (Required) Must be "ConversationMemory".

Actions & Output

• Action: This node performs both write and read operations on memory files as part of its internal, non-configurable process.
• Output: Returns a string containing generated memories.