Memory Node

Enable persistent knowledge with conversation history and knowledge bases

Memory Node

Memory gives your agent persistent knowledge and long-term context.

---

Overview

Purpose: Store knowledge, conversation history, and reference information Connection: Bottom-left handle of main agent card Color: Green Icon: 🧠 Required: Optional (0 to many)

---

What is Memory?

Memory nodes contain markdown files that provide your agent with:

• Domain knowledge
• Company information
• Conversation context
• Reference documentation
• Historical data

---

Memory Structure

Memory is organized as markdown files in folders:

memory-layer/
├── company-info.md
├── team-contacts.md
├── product-specs.md
└── faq.md

Each file contains knowledge in plain markdown format.

---

Memory Types

1. Knowledge Base

Purpose: Domain-specific information

Example: Company Knowledge Base

# TechCorp Company Information

## About Us
TechCorp is a software company specializing in...

## Products
- Product A: Cloud-based solution for...
- Product B: On-premise software for...

## Support Hours
Monday-Friday: 9am-6pm EST

Use case: Customer support agents with company context

---

2. Prime Context

Purpose: Always-available context included in every interaction

Example: Agent Guidelines

# Agent Operating Guidelines

## Communication Style
- Professional but friendly
- Concise responses
- Always cite sources

## Restrictions
- Never share customer data
- Don't make promises about features
- Escalate billing issues to finance

Use case: Ensure consistent agent behavior

---

3. Conversation History

Purpose: Maintain context across sessions

Example: Previous Interactions

# Customer: John Doe (ID: 12345)

## 2025-01-05 - Password Reset
Customer requested password reset. Issue resolved.

## 2025-01-08 - Billing Question
Asked about invoice #5678. Referred to finance team.

Use case: Personalized multi-session support

---

Creating Memory

Via Settings → Node Managers → Memory:

1. Click “Add New Memory Layer”
2. Enter name (e.g., “Company KB”)
3. Add markdown files
4. Write content in each file
5. Save to workspace library

Memory layer appears in sidebar → drag to canvas → connect to main agent.

---

Configuration

When clicked on canvas:

• Add/edit markdown files
• Organize into subfolders
• Set priority (which files load first)
• Configure memory scope (always vs. contextual)

---

Real-World Examples

Example 1: Customer Support Knowledge Base

Structure:

customer-support-kb/
├── products.md          # Product catalog
├── policies.md          # Company policies
├── troubleshooting.md   # Common issues
└── contacts.md          # Team contacts

products.md:

# Product Catalog

## TechCorp Cloud
- Pricing: $99/month
- Features: Unlimited storage, 24/7 support
- Trial: 14-day free trial

## TechCorp Enterprise
- Pricing: Custom
- Features: On-premise, dedicated support
- Contact sales for pricing

---

Example 2: Code Review Style Guide

Structure:

code-review-guide/
├── style-guide.md       # Coding standards
├── security-checklist.md # Security best practices
└── examples.md          # Good/bad examples

style-guide.md:

# Python Style Guide

## Naming Conventions
- Functions: snake_case
- Classes: PascalCase
- Constants: UPPER_SNAKE_CASE

## Imports
- Standard library first
- Third-party second
- Local imports last

## Example
\`\`\`python
import os
import sys

import requests

from .utils import helper
\`\`\`

---

Memory Patterns

Pattern 1: Layered Knowledge

[Main Agent]
     ↓
[General KB] + [Specialized KB] + [User History]

Use case: Agent with broad knowledge + domain expertise + personalization

---

Pattern 2: RAG (Retrieval-Augmented Generation)

Concept: Agent retrieves relevant memory chunks based on user query

Implementation:

1. Store large knowledge base as memory
2. Agent searches relevant sections
3. Agent uses retrieved context for response

Use case: Documentation Q&A, technical support

---

Best Practices

✅ Do:

• Use clear markdown headings
• Keep files focused (one topic per file)
• Update regularly
• Include examples and specifics
• Organize with folders

❌ Don’t:

• Store huge files (split into smaller chunks)
• Include outdated information
• Use complex formatting (keep simple)
• Duplicate information across files

---

Memory vs. System Prompt

Feature	System Prompt	Memory
When to use	Instructions, behavior	Facts, knowledge
Example	”You are helpful"	"Company founded 2020”
Changeable	Rarely	Frequently
Scope	Every interaction	Contextual retrieval

Rule: Instructions in prompt, facts in memory.

---

Next Steps

• Commands Node - Create slash commands
• MCPs Node - External integrations

---

← Skills Node | Agent Node | Commands Node →