Memory

​

Overview

• Persistent Context: Remember key details about users between conversations
• Personalized Experiences: Provide more relevant and contextual responses based on conversation history
• Reduced Repetition: Avoid asking users to re-explain their background or preferences
• Enhanced User Experience: Create more natural, human-like interactions that build on previous conversations

​

How Memory Works

​

1. Automatic Summary Generation

• Key User Context: Important background information about the user
• Preferences & Requirements: User preferences, needs, and specific requirements mentioned
• Previous Solutions: What worked or didn’t work in past interactions
• Follow-up Items: Any pending tasks or future discussion topics

​

2. Storage & Organization

• Linked to Users: Associated with the user’s external ID for easy retrieval
• Timestamped: Tracked with creation dates for temporal context
• Thread-Associated: Connected to specific conversation threads for traceability

​

3. Retrieval & Usage

• API Integration: Programmatically retrieve memory summaries using the Memory API
• Dashboard View: Review memory summaries in the Melodi web interface
• Agent Integration: Incorporate memory into your AI agent’s context for personalized responses

​

Setting Up Memory

​

Prerequisites

1. Active Threads: Conversations logged through the Create Thread API
2. User Identification: Users created with external IDs via the Create User API
3. Complete Conversations: Threads with meaningful conversation content for the AI to summarize

​

Enabling Memory (Beta)

​

Using the Memory API

​

Retrieving User Memory

curl -X GET "https://app.melodi.fyi/api/external/memory?userExternalId=user-123" \
  -H "Authorization: Bearer YOUR_PRIVATE_API_KEY"

​

Example Response

{
  "userExternalId": "user-123",
  "threadId": 456,
  "threadExternalId": "support-session-789",
  "summary": "User is implementing OAuth authentication for their e-commerce platform. They're using Node.js with Express and need help with session management. Previous discussion covered JWT token security and they decided to use refresh tokens for better security.",
  "createdAt": "2024-01-15T10:30:45.123Z"
}

​

Integrating Memory into Your Agent

import requests
from melodi import MelodiClient

def get_user_context(user_external_id):
    """Retrieve user memory to provide context for new conversations"""
    memory_response = requests.get(
        f"https://app.melodi.fyi/api/external/memory?userExternalId={user_external_id}",
        headers={"Authorization": f"Bearer {MELODI_API_KEY}"}
    )
    
    if memory_response.status_code == 200:
        memory_data = memory_response.json()
        return f"Previous conversation context: {memory_data['summary']}"
    
    return "No previous conversation history available."

def create_personalized_prompt(user_input, user_external_id):
    """Create an AI prompt that includes user memory for context"""
    context = get_user_context(user_external_id)
    
    prompt = f"""
    {context}
    
    Current user message: {user_input}
    
    Please provide a helpful response that takes into account the user's previous conversation history.
    """
    
    return prompt

​

Best Practices

​

1. Memory-Aware Conversation Design

• Reference Previous Context: When memory is available, acknowledge previous conversations naturally
• Progressive Enhancement: Design your agent to work both with and without memory context
• Privacy Considerations: Be transparent about what information is remembered

​

2. API Integration Patterns

• Check for Memory: Always attempt to retrieve memory before starting new conversations
• Graceful Fallbacks: Handle cases where no memory is available (new users, privacy settings)
• Memory Updates: Memory is automatically updated as new conversations occur

​

3. User Experience

• Transparency: Let users know when you’re using information from previous conversations
• Control: Consider providing users with ways to view or clear their memory if needed
• Contextual Relevance: Use memory to enhance conversations, not overwhelm them with old information

​

Memory Quality & Accuracy

​

What Makes Good Memory

• Actionable Information: Details that help improve future interactions
• User Preferences: Consistent patterns in user behavior and choices
• Technical Context: Relevant technical details for support or development conversations
• Relationship Building: Information that helps build rapport and trust

​

Limitations During Beta

• Processing Time: Memory summaries may take a few minutes to hours to generate after conversation completion
• Summary Length: Currently optimized for concise summaries; very long conversations may have details condensed
• Language Support: Primary optimization for English conversations
• Update Frequency: Memory reflects the most recent thread only; older conversations aren’t automatically merged

​

Getting Help

• Feature Requests: Share ideas for how memory could better serve your use case
• Quality Issues: Report problems with memory accuracy or relevance
• Integration Support: Get help integrating memory into your specific workflow

​

Related Features

• Session Outcome: Track how memory affects conversation quality
• User Intents: Understand what users are trying to accomplish across sessions
• Get Memory API: Technical documentation for the Memory API
• Create Thread API: How to create conversations that generate memory