Usage Patterns Overview

DeltaMemory supports multiple integration patterns to fit different application architectures and use cases. Understanding these patterns helps you choose the right approach for your AI application.

The Two Patterns

1. Automatic Mode

Best for: Simple chatbots, assistants with consistent memory needs

Memory operations happen transparently on every interaction. The framework automatically recalls relevant context before each LLM call and ingests the conversation afterward.

// Memory is handled automatically
const response = await generateText({
  model: deltaMemoryModel('gpt-4', { userId: 'user-123' }),
  prompt: 'What are my preferences?'
});

Pros:

• Zero friction integration
• No agent decision-making required
• Consistent memory behavior

Cons:

• Recalls on every turn (even when unnecessary)
• Higher latency and cost
• Less control over memory operations

2. Agent-Controlled Tools (Recommended)

Best for: Complex agents, multi-step workflows, cost-sensitive applications, production use

Memory operations are exposed as tools that the agent decides when to use based on the conversation context.

const response = await generateText({
  model: openai('gpt-4'),
  tools: {
    ...deltaMemoryTools(client, { userId: 'user-123' })
  },
  prompt: 'What are my preferences?'
});

Pros:

• Agent decides when memory is needed
• More efficient (fewer unnecessary recalls)
• Better for complex workflows
• Full control when needed

Cons:

• Agent might forget to use tools
• Requires good tool descriptions
• Slightly more complex setup

Choosing the Right Pattern

Pattern Comparison

// AUTOMATIC MODE
// ✓ Simple
// ✗ Always recalls
const response = await generateText({
  model: deltaMemoryModel('gpt-4', { userId }),
  prompt: message
});
 
// AGENT-CONTROLLED TOOLS (Recommended)
// ✓ Efficient
// ✓ Agent decides
// ✓ Full control
const client = new DeltaMemory({ 
  apiKey: process.env.DELTAMEMORY_API_KEY,
  baseUrl: process.env.DELTAMEMORY_URL
});
const response = await generateText({
  model: openai('gpt-4'),
  tools: { ...deltaMemoryTools(client, { userId }) },
  prompt: message
});

Next Steps

Explore each pattern in detail: