npm install @magneteco/client

import { MagnetoClient } from '@magneteco/client';

const memory = new MagnetoClient({
  appId: 'my-app',
  baseUrl: process.env.MAGNETECO_URL,
  apiKey: process.env.MAGNETECO_API_KEY,
});

await memory.memorize({
  userId: 'user-123',
  content: conversationText,
  contentType: 'conversation',
});

const context = await memory.retrieve({
  userId: 'user-123',
  query: userMessage,
  maxTokens: 2000,
});

// Inject into your agent's prompt
const prompt = `${memory.formatForPrompt(context)}\n\nUser: ${userMessage}`;

await memory.trackEvent({
  userId: 'user-123',
  eventType: 'project_completed',
  data: { projectId: 'proj-456', projectName: 'Website Redesign' },
  importance: 'high',
});

Input (conversation/event)
    ↓
Store raw resource (S3)
    ↓
Extract atomic facts (LLM)
    ↓
Classify by category
    ↓
For each category:
    - Load existing summary
    - Evolve summary with new facts (LLM handles conflicts)
    - Save updated summary
    ↓
Extract entities → Neo4j nodes
Extract relationships → Neo4j edges (with conflict detection)

Query
    ↓
Select relevant categories (LLM)
    ↓
Load category summaries
    ↓
Sufficient? → Return
    ↓
Vector search for specific items
    ↓
Graph traversal for relationships
    ↓
Apply time decay ranking
    ↓
Assemble within token budget