Categories

Auto-Categorization

When you use smart store, the LLM automatically assigns a category to each extracted fact. Categories help organize memories and enable filtered recall.

Category List

MemLib uses 9 built-in categories:

Using Categories

Filter Recall by Category

Narrow your search to a specific type of information:

// Only recall preferences
const prefs = await mem.recall({
  query: "development setup",
  category: "preference",
});

// Only recall health-related memories
const health = await mem.recall({
  query: "dietary restrictions",
  category: "health",
});

Filter Prepare by Category

When synthesizing context, focus on specific memory types:

const ctx = await mem.prepare({
  messages: [{ role: "user", content: "Let's plan dinner" }],
  category: "health",  // priority on allergies and dietary info
});

Categories in Results

Every recalled or stored memory includes its assigned category:

const result = await mem.store({
  content: "I'm allergic to shellfish and prefer vegetarian food",
});

// result.memories = [
//   { content: "Allergic to shellfish", category: "health", importance: 0.95 },
//   { content: "Prefers vegetarian food", category: "preference", importance: 0.7 },
// ]

Consolidation by Category

When memories are consolidated, they are grouped by (namespace, entity, category). This prevents unrelated facts from being merged:

• Health memories consolidate together → "Has several dietary restrictions: allergic to peanuts, prefers vegetarian..."
• Professional memories consolidate separately → "Software engineer who uses TypeScript, React, and Next.js..."

Notes

• Categories are automatically assigned by the LLM during fact extraction — you don't need to specify them.
• You cannot override the category during store — it's always inferred from the content.
• Categories are stored as plain text strings, so future versions may extend the list.
• The other category is a catch-all for information that doesn't fit the predefined types.