Embedding Modes

OpenMemory offers two primary embedding modes to balance performance and cognitive depth.

1. Simple Mode (Fast)

In Simple Mode, OpenMemory behaves like a traditional vector database. It generates a single embedding vector for the entire memory content.

• Pros: Fast, low storage usage.
• Cons: Less nuance.
• Use Case: Simple search, high-throughput logs.

const mem = new OpenMemory({
  tier: "fast" // Enables Simple Mode
});

2. Advanced Mode (HMD v2)

In Advanced Mode, OpenMemory implements the Holistic Memory Descriptor v2 specification. It generates 5 separate embeddings for each memory, corresponding to different cognitive sectors:

1. Episodic: Events and timeline.
2. Semantic: Facts and knowledge.
3. Procedural: Instructions and how-to.
4. Emotional: Sentiment and tone.
5. Reflective: Meta-analysis.

• Pros: Extremely accurate, context-aware retrieval.
• Cons: Slower ingestion (5x embedding calls), higher storage.
• Use Case: AI Agents, Personal Assistants, Complex RAG.

const mem = new OpenMemory({
  tier: "deep" // Enables Advanced Mode
});

Hybrid Mode (Smart)

The default "Smart" tier uses a hybrid approach. It generates a primary embedding and selectively generates sector embeddings based on content analysis.

const mem = new OpenMemory({
  tier: "smart" // Default
});