Embeddings

Configure semantic vectors for topic detection and memory retrieval.

Role

Embeddings power drift detection, semantic recall, topic matching, and graph expansion. All stored and queried vectors must use compatible dimensions and semantics.

Configuration

Choose one embedding model per compatible index.

Record model and dimension changes as a migration concern.

Batch document ingestion when the provider supports it.

Evaluate similarity thresholds with representative queries.

Custom implementation

embedder.ts

class MyEmbedder implements EmbedAdapter {
  async embed(text: string): Promise<number[]> {
    return provider.embed(text);
  }
}