mnemosine.vec.VectorStore.knn

Kind: method

Signature

VectorStore.knn(self, vector, top_k: int=10, metric: str='cosine')

Documentation

Return the top_k nearest embeddings to vector.

Native path: vec0 k-NN query returning (node_id, distance). Fallback path: brute-force scan with _distance. Result items are ordered nearest-first.

Args:

• vector (sequence of float): The query embedding.
• top_k (int): Number of results. Defaults to 10.
• metric (str): "cosine" (default), "l2" or "dot".

Returns:

• list of dict: Each item has node_id (str) and distance
• (float). Empty when no embeddings are stored.

Raises:

• VectorError: If top_k < 1 or the vector dimension conflicts.

Example:

>>> db.vec.knn([1.0, 0.0, 0.0], top_k=2)
[{'node_id': '01G...', 'distance': 0.0}, ...]