Expand description
Vector Store / Embedding Index — FR-013
GPU-friendly vector similarity search for hybrid neural-symbolic queries. Stores embeddings alongside graph nodes and supports nearest-neighbor search.
§Distance Metrics
- Cosine similarity
- L2 (Euclidean) distance
- Dot product (inner product)
§Index Types
- BruteForce: Exact search, O(n·d). Good for < 100K vectors.
- IVF: Inverted file index for approximate search. Good for 100K-10M vectors.
For GPU-accelerated search, the index data can be copied to device memory and the search kernel processes queries in parallel.
Structs§
- Search
Result - Result of a nearest-neighbor query.
- Vector
Entry - A vector with an associated ID and optional metadata.
- Vector
Store - Brute-force vector store (exact nearest-neighbor search).
- Vector
Store Config - Configuration for the vector store.
Enums§
- Distance
Metric - Distance metric for vector similarity.
- Vector
Store Error - Vector store errors.