Module vector_store 

Vector Memory System

Semantic vector storage for code search and memory. Local-first design - no external server required.

Uses brute-force cosine similarity search, which is efficient for collections up to ~100k vectors. For larger collections, consider integrating an HNSW library like hnsw_rs or instant-distance.

Features:

• Code chunking strategies (functions, structs, modules)
• Embedding generation interface (pluggable backends)
• Similarity search with filters
• Collection management (project, session, global)
• Persistence to disk

Structs

BoundedVectorStore

A capacity-limited wrapper around VectorStore that evicts the oldest items (FIFO order) when the total number of chunks exceeds max_items.

ChunkMetadata

Metadata for a code chunk

CodeChunk

A chunk of code with its embedding

CodeChunker

Code chunker for splitting code into meaningful pieces

CollectionStats

Statistics for a collection

MockEmbeddingProvider

Mock embedding provider for testing

SearchFilter

Filter for search queries

SearchResult

Search result with similarity score

TfIdfEmbeddingProvider

Simple TF-IDF based embedding provider (no external dependencies)

VectorCollection

Vector collection

VectorIndex

Vector index using simple brute-force search

VectorStore

Main vector store

VectorStoreStats

Statistics for vector store

Enums

ChunkType

Chunk types for code organization

CollectionScope

Collection scope for organizing chunks

EmbeddingBackend

Enum dispatch for embedding providers.

IndexHealth

Health status of a vector index

Constants

DEFAULT_MAX_ITEMS

Default maximum number of items across all collections in a bounded store.

EMBEDDING_DIM

Embedding dimension (common for small models)

MAX_CHUNKS

Maximum chunks per collection

MAX_VOCABULARY_SIZE

Maximum vocabulary size for TF-IDF provider before eviction occurs

Traits

EmbeddingProvider

Trait for embedding generation.