Class LocalVectorStore

Namespace

Glitch9.AIDevKit.Agents.Memory

Simple on-disk vector store (cosine similarity)'. Each item is two files:

• {id}.vec : raw float32 array (dimension = fixed)
• {id}.json : payload json (arbitrary metadata; role, content, timestamps, etc.)

Options:

• preloadVectors: load all vectors into memory on startup (fast search, higher RAM)
• normalizeOnWrite: store L2-normalized vectors to speed cosine (dot == cosine)

public sealed class LocalVectorStore : IVectorStore

Inheritance

object

LocalVectorStore

Implements

IVectorStore

Constructors

LocalVectorStore(string, int, bool, bool)

Creates a new local vector store at the specified root directory. If the directory does not exist, it will be created.

public LocalVectorStore(string id, int dimension = 1536, bool preloadVectors = false, bool normalizeOnWrite = true)

Parameters

id string

The unique identifier for the vector store. Use AI Agent's UniqueId.

dimension int

The dimension of a vector refers to the number of components or features that define it. This is a fixed value for the entire vector store and cannot be changed later. Each embedding AI model has a specific output dimension (e.g. 1536 for OpenAI's text-embedding-3-small), and this must match the dimension of the vectors being stored.

For Gemini models (e.g., gemini-embedding-001), the dimension can freely be set in the request (config.outputDimensionality) unlike OpenAI models that have fixed dimensions. we recommend using 1536 or 3072 for best performance and cost balance.

preloadVectors bool

Whether to preload all vectors into memory on startup.

normalizeOnWrite bool

Whether to store L2-normalized vectors to speed up cosine similarity calculations.

Exceptions

ArgumentOutOfRangeException

ArgumentNullException

Properties

Count

public int Count { get; }

Property Value

int

Dimension

public int Dimension { get; }

Property Value

int

Methods

DeleteAsync(string)

public UniTask<bool> DeleteAsync(string id)

Parameters

id string

Returns

UniTask<bool>

ExistsAsync(string)

public UniTask<bool> ExistsAsync(string id)

Parameters

id string

Returns

UniTask<bool>

SearchAsync(float[], int, float)

public UniTask<IReadOnlyList<SearchHit>> SearchAsync(float[] query, int k, float minSim = 0.2)

Parameters

query float[]

k int

minSim float

Returns

UniTask<IReadOnlyList<SearchHit>>

UpsertAsync(string, float[], object)

public UniTask UpsertAsync(string id, float[] vec, object payload)

Parameters

id string

vec float[]

payload object

Returns

UniTask