@nodetool-ai/vectorstore

SQLite-vec vector store for NodeTool — embeddings, collections, and semantic search for RAG.

A backend-agnostic vector provider abstraction with a local SQLite-vec implementation (plus Pinecone and Supabase providers), document chunking, and embedding functions for OpenAI, Ollama, Gemini, Mistral, Cohere, Voyage, and Jina.

Install

npm install @nodetool-ai/vectorstore

Exported symbols

| Symbol | Kind | Description | |---|---|---| | VectorProvider, VectorCollection | interface | Backend-agnostic vector store contract | | SqliteVecProvider | class | Local SQLite-vec provider | | PineconeProvider | class | Pinecone-backed provider | | SupabaseProvider | class | Supabase (pgvector) provider | | createVectorProviderFromEnv, getDefaultVectorProvider, setDefaultVectorProvider | function | Resolve / configure the default provider | | createSqliteVecProvider | function | Construct a local provider directly | | resolveCollection | function | Get a collection through the default provider, inferring its embedding function | | getProviderEmbeddingFunction | function | Build an embedding function for a model / provider | | OpenAIEmbeddingFunction, OllamaEmbeddingFunction, GeminiEmbeddingFunction, MistralEmbeddingFunction, CohereEmbeddingFunction, VoyageEmbeddingFunction, JinaEmbeddingFunction | class | Provider-specific embedding functions | | splitDocument | function | Chunk a document into overlapping text segments | | SqliteVecStore, VecCollection, getDefaultStore | class / function | Low-level SQLite-vec handle (prefer VectorProvider) | | VectorRecord, VectorMatch, VectorQuery, DistanceMetric, RecordMetadata | type | Query and record value types | | CollectionNotFoundError, ProviderConfigError, UnsupportedFilterError | class | Error types |

Usage

import {
  getDefaultVectorProvider,
  getProviderEmbeddingFunction,
  splitDocument
} from "@nodetool-ai/vectorstore";

const provider = getDefaultVectorProvider();
const ef = getProviderEmbeddingFunction("text-embedding-3-small", "openai");

const collection = await provider.getOrCreateCollection({
  name: "docs",
  embeddingFunction: ef,
  metadata: { embedding_model: "text-embedding-3-small", embedding_provider: "openai" }
});

const chunks = splitDocument("...long document...", "doc-1", 1000, 200);
await collection.upsert(
  chunks.map((c, i) => ({ id: `doc-1:${i}`, document: c.text }))
);

const results = await collection.query({ text: "what is nodetool?", topK: 5 });

The default provider is chosen from the environment (createVectorProviderFromEnv), defaulting to local SQLite-vec at VECTORSTORE_DB_PATH. When a collection stores embedding_model / embedding_provider metadata, resolveCollection reattaches the right embedding function for you.

Links

• NodeTool
• GitHub