Vector Search

The retrieval.vectorSearch indexes the input files using embeddings into a vector database that can be used for similarity search. This is commonly referred to as Retrieval Augmented Generation (RAG).

const files = await retrieval.vectorSearch("keyword", env.files)

The returned value is an array of files with the resconstructed content from the matching chunks.

const files = await retrieval.vectorSearch("keyword", env.files)
def("FILE", files)

Model configuration

The computation of embeddings is done through the LLM APIs using the same authorization token as the LLM API.

The default model is openai:text-embedding-ada-002 but you can override the model using embedModel.

const files = await retrieval.vectorSearch(
    "keyword",
    env.files, {
        embedModel: "ollama:all-minilm"
    })

You can further customize the embedding generation by using chunkSize and chunkOverlap.

Index name

If you modify the model or chunking configurations, you will want to create separate index databases.

const files = await retrieval.vectorSearch(
    "keyword",
    env.files, {
        indexName: "all-minilm",
        embedModel: "ollama:all-minilm"
    })

Installation requirements

The retrieval uses LLamaindex TS for indexing and searching.

The llamaindex package will be automatically installed.