Struct Victor

pub struct Victor<D> { /* private fields */ }

The main database struct. Through this you can Victor::add and Victor::search for embeddings.

Implementations

impl<D: DirectoryHandle> Victor<D>

pub fn new(root: impl Into<D>) -> Self

Create a new Victor database given a directory handle.

For example, you can use std::path::PathBuf to use the native filesystem. Or you can use crate::memory::DirectoryHandle to use an in-memory database.

pub async fn add( &mut self, content: Vec<impl Into<String>>, tags: Vec<impl Into<String>>, )

Add many documents to the database. Embeddings will be generated for each document.

victor
    .add(
        vec!["Pineapple", "Rocks"], // documents
        vec!["Pizza Toppings"],     // tags (only used for filtering)
    )
    .await;

pub async fn add_single( &mut self, content: impl Into<String>, tags: Vec<impl Into<String>>, )

Add a single document to the database. Embedding will be generated for the document. When adding many documents, it is more efficient to use add.

victor.add_single("Pepperoni pizza", vec!["Pizza Flavors"]).await;

pub async fn add_embeddings( &mut self, to_add: Vec<(impl Into<String>, Vec<f32>)>, tags: Vec<impl Into<String>>, )

Add many document/embedding pairs to the database. This is useful for adding embeddings that have already been generated.

victor.add_embeddings(vec![("Pepperoni pizza", vec![0.1, 0.2, 0.3])], vec!["Pizza Flavors"]).await;

pub async fn add_single_embedding( &mut self, content: impl Into<String>, vector: Vec<f32>, tags: Vec<impl Into<String>>, )

Add a single document/embedding pair to the database. This is useful for adding embeddings that have already been generated. When adding many documents, it is more efficient to use add_embeddings.

victor.add_single_embedding("Pepperoni pizza", vec![0.1, 0.2, 0.3], vec!["Pizza Flavors"]).await;

pub async fn search( &self, content: impl Into<String>, with_tags: Vec<impl Into<String>>, top_n: u32, ) -> Vec<NearestNeighborsResult>

Search the database for the nearest neighbors to a given document. An embedding will be generated for the document being searched for. This will return the top top_n nearest neighbors.

victor.search("Pepperoni pizza", vec!["Pizza Flavors"], 10).await;

pub async fn search_embedding( &self, vector: Vec<f32>, with_tags: Vec<impl Into<String>>, top_n: u32, ) -> Vec<NearestNeighborsResult>

Search the database for the nearest neighbors to a given embedding. This will return the top top_n nearest neighbors.

pub async fn clear_db(&mut self) -> Result<(), D::Error>

Clear the database, deleting all data.