README
- ::addQdrantStore::add
- ::clearQdrantStore::clear
- ::createQdrantStore::create
- ::deleteQdrantStore::delete
- ::lenQdrantStore::len
- ::newQdrantStore::new
- ::searchQdrantStore::search

moonqdrant

A Qdrant vector store for moonbitstack/moonrag, over Qdrant's REST API.

moonrag

qdrant

vector-store

vector-database

moonbit

Download zip

Author

moonbitstack

Version

0.1.0

License

Apache-2.0

Repository

github.com/moonbitstack/moonrag

Last updated

9 hours ago

Downloads

#moonqdrant

A Qdrant vector store for moonrag, over its REST API. Native only: it opens sockets, so it is not built for js or wasm.

moon add moonbitstack/moonqdrant

let store = @moonqdrant.QdrantStore::new("http://127.0.0.1:6333", embedder, dimension=256)

It implements moonrag's VectorStore. Like the in-memory, SQLite and MySQL stores it recomputes the cosine in this process and ranks by the same Filter::matches, so its results agree with the other backends rather than trusting the server.

See this store's page, the guide and the API.

QdrantStore

source

pub struct QdrantStore {
  base : String
  collection : String
  embedder : &@moonbitstack/moonrag.Embedder
Embedder
  dimension : Int
  counter : Int
}

impl VectorStore for QdrantStore

async fn add(self : QdrantStore, documents : Array[@moonbitstack/moonrag.Document
Document]) -> Array[String]

source

Embed documents, store them as points, and return the ids they are stored under.

async fn delete(self : QdrantStore, ids : ArrayView[String]) -> Unit

source

Delete the points stored under ids. An id not present is not an error.

async fn search(self : QdrantStore, vector : ArrayView[Double], limit~ : Int, filter? : @moonbitstack/moonrag.Filter
Filter) -> Array[@moonbitstack/moonrag.Scored
Scored]

source

The best limit documents for vector, best first. Every point is read back with its vector, then filtered, scored and ranked here — the same arithmetic the in-memory store uses.

QdrantStore::add

source

async fn QdrantStore::add(self : QdrantStore, documents : Array[@moonbitstack/moonrag.Document
Document]) -> Array[String]

QdrantStore::clear

source

async fn QdrantStore::clear(self : QdrantStore) -> Unit

Remove every stored point by recreating the collection empty.

QdrantStore::create

source

async fn QdrantStore::create(self : QdrantStore) -> Unit

Create the collection this store needs. Safe to call on every start: an existing collection of the same shape is left as it is.

QdrantStore::delete

source

async fn QdrantStore::delete(self : QdrantStore, ids : ArrayView[String]) -> Unit

QdrantStore::len

source

async fn QdrantStore::len(self : QdrantStore) -> Int64

The number of stored points.

QdrantStore::new

source

fn QdrantStore::new(base : String, embedder : &@moonbitstack/moonrag.Embedder
Embedder, dimension~ : Int, collection? : String) -> QdrantStore raise

A store over the Qdrant server at base (for example http://127.0.0.1:6333).

QdrantStore::search

source

async fn QdrantStore::search(self : QdrantStore, vector : ArrayView[Double], limit~ : Int, filter? : @moonbitstack/moonrag.Filter
Filter) -> Array[@moonbitstack/moonrag.Scored
Scored]