Module vector 

Vector memory and semantic search capabilities.

This module provides vector embeddings and semantic search functionality for agent memories. It enables similarity-based memory retrieval using vector embeddings rather than simple keyword matching.

Features

• Embedding Generation: Convert text to vector embeddings using various providers
• Semantic Search: Find similar memories based on meaning, not just keywords
• Hybrid Search: Combine semantic similarity with traditional keyword search
• Multiple Backends: Support for local and cloud vector databases

Architecture

The vector memory system is built around three core traits:

• EmbeddingProvider - Generates vector embeddings from text
• VectorStore - Stores and retrieves vectors with metadata
• VectorMemory - High-level interface combining embeddings and storage

Available Backends

• LocalVectorStore - In-memory vector store with cosine similarity search

Example

use ceylon_next::memory::vector::{EmbeddingProvider, VectorMemory, LocalVectorStore};
use std::sync::Arc;

#[tokio::main]
async fn main() {
    // Create an embedding provider (e.g., OpenAI, Ollama, etc.)
    // let embedder = OpenAIEmbeddings::new("api-key");

    // Create a vector store
    let store = LocalVectorStore::new(384); // 384-dimensional embeddings

    // Use for semantic search
    // let results = store.search(&query_vector, 5).await.unwrap();
}

Structs

CachedEmbeddings

A caching wrapper for embedding providers.

LocalVectorStore

In-memory vector store using brute-force cosine similarity search.

SearchResult

A search result with similarity score.

VectorEntry

A vector embedding with associated metadata.

Traits

EmbeddingProvider

Trait for generating vector embeddings from text.

VectorMemory

High-level interface for semantic memory operations.

VectorStore

Trait for storing and retrieving vector embeddings.

Functions

cosine_similarity

Computes the cosine similarity between two vectors.

normalize_vector

Normalizes a vector to unit length (L2 normalization).