Module embedding 

Embedding provider abstraction.

Defines a trait for text embedding and provides common utilities. Concrete backends (local ONNX, candle, OpenAI) are feature-gated.

use llm_kernel::embedding::{EmbeddingProvider, EmbeddingResult};

Re-exports

pub use catalog::EmbeddingModel;

pub use types::EmbeddingProvider;

pub use types::EmbeddingResult;

pub use types::chunk_batch;

pub use types::cosine_similarity;

pub use openai::OpenAIEmbeddingClient;

pub use fastembed::FastembedProvider;

pub use lazy::EmbeddingCache;

pub use lazy::LazyFastembedProvider;

pub use lazy::LazyOpts;

pub use lazy::ModelState;

pub use lazy::is_model_cached;

pub use qwen3::Qwen3Provider;

pub use nomic_moe::NomicMoeProvider;

pub use async_vector_index::AsyncVectorIndex;

pub use vector_index::SearchHit;

pub use vector_index::VectorIndex;

pub use qdrant::QdrantVectorIndex;

pub use elastic::ElasticsearchVectorIndex;

pub use turbovec::TurbovecIndex;

Modules

async_vector_index

Async vector index trait for remote/shared backends (needs async_trait). Async vector index trait for remote/shared backends.

catalog

Zero-dep embedding model catalog.

elastic

Elasticsearch AsyncVectorIndex (feature elastic). Elasticsearch AsyncVectorIndex (elastic feature).

fastembed

Local ONNX embedding provider via fastembed-rs.

lazy

Lazy-loading embedding provider with idle eviction and LRU query cache.

nomic_moe

Nomic V2 MoE embedding provider via fastembed-rs candle backend.

openai

OpenAI text-embedding provider (sync, via ureq).

qdrant

Qdrant AsyncVectorIndex (feature qdrant). Qdrant AsyncVectorIndex (qdrant feature).

qwen3

Qwen3 embedding provider via fastembed-rs candle backend.

turbovec

TurboQuant-backed vector index implementation.

types

Embedding types and trait definitions.

vector_index

Vector index trait and types (zero dependencies). Vector index trait for compressed approximate nearest neighbor search.