Crate qql_embed 

Shared embedding layer for QQL.

• Embedder — host-agnostic dense/sparse API (batch dense by default when overridden)
• resolve_embeddings — rewrite QUERY / UPSERT text into vectors on the AST
• sparse — local wire-compatible BM25 sparse vectors (no network)

Used by qql (runtime HttpEmbedder), qql-edge (FastEmbedder), and qql-wasm (fetch / JS adapters). No Qdrant I/O and no HTTP client live here.

Re-exports

pub use embedder::Embedder;

pub use embedder::EmbedderBound;

pub use embedder::JointEmbeddingOutput;

pub use embedder::SparseEmbedder;

pub use embedder::cross_rerank_unsupported_error;

pub use embedder::dense_model_unsupported_error;

pub use embedder::image_unsupported_error;

pub use embedder::multi_unsupported_error;

pub use embedder::sparse_model_unsupported_error;

pub use resolve::DENSE_VECTOR_NAME;

pub use resolve::SPARSE_VECTOR_NAME;

pub use resolve::resolve_embeddings;

pub use sparse::Bm25Params;

pub use sparse::SparseVector;

pub use topology::TopologyNames;

pub use topology::query_needs_kind_resolution;

pub use topology::resolve_query_vector_kinds;

pub use topology::resolve_query_vector_kinds_simple;

Modules

embedder

Host-agnostic Embedder trait, joint embedding output, and error helpers.

resolve

AST rewriting: resolve TEXT inputs to vectors via the configured embedder.

sparse

Local wire-compatible BM25 sparse vectors (no network). Client-side BM25 sparse embeddings, wire-compatible with Qdrant’s qdrant/bm25 model defaults.

topology

USING vector-kind resolution from collection topology (dense/sparse/multi). Resolve USING vector kinds from collection topology before embedding.