Module embeddings 

Text Embedding Providers

This module provides a pluggable embedding system for AgenticDB.

Available Providers

• HashEmbedding: Fast hash-based placeholder (default, not semantic)
• OnnxEmbedding: Real semantic embeddings using ONNX Runtime (feature: onnx-embeddings) ✅ RECOMMENDED
• LatticeEmbedding: Real semantic embeddings using lattice-embed, pure-Rust native inference (feature: lattice-embeddings)
• CandleEmbedding: Real embeddings using candle-transformers (feature: real-embeddings)
• ApiEmbedding: External API calls (OpenAI, Anthropic, Cohere, etc.)

Usage

use ruvector_core::embeddings::{EmbeddingProvider, HashEmbedding};

// Default: Hash-based (fast, but not semantic)
let hash_provider = HashEmbedding::new(384);
let embedding = hash_provider.embed("hello world")?;

ONNX Embeddings (Recommended for Production)

use ruvector_core::embeddings::{EmbeddingProvider, OnnxEmbedding};

// Real semantic embeddings using all-MiniLM-L6-v2
let provider = OnnxEmbedding::from_pretrained("sentence-transformers/all-MiniLM-L6-v2")?;
let embedding = provider.embed("hello world")?;
// "dog" and "cat" WILL be similar (semantic understanding!)

Structs

ApiEmbedding

API-based embedding provider (OpenAI, Anthropic, Cohere, etc.)

EmbeddingSpaceAccess

Capabilities retained when opening a corpus with an incompatible active embedding space. Vector inspection never invokes the active embedder.

EmbeddingSpaceIdentity

Complete immutable identity of a retrieval vector space.

HashEmbedding

Hash-based embedding provider (placeholder, not semantic)

PromptTemplates

The exact query and passage prompt templates a registered model owns.

SymmetricEmbeddingAdapter

Explicit migration wrapper for symmetric embedding implementations.

Enums

EmbeddingDistanceMetric

EmbeddingRole

The role of text at a retrieval boundary.

EmbeddingRolePolicy

Whether a model requires distinct query and passage transforms.

OutputDtype

PoolingStrategy

PoolingStrategyName

PrefixPolicy

Constants

EMBEDDING_SPACE_FORMAT_REVISION

Revision of this crate’s embedding semantics, and the only ruvector-owned version number that may reach runtime_revision and therefore embedding_space_id.

Traits

EmbeddingProvider

Trait for text embedding providers

Functions

apply_prompt_template

Apply the identity’s registered prompt template for role to raw text.

build_revision

Build provenance of the running binary — informational only.

default_space_identity

A non-attested symmetric identity for providers with no registry fixture (hash placeholders, opaque fleet backends, tests).

embedding_cache_key

legacy_embed_call_count

registry_identity

Build the identity a registered model must have, from the exact registry fixture for model_id.

registry_prompt_templates

The exact prompt templates registered for model_id.

Type Aliases

BoxedEmbeddingProvider

Type-erased embedding provider for dynamic dispatch