Crate erio_embedding 

erio-embedding

erio-embedding defines embedding engine abstractions and concrete implementations used by Erio retrieval pipelines.

It supports local model execution via GemmaEmbedding and includes additional configuration/types for embedding tasks.

Quickstart

Set ERIO_MODEL_DIR to a directory containing local model assets before constructing GemmaEmbedding.

export ERIO_MODEL_DIR=/absolute/path/to/local/embedding-model

use erio_embedding::{EmbeddingConfig, EmbeddingEngine, GemmaEmbedding};

let _engine: Box<dyn EmbeddingEngine> = Box::new(GemmaEmbedding::new(EmbeddingConfig::default())?);

docs.rs behavior

Real usage requires ERIO_MODEL_DIR to point to valid local model assets. During docs.rs builds (DOCS_RS=1), the build script skips model download and validation by setting ERIO_MODEL_DIR to a dummy directory, so docs/examples are for API reference and are not runnable there.

API tour

• Engine API: EmbeddingEngine
• Implementations: GemmaEmbedding, RemoteEmbedding
• Config/types: EmbeddingConfig, TaskType
• Error type: EmbeddingError
• Modules: engine, gemma, remote, config, task, model, error

Related crates

• Used by erio-context-store for vector ingestion and semantic search.
• Often paired with erio-llm-client in RAG workflows.
• Docs: https://docs.rs/erio-embedding
• Source: https://github.com/NomanworkGroup/erio/tree/main/crates/embedding

Compatibility

• MSRV: Rust 1.93
• License: Apache-2.0 Erio Embedding - embedding engine abstraction for computing vector embeddings.

Local embedding engines (e.g. GemmaEmbedding) load model files from ERIO_MODEL_DIR. The embedding crate’s build script fetches the default model assets at build time.

Re-exports

pub use config::EmbeddingConfig;

pub use engine::EmbeddingEngine;

pub use error::EmbeddingError;

pub use gemma::GemmaEmbedding;

pub use remote::RemoteEmbedding;

pub use task::TaskType;

Modules

config

Configuration for embedding engines.

engine

Embedding engine trait definition.

error

Embedding-specific error types.

gemma

Local embedding engine using the GGUF-quantized EmbeddingGemma model.

model

GGUF-quantized EmbeddingGemma model internals.

remote

Remote embedding engine calling an OpenAI-compatible embedding API.

task

Task type definitions and prompt formatting for embedding models.