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};

# fn main() -> Result<(), erio_embedding::EmbeddingError> {
let _engine: Box<dyn EmbeddingEngine> = Box::new(GemmaEmbedding::new(EmbeddingConfig::default())?);
# Ok(())
# }

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