Skip to Content
DocumentationAdvancedRetrieval-Augmented Generation (RAG)

Retrieval-Augmented Generation (RAG)

Alith supports Retrieval-Augmented Generation (RAG), a technique that combines retrieval of relevant information from a knowledge base with text generation. This allows agents to provide more accurate and context-aware responses by leveraging external data.

RAG with Memory

use alith::{Agent, EmbeddingsBuilder, InMemoryStorage, LLM};
 
#[tokio::main]
async fn main() -> Result<(), anyhow::Error> {
    let model = LLM::from_model_name("gpt-4")?;
    let embeddings_model = model.embeddings_model("text-embedding-ada-002");
    let data = EmbeddingsBuilder::new(embeddings_model.clone())
        .documents(vec!["doc0", "doc1", "doc2"])
        .unwrap()
        .build()
        .await?;
    let storage = InMemoryStorage::from_multiple_documents(embeddings_model, data);
 
    let agent = Agent::new("simple agent", model, vec![])
        .preamble(
            r#"
You are a dictionary assistant here to assist the user in understanding the meaning of words.
You will find additional non-standard word definitions that could be useful below.
"#,
        )
        .store_index(1, storage);
    let response = agent.prompt("What does \"glarb-glarb\" mean?").await?;
 
    println!("{}", response);
 
    Ok(())
}

RAG with Vector Database

use alith::store::qdrant::{
    CreateCollectionBuilder, Distance, QdrantClient, QdrantStorage, VectorParamsBuilder,
    DEFAULT_COLLECTION_NAME,
};
use alith::{Agent, EmbeddingsBuilder, LLM};
 
#[tokio::main]
async fn main() -> Result<(), anyhow::Error> {
    let model = LLM::from_model_name("gpt-4")?;
    let embeddings_model = model.embeddings_model("text-embedding-ada-002");
    let data = EmbeddingsBuilder::new(embeddings_model.clone())
        .documents(vec!["doc0", "doc1", "doc2"])
        .unwrap()
        .build()
        .await?;
 
    let client = QdrantClient::from_url("http://localhost:6334").build()?;
 
    if !client.collection_exists(DEFAULT_COLLECTION_NAME).await? {
        client
            .create_collection(
                CreateCollectionBuilder::new(DEFAULT_COLLECTION_NAME)
                    .vectors_config(VectorParamsBuilder::new(1536, Distance::Cosine)),
            )
            .await?;
    }
 
    let storage = QdrantStorage::from_multiple_documents(client, embeddings_model, data).await?;
 
    let agent = Agent::new("simple agent", model, vec![])
        .preamble(
            r#"
You are a dictionary assistant here to assist the user in understanding the meaning of words.
You will find additional non-standard word definitions that could be useful below.
"#,
        )
        .store_index(1, storage);
    let response = agent.prompt("What does \"glarb-glarb\" mean?").await?;
 
    println!("{}", response);
 
    Ok(())
}
Last updated on
StoreChain of Thought (CoT)