Embedding stores

Once the embeddings have been generated, we need to store them in a database. However, not all databases are suitable for this task. The main problem is that the embeddings are high-dimensional vectors, which makes it difficult to store them in a database. It is also required that the database has a way to perform similarity searches on the embeddings.

Only a subset of databases can be used to store embeddings. ELLMental provides built-in clients to use different embedding stores, you just have to choose which one you want to use. Then, provide the required configuration for that store and eLLMental will take care of the rest.

Local store

This is the most basic of the stores. It stores the embeddings in a local file using the Chroma client. This is useful for testing purposes, but it is not recommended for production environments.

EMBEDDING_STORE="LOCAL"
STORE_PATH="<STORE_PATH>"

• STORE_PATH is the path where the ChromaDB will be generated locally.

Chroma store

In case you want to use ChromaDB as a production database, you can use this store. It uses the Chroma client to store the embeddings in a ChromaDB database. This Chroma database doesn't have to be local, it can be a remote ChromaDB, which is recommended for production environments.

EMBEDDING_STORE="CHROMA"
CHROMA_URL="<CHROMA_URL>"
CHROMA_PORT="<CHROMA_PORT>"
CHROMA_COLLECTION="<CHROMA_COLLECTION>"

• CHROMA_URL is the URL of the ChromaDB database.
• CHROMA_PORT is the port of the ChromaDB database.
• CHROMA_COLLECTION is the collection where the embeddings are stored.

Supabase store

EMBEDDING_STORE="SUPABASE"
SUPABASE_URL="<SUPABASE_URL>"
SUPABASE_KEY="<SUPABASE_KEY>"
SUPABASE_TABLE="<SUPABASE_TABLE>"
SUPABASE_FUNCTION="<SUPABASE_FUNCTION>"

• SUPABASE_URL is the URL of the Supabase database.
• SUPABASE_KEY is the access key of the Supabase database.
• SUPABASE_TABLE is the table where the embeddings are stored.
• SUPABASE_FUNCTION is the function that is used to perform similarity searches.

You can check out how to set up your Supabase database here.

Pinecone store

Another alternative could be Pinecone. It is a vector-oriented database created specifically for semanitc search. It is a cloud-based service, so you don't have to worry about setting up a database by yourself.

EMBEDDING_STORE="PINECONE"
PINECONE_KEY="<PINECONE_KEY>"
PINECONE_ENVIRONMENT="<PINECONE_ENVIRONMENT>"
PINECONE_INDEX="<PINECONE_INDEX>"

• PINECONE_KEY is the access key of the Pinecone database.
• PINECONE_ENVIRONMENT is the environment where the Pinecone project is placed.
• PINECONE_INDEX is the Pinecone index (database) where embeddings are stored and queried.

You can check out how to set up your Pinecone database here.