5.8 KiB

Raw Permalink Blame History

Oracle Database Embedding Store

This module implements EmbeddingStore using Oracle Database.

Requirements

Oracle Database 23.4 or newer

Installation

<dependency>
    <groupId>dev.langchain4j</groupId>
    <artificatId>langchain4j-oracle</artificatId>
    <version>0.1.0</version>
</dependency>

Usage

Instances of this store can be created by configuring a builder. The builder requires that a DataSource and an embedding table be provided. The distance between two vectors is calculated using cosine similarity which measures the cosine of the angle between two vectors.

It is recommended to configure a DataSource with pools connections, such as the Universal Connection Pool or Hikari. A connection pool will avoid the latency of repeatedly creating new database connections.

If an embedding table already exists in your database provide the table name.

EmbeddingStore embeddingStore = OracleEmbeddingStore.builder()
   .dataSource(myDataSource)
   .embeddingTable("my_embedding_table")
   .build();

If the table does not already exist, it can be created by passing a CreateOption to the builder.

EmbeddingStore embeddingStore = OracleEmbeddingStore.builder()
   .dataSource(myDataSource)
   .embeddingTable("my_embedding_table", CreateOption.CREATE_IF_NOT_EXISTS)
   .build();

By default the embedding table will have the following columns:

Name	Type	Description
id	VARCHAR(36)	Primary key. Used to store UUID strings which are generated when the embedding store
embedding	VECTOR(*, FLOAT32)	Stores the embedding
text	CLOB	Stores the text segment
metadata	JSON	Stores the metadata

If the columns of your existing table do not match the predefined column names or you would like to use different column names, you can use a EmbeddingTable builder to configure your embedding table.

OracleEmbeddingStore embeddingStore =
OracleEmbeddingStore.builder()
    .dataSource(myDataSource)
    .embeddingTable(EmbeddingTable.builder()
            .createOption(CREATE_OR_REPLACE) // use NONE if the table already exists
            .name("my_embedding_table")
            .idColumn("id_column_name")
            .embeddingColumn("embedding_column_name")
            .textColumn("text_column_name")
            .metadataColumn("metadata_column_name")
            .build())
    .build();

The builder allows you to create indexes on the embedding and metadata columns of the EmbeddingTable by providing an instance of the Index class. Two builders allow you to create instances of the Index class: IVFIndexBuilder and JSONIndexBuilder.

IVFIndexBuilder allows you to configure an IVF (Inverted File Flat) index on the embedding column of the EmbeddingTable.

OracleEmbeddingStore embeddingStore =
    OracleEmbeddingStore.builder()
        .dataSource(myDataSource)
        .embeddingTable(EmbeddingTable.builder()
            .createOption(CreateOption.CREATE_OR_REPLACE) // use NONE if the table already exists
            .name("my_embedding_table")
            .idColumn("id_column_name")
            .embeddingColumn("embedding_column_name")
            .textColumn("text_column_name")
            .metadataColumn("metadata_column_name")
            .build())
        .index(Index.ivfIndexBuilder().createOption(CreateOption.CREATE_OR_REPLACE).build())
        .build();

JSONIndexBuilder allows you to configure a function-based index on keys of the metadata column of the EmbeddingTable.

OracleEmbeddingStore.builder()
    .dataSource(myDataSource)
    .embeddingTable(EmbeddingTable.builder()
        .createOption(CreateOption.CREATE_OR_REPLACE) // use NONE if the table already exists
        .name("my_embedding_table")
        .idColumn("id_column_name")
        .embeddingColumn("embedding_column_name")
        .textColumn("text_column_name")
        .metadataColumn("metadata_column_name")
        .build())
    .index(Index.jsonIndexBuilder()
        .createOption(CreateOption.CREATE_OR_REPLACE)
        .key("name", String.class, JSONIndexBuilder.Order.ASC)
        .key("year", Integer.class, JSONIndexBuilder.Order.DESC)
        .build())
    .build();

For more information about Oracle AI Vector Search refer to the documentation.

Chat Memory Store

This module also provides OracleChatMemoryStore, a simple persistent implementation of ChatMemoryStore.

Create a table:

CREATE TABLE chat_memory (
    memory_id VARCHAR2(255) PRIMARY KEY,
    content CLOB NOT NULL
);

Use it in chat memory:

ChatMemoryStore store = OracleChatMemoryStore.builder()
   .dataSource(myDataSource)
   .tableName("chat_memory")
   .build();

ChatMemory chatMemory = MessageWindowChatMemory.builder()
   .id("conversation-1")
   .maxMessages(10)
   .chatMemoryStore(store)
   .build();

OracleChatMemoryStore stores one row per memory id, with all messages serialized as JSON in the content column.

Running the Test Suite

By default, integration tests will run a docker image of Oracle Database using TestContainers. Alternatively, the tests can connect to an Oracle Database if the following environment variables are configured:

ORACLE_JDBC_URL : Set to an Oracle JDBC URL, such as jdbc:oracle:thin@example:1521/serviceName
ORACLE_JDBC_USER : Set to the name of a database user. (Optional)
ORACLE_JDBC_PASSWORD : Set to the password of a database user. (Optional)

5.8 KiB Raw Permalink Blame History