langchain-vedb 0.1.3

An integration package connecting VeDB and LangChain

VeDB

This notebook covers how to get started with the VeDB.

Setup

To access VeDB you'll need to create a/an ByteDance account, get an API key, and install the langchain-vedb integration package.

%pip install -qU "langchain-vedb>=MINIMUM_VERSION"

Initialization

import EmbeddingTabs from "@theme/EmbeddingTabs";

extra_column = {"source": "VARCHAR(255)"}   # Define the field information to be added to metadata

client_config = {"host": 'xxx.xxx.xxx.xxx',  # e.g., '192.168.1.88'
                 "port": 12345,
                 "user": 'vector_user',
                 "password": '123456',  # Leave empty if no password was set during authorization
                 "database": 'vectorStoreTest',  # Use your actual database name
                 "table_name": 'langchain_vector_test',  # Set an existing table name or initialize a new one
                 "charset": 'utf8mb4',
                 "metadata_columns": extra_column}     # Additional metadata columns
vector_store = ByteDanceVectorStore(embeddings, client_config)

Manage vector store

Add items to vector store

from langchain_core.documents import Document

document_1 = Document(page_content="foo", metadata={"source": "https://example.com"})

document_2 = Document(page_content="bar", metadata={"source": "https://example.com"})

document_3 = Document(page_content="baz", metadata={"source": "https://example.com"})

documents = [document_1, document_2, document_3]

vector_store.add_documents(documents=documents, ids=["1", "2", "3"])

You can also use the from_text function to directly instantiate a vector store from a document.

docs = ["foo","bar","bar"]

metadatas = [{"source": "https://example.com"},{"source": "https://example.com"},{"source": "https://example.com"}]

vector_store = ByteDanceVectorStore.from_texts(texts=docs, embedding=embeddings, metadatas=metadatas, client_config=client_config)

Delete items from vector store

vector_store.delete(ids=["3"])

Update items in vector store

updated_document_1 = Document(
    id = "1", page_content="qux", metadata={"source": "https://another-example.com"}
)

vector_store.update_document(document=updated_document_1)

Get documents by ids

You can directly get a specific document by its ID.

results = vector_store.get_by_ids(["1", "2", "3"])
for doc in results:
     print(f"* {doc.id} {doc.page_content} [{doc.metadata}]")

Query vector store

Once your vector store has been created and the relevant documents have been added, you will most likely wish to query it during the running of your chain or agent.

Filtering Support

The vectorstore supports a set of filters that can be applied against the metadata fields of the documents.

Operator	Meaning/Category
$eq	Equality (==)
$ne	Inequality (!=)
$lt	Less than (<)
$lte	Less than or equal (<=)
$gt	Greater than (>)
$gte	Greater than or equal (>=)
$in	Special Case (in)
$nin	Special Case (not in)
$between	Special Case (between)
$exists	Exists (IS [NOT] NULL)
$and	Logical (and)
$or	Logical (or)

Query directly

A simple similarity search can be performed in the following way.

results = vector_store.similarity_search(
    query="thud", k=1, filter={"source": "https://example.com"}
)
for doc in results:
    print(f"* {doc.page_content} [{doc.metadata}]")

You can run the following to execute a similarity search and get the corresponding scores

results = vector_store.similarity_search_with_score(
    query="thud", k=1, filter={"source": "https://example.com"}    
)
for doc, score in results:
    print(f"* [SIM={score:3f}] {doc.page_content} [{doc.metadata}]")

If you provide a dict with multiple fields, but no operators, the top level will be interpreted as a logical AND filter

results = vector_store.similarity_search_with_score(
    query="thud", 
    k=10, 
    filter={"year": {"$in": [2004, 2022]}, "topic": {"$eq": "magic"}},
)

results = vector_store.similarity_search_with_score(
    query="thud", 
    k=10, 
    filter={
         "$and":[
         {"year": {"$in": [2004, 2025]}}, 
         {"topic": {"$eq": "magic"}},
         ]
     },
)

Limitations

After adding documents, the system builds the ANN index asynchronously in the background. Therefore, you need to wait a short period between adding documents and performing similarity searches. You can use sleep to pause for a while. To prevent potential crashes, you can run the following between adding documents and performing a similarity search.

vector_store.close()

---

❗️Needs to be modified

API reference

For detailed documentation of all ByteDanceVectorStore features and configurations head to the API reference: https://api.python.langchain.com/en/latest/vectorstores/langchain_vedb_link.vectorstores.VeDB.html

Related

• Vector store conceptual guide
• Vector store how-to guides