Back to Crewai

Db2 Vector Search Tool

docs/v1.15.11/en/tools/database-data/db2searchtool.mdx

1.15.115.7 KB
Original Source

DB2VectorSearchTool

Description

Perform semantic vector similarity searches against IBM Db2 tables using the native VECTOR_DISTANCE function. Supports configurable distance metrics, OpenAI or custom embeddings, metadata filtering, and result shaping.

Installation

bash
pip install ibm_db openai

Or with uv:

bash
uv add ibm_db openai

Environment Variables

bash
OPENAI_API_KEY=your_openai_key          # Required when using default OpenAI embeddings
DB2_CONNECTION_STRING=DATABASE=TESTDB;HOSTNAME=localhost;PORT=50000;PROTOCOL=TCPIP;UID=db2user;PWD=password;

Basic Usage

python
from crewai import Agent
from crewai_tools import DB2VectorSearchTool

tool = DB2VectorSearchTool(
    connection_string="DATABASE=TESTDB;HOSTNAME=localhost;PORT=50000;PROTOCOL=TCPIP;UID=db2user;PWD=password;",
    table_name="documents",
    vector_column="embedding",
)

agent = Agent(
    role="Research Assistant",
    goal="Find relevant information in documents",
    tools=[tool],
)

Full Semantic Search Workflow

python
import os
from dotenv import load_dotenv
from crewai import Agent, Task, Crew, Process
from crewai_tools import DB2VectorSearchTool

load_dotenv()

db2_tool = DB2VectorSearchTool(
    connection_string=os.getenv("DB2_CONNECTION_STRING"),
    table_name="documents",
    vector_column="embedding",
    return_columns=["content", "category"],
    limit=3,
    distance_metric="COSINE",
    max_distance=0.35,
)

search_agent = Agent(
    role="Senior Semantic Search Agent",
    goal="Find and analyse documents based on semantic search",
    backstory="You are an expert research assistant who can find relevant information using semantic search in a Db2 database.",
    tools=[db2_tool],
    verbose=True,
)

answer_agent = Agent(
    role="Senior Answer Assistant",
    goal="Generate answers based on retrieved context",
    backstory="You are an expert assistant who generates answers from provided context.",
    tools=[db2_tool],
    verbose=True,
)

search_task = Task(
    description="""Search for relevant documents about {query}.
    Include the relevant information found, vector distances, and returned fields.""",
    agent=search_agent,
)

answer_task = Task(
    description="Given the retrieved Db2 context, generate a final answer.",
    agent=answer_agent,
)

crew = Crew(
    agents=[search_agent, answer_agent],
    tasks=[search_task, answer_task],
    process=Process.sequential,
    verbose=True,
)

result = crew.kickoff(inputs={"query": "What is the role of X in the document?"})
print(result)

Tool Parameters

ParameterTypeDefaultDescription
connection_stringstrrequiredDb2 connection string. Format: DATABASE=x;HOSTNAME=x;PORT=50000;PROTOCOL=TCPIP;UID=x;PWD=x;
table_namestr"documents"Table to search. Supports schema.table notation.
vector_columnstr"embedding"Column storing the vector embeddings.
embedding_modelstr"text-embedding-3-large"OpenAI model used when no custom embedding function is provided.
return_columnslist[str]["content"]Columns to include in each result. Must contain at least one entry.
limitint3Maximum number of results (1–100).
distance_metricstr"COSINE"Db2 distance metric. See supported values below.
max_distancefloat | NoneNoneDrop results whose distance exceeds this value.
custom_embedding_fnCallable[[str], list[float]] | NoneNoneCustom embedding function. Overrides OpenAI when provided.

Supported Distance Metrics

The following values map directly to the Db2 VECTOR_DISTANCE function:

  • COSINE
  • EUCLIDEAN
  • EUCLIDEAN_SQUARED
  • DOT
  • HAMMING
  • MANHATTAN

Reference: IBM Db2 VECTOR_DISTANCE documentation

Schema Parameters (per query)

ParameterTypeRequiredDescription
querystrThe search query.
filter_bystr | NoneColumn name for metadata filtering. Must be paired with filter_value.
filter_valueAny | NoneValue to filter on. Must be paired with filter_by.

Return Format

json
{
  "success": true,
  "results": [
    {
      "distance": 0.1401,
      "data": {
        "content": "Document content here",
        "category": "research"
      }
    }
  ]
}

On error:

json
{
  "success": false,
  "error": "Description of what went wrong",
  "error_type": "ExceptionClassName"
}

Metadata Filtering

python
result = db2_tool.run(
    query="machine learning",
    filter_by="category",
    filter_value="research",
)

filter_by and filter_value must always be provided together. Providing only one raises a validation error.

Custom Embeddings

Use any embedding model by supplying a custom_embedding_fn:

python
from sentence_transformers import SentenceTransformer
from crewai_tools import DB2VectorSearchTool

model = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")

def custom_embeddings(text: str) -> list[float]:
    return model.encode(text).tolist()

tool = DB2VectorSearchTool(
    connection_string="DATABASE=TESTDB;HOSTNAME=localhost;PORT=50000;PROTOCOL=TCPIP;UID=db2user;PWD=password;",
    table_name="documents",
    custom_embedding_fn=custom_embeddings,
)

When custom_embedding_fn is provided, OPENAI_API_KEY is not required.

Security Features

  • SQL identifier validation (table, column names must match ^[A-Za-z][A-Za-z0-9_]*(\.[A-Za-z][A-Za-z0-9_]*)?$)
  • Parameterised SQL queries — values never interpolated into SQL strings
  • Distance metric whitelist — only valid Db2 metric names accepted