IBM watsonx.ai

WatsonxToolkit is a wrapper for IBM watsonx.ai Toolkit.

This example shows how to use watsonx.ai Toolkit using LangChain.

Overview

Integration details

Class	Package	Serializable	JS support	Package downloads	Package latest
WatsonxToolkit	langchain-ibm	❌	✅

Setup

To access IBM watsonx.ai toolkit you'll need to create an IBM watsonx.ai account, get an API key, and install the langchain-ibm integration package.

Credentials

This cell defines the WML credentials required to work with watsonx Toolkit.

Action: Provide the IBM Cloud user API key. For details, see documentation.

import os
from getpass import getpass

watsonx_api_key = getpass()
os.environ["WATSONX_APIKEY"] = watsonx_api_key

Additionaly you are able to pass additional secrets as an environment variable.

import os

os.environ["WATSONX_URL"] = "your service instance url"
os.environ["WATSONX_TOKEN"] = "your token for accessing the service instance"

Installation

The LangChain IBM integration lives in the langchain-ibm package:

!pip install -qU langchain-ibm

Instantiation

Initialize the WatsonxToolkit class.

from langchain_ibm import WatsonxToolkit

watsonx_toolkit = WatsonxToolkit(
    url="https://us-south.ml.cloud.ibm.com",
)

API Reference:WatsonxToolkit

For certain requirements, there is an option to pass the IBM's APIClient object into the WatsonxToolkit class.

from ibm_watsonx_ai import APIClient

api_client = APIClient(...)

watsonx_toolkit = WatsonxToolkit(
    watsonx_client=api_client,
)

Tools

Get all tools

It is possible to get all available tools as a list of WatsonxTool objects.

watsonx_toolkit.get_tools()

[WatsonxTool(name='GoogleSearch', description='Search for online trends, news, current events, real-time information, or research topics.', args_schema=<class 'langchain_ibm.toolkit.ToolArgsSchema'>, agent_description='Search for online trends, news, current events, real-time information, or research topics.', tool_config_schema={'title': 'config schema for GoogleSearch tool', 'type': 'object', 'properties': {'maxResults': {'title': 'Max number of results to return', 'type': 'integer', 'minimum': 1, 'maximum': 20}}}, watsonx_client=<ibm_watsonx_ai.client.APIClient object at 0x127e0f490>),
 WatsonxTool(name='WebCrawler', description='Useful for when you need to summarize a webpage. Do not use for Web search.', args_schema=<class 'langchain_ibm.toolkit.ToolArgsSchema'>, agent_description='Useful for when you need to summarize a webpage. Do not use for Web search.', tool_input_schema={'type': 'object', 'properties': {'url': {'title': 'url', 'description': 'URL for the webpage to be scraped', 'type': 'string', 'pattern': '^(https?:\/\/)?([\da-z\.-]+)\.([a-z\.]{2,6})([\/\w \.-]*)*\/?$'}}, 'required': ['url']}, watsonx_client=<ibm_watsonx_ai.client.APIClient object at 0x127e0f490>),
 WatsonxTool(name='SDXLTurbo', description='Generate an image from text using Stability.ai', args_schema=<class 'langchain_ibm.toolkit.ToolArgsSchema'>, agent_description='Generate an image from text. Not for image refining. Use very precise language about the desired image, including setting, lighting, style, filters and lenses used. Do not ask the tool to refine an image.', watsonx_client=<ibm_watsonx_ai.client.APIClient object at 0x127e0f490>),
 WatsonxTool(name='Weather', description='Find the weather for a city.', args_schema=<class 'langchain_ibm.toolkit.ToolArgsSchema'>, agent_description='Find the weather for a city.', tool_input_schema={'type': 'object', 'properties': {'location': {'title': 'location', 'description': 'Name of the location', 'type': 'string'}, 'country': {'title': 'country', 'description': 'Name of the state or country', 'type': 'string'}}, 'required': ['location']}, watsonx_client=<ibm_watsonx_ai.client.APIClient object at 0x127e0f490>),
 WatsonxTool(name='RAGQuery', description='Search the documents in a vector index.', args_schema=<class 'langchain_ibm.toolkit.ToolArgsSchema'>, agent_description='Search information in documents to provide context to a user query. Useful when asked to ground the answer in specific knowledge about {indexName}', tool_config_schema={'title': 'config schema for RAGQuery tool', 'type': 'object', 'properties': {'vectorIndexId': {'title': 'Vector index identifier', 'type': 'string'}, 'projectId': {'title': 'Project identifier', 'type': 'string'}, 'spaceId': {'title': 'Space identifier', 'type': 'string'}}, 'required': ['vectorIndexId'], 'oneOf': [{'required': ['projectId']}, {'required': ['spaceId']}]}, watsonx_client=<ibm_watsonx_ai.client.APIClient object at 0x127e0f490>)]

Get a tool

You can also get a specific WatsonxTool by name.

google_search = watsonx_toolkit.get_tool(tool_name="GoogleSearch")

Invocation

Invoke the tool with a simple input

search_result = google_search.invoke(input="IBM")
search_result

{'output': '[{"title":"IBM - United States","description":"Technology & Consulting. From next-generation AI to cutting edge hybrid cloud solutions to the deep expertise of IBM Consulting, IBM has what it takes to help\xa0...","url":"https://www.ibm.com/us-en"},{"title":"IBM - Wikipedia","description":"International Business Machines Corporation (using the trademark IBM), nicknamed Big Blue, is an American multinational technology company headquartered in\xa0...","url":"https://en.wikipedia.org/wiki/IBM"},{"title":"IBM Envizi ESG Suite","description":"Envizi systemizes the capture, transformation and consolidation of disparate sustainability data into a single source of truth and delivers actionable insights.","url":"https://www.ibm.com/products/envizi"},{"title":"IBM Research","description":"Tools + Code · BeeAI Framework. Open-source framework for building, deploying, and serving powerful agentic workflows at scale. · Docling. An open-source tool\xa0...","url":"https://research.ibm.com/"},{"title":"IBM SkillsBuild: Free Skills-Based Learning From Technology Experts","description":"IBM SkillsBuildPower your future in tech with job skills, courses, and credentials—for free. Power your future in tech with job skills, courses, and credentials\xa0...","url":"https://skillsbuild.org/"},{"title":"IBM | LinkedIn","description":"Locations · Primary. International Business Machines Corp. · 590 Madison Ave · 90 Grayston Dr · Plaza Independencia 721 · 388 Phahon Yothin Road · Jalan Prof.","url":"https://www.linkedin.com/company/ibm"},{"title":"International Business Machines Corporation (IBM)","description":"PROFITABILITY_AND_INCOME_STATEMENT · 9.60% · (TTM). 3.06% · (TTM). 24.06% · (TTM). 62.75B · (TTM). 6.02B · (TTM). 6.41. BALANCE_SHEET_AND_CASH_FLOW. (MRQ).","url":"https://finance.yahoo.com/quote/IBM/"},{"title":"Zurich - IBM Research","description":"The location in Zurich is one of IBM\'s 12 global research labs. IBM has maintained a research laboratory in Switzerland since 1956.","url":"https://research.ibm.com/labs/zurich"},{"title":"IBM (@ibm) • Instagram photos and videos","description":"Science, Technology & Engineering. We partner with developers, data scientists, CTOs and other creators to make the world work better.","url":"https://www.instagram.com/ibm/?hl=en"},{"title":"IBM Newsroom","description":"News and press releases from around the IBM world. Media contacts. Sources by topic and by region. IBM Media center. Explore IBM\'s latest and most popular\xa0...","url":"https://newsroom.ibm.com/"}]'}

To fetch a list of received results, you can execute the below cell.

import json

output = json.loads(search_result.get("output"))
output

Invoke the tool with a configuration

To check if a tool has a config schema and view its properties you can look at the tool's tool_config_schema.

In this example, the tool has a config schema that contains maxResults parameter to set maximum number of results to be returned.

google_search.tool_config_schema

{'title': 'config schema for GoogleSearch tool',
 'type': 'object',
 'properties': {'maxResults': {'title': 'Max number of results to return',
   'type': 'integer',
   'minimum': 1,
   'maximum': 20}}}

To set tool_config parameters, you need to use set_tool_config() method and pass correct dict according to above tool_config_schema.

import json

config = {"maxResults": 3}
google_search.set_tool_config(config)

search_result = google_search.invoke(input="IBM")
output = json.loads(search_result.get("output"))

There is supposed to be maximum 3 results.

print(len(output))

Invoke the tool with an input schema

We need to get another tool (with an input schema) for the example purpose.

weather_tool = watsonx_toolkit.get_tool("Weather")

To check if a tool has an input schema and view its properties, you can look at the tool's tool_input_schema.

In this example, the tool has an input schema that contains one required and one optional parameter.

weather_tool.tool_input_schema

{'type': 'object',
 'properties': {'location': {'title': 'location',
   'description': 'Name of the location',
   'type': 'string'},
  'country': {'title': 'country',
   'description': 'Name of the state or country',
   'type': 'string'}},
 'required': ['location']}

To correctly pass an input to invoke(), you need to create an invoke_input dictionary with required parameter as a key with its value.

invoke_input = {
    "location": "New York",
}

weather_result = weather_tool.invoke(input=invoke_input)
weather_result

{'output': 'Current weather in New York:\nTemperature: 0°C\nRain: 0mm\nRelative humidity: 63%\nWind: 7.6km/h\n'}

This time the output is a single string value. To fetch and print it you can execute the below cell.

output = weather_result.get("output")
print(output)

Current weather in New York:
Temperature: 0°C
Rain: 0mm
Relative humidity: 63%
Wind: 7.6km/h

Invoke the tool with a ToolCall

We can also invoke the tool with a ToolCall, in which case a ToolMessage will be returned:

invoke_input = {
    "location": "Los Angeles",
}
tool_call = dict(
    args=invoke_input,
    id="1",
    name=weather_tool.name,
    type="tool_call",
)
weather_tool.invoke(input=tool_call)

ToolMessage(content='{"output": "Current weather in Los Angeles:\\nTemperature: 8.6°C\\nRain: 0mm\\nRelative humidity: 61%\\nWind: 8.4km/h\\n"}', name='Weather', tool_call_id='1')

Use within an agent

from langchain_ibm import ChatWatsonx

llm = ChatWatsonx(
    model_id="meta-llama/llama-3-3-70b-instruct",
    url="https://us-south.ml.cloud.ibm.com",
    project_id="PASTE YOUR PROJECT_ID HERE",
)

API Reference:ChatWatsonx

from langgraph.prebuilt import create_react_agent

tools = [weather_tool]
agent = create_react_agent(llm, tools)

API Reference:create_react_agent

example_query = "What is the weather in Boston?"

events = agent.stream(
    {"messages": [("user", example_query)]},
    stream_mode="values",
)
for event in events:
    event["messages"][-1].pretty_print()

================================[1m Human Message [0m=================================

What is the weather in Boston?
==================================[1m Ai Message [0m==================================
Tool Calls:
  Weather (chatcmpl-tool-6a6c21402c824e43bdd2e8ba390af4a8)
 Call ID: chatcmpl-tool-6a6c21402c824e43bdd2e8ba390af4a8
  Args:
    location: Boston
=================================[1m Tool Message [0m=================================
Name: Weather

{"output": "Current weather in Boston:\nTemperature: -1°C\nRain: 0mm\nRelative humidity: 53%\nWind: 8.3km/h\n"}
==================================[1m Ai Message [0m==================================

The current weather in Boston is -1°C with 0mm of rain, a relative humidity of 53%, and a wind speed of 8.3km/h.

API reference

For detailed documentation of all WatsonxToolkit features and configurations head to the API reference.

Tool conceptual guide
Tool how-to guides

Overview​

Integration details​

Setup​

Credentials​

Installation​

Instantiation​

Tools​

Get all tools​

Get a tool​

Invocation​

Invoke the tool with a simple input​

Invoke the tool with a configuration​

Invoke the tool with an input schema​

Invoke the tool with a ToolCall​

Use within an agent​

API reference​

Related​

Was this page helpful?