Add Tools to a Workflow

The Customizing a Workflow tutorial demonstrates how to customize a workflow by overriding parameters. This tutorial will demonstrate how to add new tools to a workflow. Adding a new tool to a workflow requires copying and modifying the workflow configuration file, which, in effect, creates a new customized workflow.

AIQ toolkit includes several built-in tools (functions) that can be used in any workflow. To query for a list of installed tools, run the following command:

aiq info components -t function

The current workflow defines a tool to query the LangSmith User Guide. This is defined in the tools section of the configuration file:

functions:
  webpage_query:
    _type: webpage_query
    webpage_url: https://docs.smith.langchain.com
    description: "Search for information about LangSmith. For any questions about LangSmith, you must use this tool!"
    embedder_name: nv-embedqa-e5-v5
    chunk_size: 512

However, the workflow is unaware of some related technologies, such as LangGraph, if you run:

aiq run --config_file examples/simple/configs/config.yml --input "How does LangSmith interact with tools like LangGraph?"

The output will be similar to the following:

Workflow Result:
["Unfortunately, I couldn't find any information about LangSmith's interaction with LangGraph. The user guide does not mention LangGraph, and I couldn't find any relevant information through the webpage queries."]

You can solve this by updating the workflow to also query the LangGraph Quickstart guide.

To do this, create a copy of the original workflow configuration file. To add the LangGraph query tool to the workflow, update the YAML file updating the functions section from:

functions:
  webpage_query:
    _type: webpage_query
    webpage_url: https://docs.smith.langchain.com
    description: "Search for information about LangSmith. For any questions about LangSmith, you must use this tool!"
    embedder_name: nv-embedqa-e5-v5
    chunk_size: 512

to:

functions:
  langsmith_query:
    _type: webpage_query
    webpage_url: https://docs.smith.langchain.com
    description: "Search for information about LangSmith. For any questions about LangSmith, you must use this tool!"
    embedder_name: nv-embedqa-e5-v5
    chunk_size: 512
  langgraph_query:
    _type: webpage_query
    webpage_url: https://langchain-ai.github.io/langgraph/tutorials/introduction
    description: "Search for information about LangGraph. For any questions about LangGraph, you must use this tool!"
    embedder_name: nv-embedqa-e5-v5
    chunk_size: 512

Since you now have two instances of the webpage_query tool, you need to update the name of the first tool to langsmith_query.

Finally, update the workflow.tool_names section to include the new tool from:

workflow:
  _type: react_agent
  tool_names: [webpage_query, current_datetime]

to:

workflow:
  _type: react_agent
  tool_names: [langsmith_query, langgraph_query, current_datetime]

Note

The resulting YAML is located at examples/documentation_guides/workflows/custom_workflow/custom_config.yml in the AIQ toolkit repository.

When you rerun the workflow with the updated configuration file:

aiq run --config_file examples/documentation_guides/workflows/custom_workflow/custom_config.yml \
  --input "How does LangSmith interact with tools like LangGraph?"

We should receive output similar to:

Workflow Result:
['LangSmith interacts with LangGraph as part of an out-of-the-box solution for building complex, production-ready features with LLMs. LangGraph works in conjunction with LangSmith to provide this solution, and they are both part of the LangChain ecosystem.']

Alternate Method Using a Web Search Tool

Adding individual web pages to a workflow can be cumbersome, especially when dealing with multiple web pages. An alternative method is to use a web search tool. One of the tools available in AIQ toolkit is the tavily_internet_search tool, which utilizes the Tavily Search API.

The tavily_internet_search tool is part of the aiqtoolkit[langchain] package, to install the package run:

# local package install from source
uv pip install -e '.[langchain]'

Prior to using the tavily_internet_search tool, create an account at tavily.com and obtain an API key. Once obtained, set the TAVILY_API_KEY environment variable to the API key:

export TAVILY_API_KEY=<YOUR_TAVILY_API_KEY>

We will now update the functions section of the configuration file replacing the two webpage_query tools with a single tavily_internet_search tool entry:

functions:
  internet_search:
    _type: tavily_internet_search
  current_datetime:
    _type: current_datetime

Next, update the workflow.tool_names section to include the new tool:

workflow:
  _type: react_agent
  tool_names: [internet_search, current_datetime]

The resulting configuration file is located at examples/documentation_guides/workflows/custom_workflow/search_config.yml in the AIQ toolkit repository.

When you re-run the workflow with the updated configuration file:

aiq run --config_file examples/documentation_guides/workflows/custom_workflow/search_config.yml \
  --input "How does LangSmith interact with tools like LangGraph?"

Which will then yield a slightly different result to the same question:

Workflow Result:
['LangSmith interacts with LangGraph through the LangChain ecosystem, which provides the foundation for building LLM applications. LangGraph provides real-time monitoring, tracing, and debugging capabilities, and it can be used in conjunction with LangSmith to build robust agentic applications.']