How do I create a chain that selects a tools, runs it, and returns the response of the tool directly without any more processing?

[avalatea]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

def get_tool_results_chain() -> Ingredients_Object:
  """
  This function should return the Ingredients_Object that was created by the tool, rather than take the response of the tool, give it to the llm, and have the llm remake the Ingredients_Object.
  """
  user_query = "Can you tell me what ice cream flavors are the best?"
  llm = <some_llm>
  llm.bind_tools(get_ingredients_tool)
  prompt = <some_prompt>

  return {"user_input": user_query} | prompt | llm


def get_ingredients_tool(food) -> Ingredients_Object:
  db = food_db
  ingredients = db.query(food)
  return ingredients

Description

• Objective: Create a function using Langchain that directly returns the result of a tool (the Ingredients_Object) without involving the LLM in remaking or processing the tool's result.

• Current Implementation:

  • A user query initiates the process.
  • An LLM is set up and is supposed to bind a tool for executing a task.
  • A prompt is presumably formulated to engage the LLM.
  • The process involves the LLM in handling the tool's output, which is not desired.

• Desired Workflow:

  • Step 1: Accept user input and determine the appropriate tool and its arguments.
  • Step 2: Directly use the tool to generate results (e.g., get_ingredients_tool should create the Ingredients_Object).
  • Step 3: Skip any further involvement of the LLM—no processing, no reinterpretation of the tool's output.
  • Expected Output: Return the Ingredients_Object produced by the tool directly.

• Current Challenge:

  • The Langchain setup currently routes the tool’s results through the LLM.
  • This leads to unnecessary processing by the LLM instead of directly yielding the tool's results.

• Goal: Configure Langchain such that the LLM acts only in selecting the appropriate tool and argument, with no further manipulation or remaking of the tool's outcomes.

System Info

langchain==0.3.6
langchain-anthropic==0.2.1
langchain-community==0.3.0
langchain-core==0.3.14
langchain-google-community==2.0.0
langchain-mongodb==0.2.0
langchain-openai==0.2.1
langchain-text-splitters==0.3.0

Python 3.12.7

macos

[feijoes]

@avalatea maybe you could try using the StructuredOutputParser to ensure a specific output (just like the tool returns)?rn)?