Is it possible to define sequence of Tools usage?

[mraguth]

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

tools = [repl, retriever_tool, add, subtract, multiply, divide, modulus, exponentiate, integer_divide,tavily]

Description

In case I have multiple tools in the 'tools', is that possible for me to define the path? Example: I like to use 'retriever_tool' first and 'repl' later... is that possible to define that by human in the code?

tools = [repl, retriever_tool, add, subtract, multiply, divide, modulus, exponentiate, integer_divide,tavily]

System Info

Name: langgraph
Version: 0.2.34

Name: langchain
Version: 0.3.2

[eyurtsev]

Yes, but you should do this by combining together several different chains or agents.

• Use a chain if you know the number of steps that each interaction with the user involves.
• Use agents if you don't know how many steps a given step of the interaction will involve (e.g., maybe the agent will need to use add, multiply, divide and finally multiply again, or maybe the agent will need to use the retrieve tool ~5 times with different queries).

If you need to use agents, then I'd recommend using langgraph for this:

https://langchain-ai.github.io/langgraph/concepts/multi_agent/#multi-agent-collaboration
https://langchain-ai.github.io/langgraph/how-tos/subgraph/

If you can do it using pure chains, then you can just use LCEL and chain everything together

[mraguth]

@eyurtsev Thank you for your suggestion. I am using all tools for one agent. However, I like create another agent and split my tools. Can you help me in giving simple example please? My one agent example is like below...

from langgraph.prebuilt import create_react_agent
from langgraph.prebuilt.chat_agent_executor import AgentState

model = llm

# For this tutorial we will use custom tool that returns pre-defined values for weather in two cities (NYC & SF)

from typing import Literal

from langchain_core.tools import tool
# Define the graph

from langgraph.prebuilt import create_react_agent

prompt1 = ChatPromptTemplate.from_messages(
    [
        ("system", template),
        ("placeholder", "{messages}"),
        MessagesPlaceholder("chat_history", optional=True),
    ]
)

def _modify_state_messages(state: AgentState):
    return prompt.invoke({"messages": state["messages"]}).to_messages() + [
        ("user", "I am AI Agent.")
    ]

graph = create_react_agent(model, tools=tools,state_modifier=prompt1)

from IPython.display import Image, display

display(Image(graph.get_graph().draw_mermaid_png()))

def print_stream(stream):
    for s in stream:
        message = s["messages"][-1]
        if isinstance(message, tuple):
            print(message)
        else:
            message.pretty_print()

inputs = {"messages": [("user", output_query_str)]}
print_stream(graph.stream(inputs, stream_mode="values"))

[eyurtsev]

Hi @mraguth, you should read through the langgraph tutorials to figure out how to build this. It shouldn't be hard, but I don't have time to create examples beyond the documentation we already have (since every user has their own version of what they'd like to build).

I'd recommend starting with the quick start tutorial: https://langchain-ai.github.io/langgraph/tutorials/introduction/

If you don't want to use any advanced features, you should be able to represent what you have as a larger graph that has distinct states.

If you want to use some more advanced features, you can figure out how to capture this via sub-graphs (take a look through the docs to find sub-graphs)