Tool Calling with Ollama

Installation

1. Install python. https://www.python.org/downloads/
   (You may skip step 3 if you don't plan to serve LLM locally)
2. Install Ollama.
   1. https://ollama.com/download
   2. After installing, pull a model file, like ollama pull llama3.2
   3. Model files that currently support tool usage can be found here https://ollama.com/search?c=tools
   4. Make sure to check whether the model fits snuggly on your GPU for good performance
3. Install PyTorch for your system. https://pytorch.org/get-started/locally/
4. Pull this repo.
   1. Run pip install -e . in the root directory of the repo.
   2. If you run into problems add the repo root dir to PYTHONPATH.

Tool Calling

There are examples of how to use this in /ToolCalling/Examples/

1. In the Lib module there is a class called ToolChain. Import it and add modules or objects that hold methods and properties to an instance of it.

import ToolCalling.Examples.tool_list as tool_list
from Lib.function_schematizer import ToolChain

tool_chain = ToolChain()
tool_chain.add_module(tool_list)

# Make sure to use type hints and docstrings for the agent to be able to understand them

Examples of tools to try

• API calls
• Calculator
• Pass the user to someone else
• import playwright and go bananas

Workshop

Look through the different directories and get inspired by the examples

Then choose your own project and implement it!

Topic Suggestions:

• Refine answer by asking the user to clarify what they need.
• Solve math problems by giving the agent a calculator tool and maybe some step by step reasoning.
• Make a bot that translates text and then iteratively improves it.
• Give the bot access to your API of choice.
• import playwright and let the LLM control a web page