Quick Start Guide

Basic Usage

1. Create a Toolbox:

   from ai_agent_toolbox import Toolbox, XMLParser, XMLPromptFormatter
   from examples.util import anthropic_llm_call, anthropic_stream
   
   toolbox = Toolbox()
   parser = XMLParser()
   formatter = XMLPromptFormatter()
   

2. Define and Add Tools:

   def thinking(thoughts=""):
       print(f"Thinking: {thoughts}")
   
   toolbox.add_tool(
       name="thinking",
       fn=thinking,
       args={
           "thoughts": {
               "type": "string",
               "description": "Thoughts to process"
           }
       },
       description="For thinking out loud"
   )
   

3. Set up your system prompt:

   system = "You are a thoughtful AI.\n"
   system += formatter.usage_prompt(toolbox) # Add the instructions to use the tools in the toolbox
   prompt = "Think about something"
   

4. Generate a response from your LLM:

   response = anthropic_llm_call(
       prompt=prompt,
       system_prompt=system,
   )
   

5. Parse the response:

   events = parser.parse(response)
   
   for event in events:
       toolbox.use(event)
   

6. (optional) Use async streaming for calling tools ASAP:

   import asyncio
   
   async def main():
       # Stream the response
       async for chunk in anthropic_stream(system, prompt):
           for event in parser.parse_chunk(chunk):
               if event.is_tool_call:
                   await toolbox.use_async(event)
   
       # Make sure to reset the parser and handle any unclosed tool tags
       events = parser.flush()
   
       # Optionally you can handle unclosed tool tags, or just ignore them
       for event in events:
           if event.is_tool_call:
               await toolbox.use_async(event)
   
   
   asyncio.run(main())