AI Functions

class loopgpt.aifunc.aifunc

This class implements a decorator used to create AI functions. Functions decorated by @loopgpt.aifunc() need only have a valid function signature and docstring. The function can then be called normally, and the AI agent will emulate the function’s execution and return a valid python object.

Parameters:

tools (List[BaseTool], optional) – List of tools to be used by the agent to collect data. Defaults to None.

An AI function is executed by an agent.

  • If an agent argument is passed to the decorated function, that agent will be used for execution.

  • If such an argument is not passed and the AI function is called in the with context of an agent, that agent will be used for execution.

  • If neither of the above conditions are met, a new agent will be created and used for execution.
    • The model and embedding provider to be used by this new agent can be passed as arguments to the decorated function:

    • If these arguments are not passed and loopgpt.set_aifunc_args() has been called, the model and embedding provider passed to that function will be used globally.

    • If neither of the above conditions are met, the default model (GPT-3.5-Turbo) and embedding provider (OpenAIEmbeddingProvider) will be used.

Note

Always create agents using loopgpt.empty_agent when using them in conjunction with AI functions.

Examples

>>> @loopgpt.aifunc()
... def shakespearify(text: str) -> str:
...     '''Applies a shakespearian style to the given text and returns it.
...
...     Args:
...         text (str): Text to apply shakespearian style to.
...
...     Returns:
...         str: Text with shakespearian style.
...
...     '''
...
>>> shakespearify("Hey man, how you doin? I was just heading to the store ya know")
'Hark, good sir! How art thou faring? I was but making my way to the market, dost thou know.'

>>> @loopgpt.aifunc(tools=[GoogleSearch])
... def find_age(celeb: str) -> int:
...     '''Searches Google for the celebrity's age and returns it.
...
...     Args:
...         celeb (str): Name of the celebrity.
...
...     Returns:
...         int: Age of the celebrity.
...
...     '''
...
>>> find_age("Robert De Niro") + find_age("Al Pacino")
162

Using AI functions in the with context of an agent:

>>> import loopgpt
>>> from loopgpt.tools import GoogleSearch, Browser
>>> @loopgpt.aifunc()
... def outline_maker(topic: str) -> str:
...     '''Writes an outline of the given topic.
...
...     Args:
...         topic (str): Topic to write an outline about.
...
...     Returns:
...         str: Outline of the topic.
...
...     '''
...
>>> search = GoogleSearch()
>>> browser = Browser()
>>> agent = loopgpt.empty_agent()
>>>
>>> with agent:    # the agent will "watch" the searching and the browsing in the with block
>>>     results, links = search("SVB Banking Crisis")
>>>     for i in range(2):
>>>         browser(links[i])
>>>
>>>     outline = outline_maker("SVB Banking Crisis")    # this AI function can access the memory of 'agent'
>>> print(outline)
1. The collapse of Silicon Valley Bank (SVB) and its impact on the crypto market.
2. The closure of SVB leading to a bank run at Signature Bank.
3. Regulators intervening to prevent a larger financial meltdown.
4. The FDIC attempting to make all depositors whole, regardless of insurance.
5. Government investigation into SVB's failure and stock sales by financial officers.
6. Moody's downgrading the outlook on the U.S. banking system.
7. Other banks being placed under review for a downgrade.
8. Proposed legislation by Sen. Elizabeth Warren and Rep. Katie Porter to strengthen bank regulations.
9. Banking crisis reaching Europe with Credit Suisse losing share value.