talkabout 0.1.3

AI-powered object query proxy using Claude AI

talkabout

(inspired by https://github.com/awwaiid/gremllm)

Ask anything about any Python object using natural language and get Python code executed automatically. Passes the object's repr() and your query to Claude API. For now, Claude responds to the system query with single-line code statements only

Installation

pip install talkabout

Setup

You'll need an Anthropic API key. Set it as an environment variable:

export ANTHROPIC_API_KEY="your-api-key-here"

Or pass it directly when creating a Talk instance:

from talkabout import Talk
talk = Talk(my_object, api_key="your-api-key-here")

Usage

A simple example:

import numpy as np
from talkabout import Talk

x = np.random.uniform(size=50)

talk = Talk(x)
talk('90th percentile')

# printed output
Executing code: np.percentile(obj, 90)

Out[1]: np.float64(0.8442100946036629)

You can also use the .chat() method to just ask a question and get a textual response:

import torch
from talkabout import Talk

optimizer = torch.optim.Adam([torch.Tensor([0.0])], lr=0.001)
talk = Talk(optimizer)
talk.chat('is this the best optimiser ever?')

# printed output
Out[6]: 'Let me help you evaluate the Adam optimizer:\n\nAdam is generally considered ...'

A more complicated example - inspect financials of a company using Yahoo Finance API (this use-case actually motivated this package, because those yfinance dataframes are messy):

import yfinance as yf
from talkabout import Talk

pypl = yf.Ticker('PYPL')

talk = Talk(pypl)
talk('qoq Oper CF over debt; use .loc')

# printed output
Executing code: pypl.quarterly_cash_flow.loc['Operating Cash Flow'].pct_change() / pypl.quarterly_balancesheet.loc['Total Debt']

Out[6]:
2025-03-31             NaN
2024-12-31    1.076823e-10
2024-09-30   -3.265984e-11
2024-06-30   -5.669014e-12
2024-03-31    2.654644e-11
2023-12-31             NaN
dtype: float64