hamtaa-texttools 1.1.7

A high-level NLP toolkit built on top of modern LLMs.

TextTools

📌 Overview

TextTools is a high-level NLP toolkit built on top of modern LLMs.

It provides both sync (TheTool) and async (AsyncTheTool) APIs for maximum flexibility.

It provides ready-to-use utilities for translation, question detection, keyword extraction, categorization, NER extraction, and more — designed to help you integrate AI-powered text processing into your applications with minimal effort.

---

✨ Features

TextTools provides a rich collection of high-level NLP utilities built on top of LLMs.
Each tool is designed to work with structured outputs (JSON / Pydantic).

• categorize() - Classifies text into Islamic studies categories
• is_question() - Binary detection of whether input is a question
• extract_keywords() - Extracts keywords from text
• extract_entities() - Named Entity Recognition (NER) system
• summarize() - Text summarization
• text_to_question() - Generates questions from text
• merge_questions() - Merges multiple questions with different modes
• rewrite() - Rewrites text with different wording/meaning
• subject_to_question() - Generates questions about a specific subject
• translate() - Text translation between languages
• run_custom() - Allows users to define a custom tool with arbitrary BaseModel

---

⚙️ with_analysis, logprobs, output_lang, user_prompt, temperature and validator parameters

TextTools provides several optional flags to customize LLM behavior:

• with_analysis=True → Adds a reasoning step before generating the final output. Useful for debugging, improving prompts, or understanding model behavior.
  Note: This doubles token usage per call because it triggers an additional LLM request.

• logprobs=True → Returns token-level probabilities for the generated output. You can also specify top_logprobs=<N> to get the top N alternative tokens and their probabilities.

• output_lang="en" → Forces the model to respond in a specific language. The model will ignore other instructions about language and respond strictly in the requested language.

• user_prompt="..." → Allows you to inject a custom instruction or prompt into the model alongside the main template. This gives you fine-grained control over how the model interprets or modifies the input text.

• temperature=0.0 → Determines how creative the model should respond. Takes a float number from 0.0 to 1.0.

• validator=validation_function → Forces TheTool to validate the output result based on your custom validator. Validator should return bool (True if there were no problem, False if the validation failed.) If validator failed, TheTool will retry to get another output by modifying temperature.

All these parameters can be used individually or together to tailor the behavior of any tool in TextTools.

Note: There might be some tools that don't support some of the parameters above.

---

🧩 ToolOutput

Every tool of TextTools returns a ToolOutput object which is a BaseModel with attributes:

• result → The output of LLM (type=Any)
• analysis → The reasoning step before generating the final output (type=str)
• logprobs → Token-level probabilities for the generated output (type=list)
• errors → Any error that have occured during calling LLM (type=str)

None: You can use repr(ToolOutput) to see details of an output.

---

🚀 Installation

Install the latest release via PyPI:

pip install -U hamtaa-texttools

---

Sync vs Async

Tool	Style	Use case
TheTool	Sync	Simple scripts, sequential workflows
AsyncTheTool	Async	High-throughput apps, APIs, concurrent tasks

---

⚡ Quick Start (Sync)

from openai import OpenAI
from texttools import TheTool

# Create your OpenAI client
client = OpenAI(base_url = "your_url", API_KEY = "your_api_key")

# Specify the model
model = "gpt-4o-mini"

# Create an instance of TheTool
the_tool = TheTool(client=client, model=model)

# Example: Question Detection
detection = the_tool.is_question("Is this project open source?", logprobs=True, top_logprobs=2)
print(detection.result)
print(detection.logprobs)
# Output: True + logprobs

# Example: Translation
translation = the_tool.translate("سلام، حالت چطوره؟" target_language="English", with_analysis=True)
print(translation.result)
print(translation.analysis)
# Output: "Hi! How are you?"  + analysis

---

⚡ Quick Start (Async)

import asyncio
from openai import AsyncOpenAI
from texttools import AsyncTheTool

async def main():
    # Create your AsyncOpenAI client
    async_client = AsyncOpenAI(base_url="your_url", api_key="your_api_key")

    # Specify the model
    model = "gpt-4o-mini"

    # Create an instance of AsyncTheTool
    async_the_tool = AsyncTheTool(client=async_client, model=model)
    
    # Example: Async Translation and Keyword Extraction
    translation_task = async_the_tool.translate("سلام، حالت چطوره؟", target_language="English")
    keywords_task = async_the_tool.extract_keywords("Tomorrow, we will be dead by the car crash")

    (translation, keywords) = await asyncio.gather(translation_task, keywords_task)
    print(translation.result)
    print(keywords.result)

asyncio.run(main())

---

👍 Use Cases

Use TextTools when you need to:

• 🔍 Classify large datasets quickly without model training
• 🌍 Translate and process multilingual corpora with ease
• 🧩 Integrate LLMs into production pipelines (structured outputs)
• 📊 Analyze large text collections using embeddings and categorization

---

📚 Batch Processing

Process large datasets efficiently using OpenAI's batch API.

Quick Start

from texttools import BatchJobRunner, BatchConfig

# Configure your batch job
config = BatchConfig(
    system_prompt="Extract entities from the text",
    job_name="entity_extraction",
    input_data_path="data.json",
    output_data_filename="results.json",
    model="gpt-4o-mini"
)

# Define your output schema
class Output(BaseModel):
    entities: list[str]

# Run the batch job
runner = BatchJobRunner(config, output_model=Output)
runner.run()

---

🤝 Contributing

Contributions are welcome!
Feel free to open issues, suggest new features, or submit pull requests.

---

License

This project is licensed under the MIT License - see the LICENSE file for details.