datatune - GitLite

KO | EN

gitlite — search

#javascript #python #hacktoberfest #react #go #deep-learning #java #awesome #cli #kubernetes #security #pytorch #ios #open-source

datatune

(★ 151)

Agentic data transformation on infinite amounts of data

File Explorer

Download Latest Version (.zip)

- - docs.yml
  - tests.yml
- - __init__.py
  - agent.py
  - runtime.py
- - deduplication.py
  - filter.py
  - map.py
  - reduce.py
- - __init__.py
  - llm.py
  - model_rate_limits.py
- __init__.py
- logger.py
- - Agent.md
  - conf.py
  - DataLoaders.md
  - Filter.md
  - index.md
  - LLM.md
  - Map.md
  - Op.md
  - Reduce.md
- make.bat
- Makefile
- agents.ipynb
- data_anonymization.ipynb
- deduplication.ipynb
- Getting_started.ipynb
- ibis_core.ipynb
- - combined.csv
  - dask_only.csv
  - datatune_only.csv
  - test_filter.csv
  - test_filter_result.csv
  - test_map.csv
- ibis_core_test.py
- test_agent.py
- test_combination.py
- test_error.py
- test_filter.py
- test_ibis_core.py
- test_map.py
- test_optimized_batch_completion.py
.gitignore
CNAME
how it works.png
LICENSE
pyproject.toml
README.md

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# 🎵 Datatune
[![PyPI version](https://img.shields.io/pypi/v/datatune.svg)](https://pypi.org/project/datatune/)
[![License](https://img.shields.io/github/license/vitalops/datatune)](https://github.com/vitalops/datatune/blob/main/LICENSE)
[![PyPI Downloads](https://static.pepy.tech/badge/datatune)](https://pepy.tech/projects/datatune)
[![Docs](https://img.shields.io/badge/docs-docs.datatune.ai-blue)](https://docs.datatune.ai)
[![Discord](https://img.shields.io/badge/Discord-7289da?logo=discord&logoColor=white)](https://discord.gg/3RKA5AryQX)
[![Paper](https://img.shields.io/badge/paper-DATA%202026-b31b1b)](https://www.scitepress.org/Link.aspx?doi=10.5220%2f0015162300004091)

Scalable Data Transformations with row-level intelligence.

Datatune is not just another Text to SQL tool. With datatune, LLMs and Agents can have full access to infinite amount of data, and apply semantic intelligence in every record.

## How It Works
[Click here to understand how Datatune works](https://blog.datatune.ai/how-to-make-llms-work-on-large-amounts-of-data)

## Paper
Datatune's framework and evaluation are described in our peer-reviewed paper, **["Beyond Context Windows: Data Transformations at Scale with LLMs"](https://www.scitepress.org/Link.aspx?doi=10.5220%2f0015162300004091)**, presented at DATA 2026 (15th International Conference on Data Science, Technology and Applications), Porto, Portugal.

## Installation
```bash
pip install datatune
```
## Quick Start
```python
import datatune as dt
from datatune.llm.llm import OpenAI
import dask.dataframe as dd
llm = OpenAI(model_name="gpt-3.5-turbo")
df = dd.read_csv("products.csv")
# Extract categories using natural language
mapped = dt.map(
    prompt="Extract categories from the description and name of product.",
    output_fields=["Category", "Subcategory"],
    input_fields=["Description", "Name"]
)(llm, df)
# Filter with simple criteria
filtered = dt.filter(
    prompt="Keep only electronics products",
    input_fields=["Name"]
)(llm, mapped)
# Save results
result = dt.finalize(filtered)
result.compute().to_csv("electronics_products.csv")
```
## 🤖 Agents - Even Simpler
Let AI automatically figure out the transformation steps for you:
```python
import datatune as dt
from datatune.llm.llm import OpenAI
llm = OpenAI(model_name="gpt-3.5-turbo")
agent = dt.Agent(llm)
# Just describe what you want - the agent handles map, filter, and more
df = agent.do("Add ProfitMargin column and keep only African organizations", df)
result = dt.finalize(df)
```
The agent automatically:
- Determines which operations to use (map, filter, etc.)
- Chains multiple transformations
- Handles complex multi-step tasks from a single prompt
- Generates and executes Python code along with row-level primitives (Map, Filter, etc) if required. 
## Supported LLMs
```python
# OpenAI
from datatune.llm.llm import OpenAI
llm = OpenAI(model_name="gpt-3.5-turbo")
# Ollama (local)
from datatune.llm.llm import Ollama
llm = Ollama()
# Azure
from datatune.llm.llm import Azure
llm = Azure(model_name="gpt-3.5-turbo", api_key=api_key)
```
## Data Sources
Works with **Dask** and **Ibis** (DuckDB, PostgreSQL, BigQuery, and more):
```python
# Dask
import dask.dataframe as dd
df = dd.read_csv("data.csv")
# Ibis + DuckDB
import ibis
con = ibis.duckdb.connect("data.duckdb")
table = con.table("my_table")
```
## Learn More
- **[Documentation](https://docs.datatune.ai/)** - Complete guides and API reference
- **[Paper](https://www.scitepress.org/Link.aspx?doi=10.5220%2f0015162300004091)** - "Beyond Context Windows: Data Transformations at Scale with LLMs" (DATA 2026)
- **[Examples](https://github.com/vitalops/datatune/tree/main/examples)** - Real-world use cases
- **[Discord](https://discord.gg/3RKA5AryQX)** - Community support
- **[Issues](https://github.com/vitalops/datatune/issues)** - Report bugs or request features

## Citation
If you use Datatune in your research, please cite:
```bibtex
@conference{data26,
author={Abhigith Abraham and Fariz Rahman and Fadil Rahman and Monika Janchevska and Tanaya Amar},
title={Beyond Context Windows: Data Transformations at Scale with LLMs},
booktitle={Proceedings of the 15th International Conference on Data Science, Technology and Applications - Volume 1: DATA},
year={2026},
pages={910-921},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0015162300004091},
isbn={978-989-758-854-9},
issn={2184-285X},
}
```

## License
MIT License