`xtrllm` - Extract Structured Data using LLMs

A lightweight Python framework for portable, versioned, reusable LLM extraction tasks.

xtrllm separates two pieces that often get tangled together:

The engine - prompt -> structured output -> validated result
The task - schema + prompt strategy + domain-specific extraction rules

The engine ships with the package. The tasks live wherever your project keeps its extraction logic, and can be loaded at runtime.

Installation

Install the latest stable release from PyPI:

pip install xtrllm

Basic Shape

Register a directory of task files, choose a task and model, then call the extractor with the inputs expected by that task.

from xtrllm import LLMXtractor, load_tasks

load_tasks("eulex/tasks", namespace="eulex")

extractor = LLMXtractor(
    task="eulex/classify_actor",
    model="gpt-4.1-mini",
)

result = extractor("Tribunale di Milano")
print(result.model_dump(mode="json"))
# {"actor_type": "ACT_TYPE_NAT_COURT"}

Each result is a Pydantic model, so downstream code receives typed, schema-validated data rather than raw model text.

For a complete walkthrough, including a full task definition and batch processing with a DataFrame, see the Quickstart.

Citation

If you use this framework in academic research, please cite:

@misc{mandujano2026xtrllm,
  author       = {Mauricio Mandujano Manríquez},
  title        = {`xtrllm` - Extract Structured Data using LLMs},
  year         = {2026},
  howpublished = {\url{https://github.com/mauriciomm7/xtrllm}},
  note         = {GitHub repository}
}

Acknowledgments

This project stands on the shoulders of excellent open-source tools and services:

Extraction Logo Icon made by Freepik from Flaticon.
CI/CD Automation - GitHub Actions Powered by GitHub Actions, enabling automated build, test, and deployment pipelines directly within the GitHub ecosystem.
LLM CLI & Python Library llm by Simon Willison - a CLI tool and Python library for interacting with OpenAI, Anthropic Claude, Google Gemini, Meta Llama, and local language model APIs.
Pydantic Built with Pydantic - data validation and settings management using Python type annotations, enabling robust schema enforcement and structured data handling.

License

This project is licensed under the MIT License.

xtrllm - Extract Structured Data using LLMs