Skip to content

📊 LLM Extractinator Studio

A simple, point-and-click interface to design, test, and run extraction tasks—no coding required. Built with Streamlit, the Studio lets you go from dataset to results in minutes.

🚀 How to Launch

Once installed, open your terminal and run:

launch-extractinator

This starts the app locally and opens it in your browser.

💡 By default, it runs on port 8501. Use --port to choose a different one:

bash launch-extractinator --port 8502

🧭 What You Can Do

The Studio is organized into three main tabs:

1️⃣ Quick‑start

Run an existing Task JSON with a few clicks.

  • Select Task – Dropdown shows all saved tasks in tasks/
  • Preview – Expand to review the JSON
  • ✅ Use this Task – Loads the task and highlights the ▶️ Run tab

📝 Tip: Save and version-control tasks to share or rerun later.

2️⃣ Build Task

Create a new task in 3 guided steps:

🪄 Step-by-step Wizard

Step What it Does
1. Files Pick your dataset, parser, and optional examples
2. Description Add a title and choose which column has the input text
3. Review & Save Preview the full JSON and save it to tasks/

✨ Features:

  • ✅ Supports CSV and JSON datasets
  • ✅ Drag and drop or browse files from your project
  • ✅ Auto-preview for tables and code

🔧 Parser Builder

Still need to build your OutputParser? Click 🛠️ Open Parser Builder in the sidebar.

  • Design a Pydantic schema visually
  • Set types and nesting
  • Export as a ready-to-use OutputParser file

Once saved, place it in tasks/parsers/ and reference it in your Task JSON. For more details, see the Parser documentation.

3️⃣ Run

The final step—run your task and see live output.

Key Features

  • Switch Tasks easily without leaving the tab
  • Choose Model – Any model supported by Ollama or your local setup
  • Enable Reasoning Mode – Toggle special model settings if needed

Advanced Settings (Optional)

Split into two subtabs:
- General – run name, repetitions, seed, verbosity - Sampling & Limits – temperature, top-k, context length, examples

Run & Monitor

  • Click 🚀 Run to start
  • Logs stream live in real-time
  • Success or failure reported at the end

📁 Project Structure Overview

project-root/
├── data/            # Datasets (.csv/.json)
├── examples/        # Few-shot examples (optional)
├── tasks/
│   ├── Task001.json # Task configs
│   └── parsers/
│       └── your_parser.py  # OutputParser files

💡 Tips

  • Dark Mode – Enable via ☰ → Settings → Theme
  • Reset Session – Use the sidebar button to clear everything

🙌 Feedback & Contributions

Spotted a bug or want to help improve the Studio?
Open an issue or PR on GitHub—code lives in llm_extractinator/gui.py.

We welcome all contributions! 🎉