LLM Classifier GUI

A desktop GUI application for text classification using LLMs. Supports multiple inference backends: Ollama, vLLM, SGLang, and llama.cpp.

Built with Flet and powered by ollama-classifier.

Features

• Multiple backends: Ollama, vLLM, SGLang, llama.cpp (local or remote)
• Load data from CSV or Excel files
• Flexible classification schema: define labels manually or load from a CSV/Excel file
• Two classification methods:
  • Classify: single-call prediction with confidence score
  • Score: multi-call evaluation with softmax probabilities for all labels
• Output format options:
  • Top label only: prediction + confidence
  • All labels: each label becomes a column with its probability
• Batch size control: process multiple texts concurrently for speed
• Save results to Excel (merged with original data)
• Dark/Light mode toggle
• Secure storage for API keys

Prerequisites

1. Python 3.11+
2. At least one inference backend:
   • Ollama (local)
   • vLLM server
   • SGLang server
   • llama.cpp server

Linux system dependencies

Installation

# Via uvx (recommended — no installation needed)
uvx ollama-classifier-gui

# Or via pip
pip install ollama-classifier-gui
ollama-classifier-gui

Run from source

uv sync
uv run python -m ollama_classifier_gui.main

Usage

1. Settings tab

Configure your inference backend:

• Select the backend type (Ollama, vLLM, SGLang, llama.cpp)
• Set the endpoint URL (auto-fills with defaults)
• Enter the model name (or use "Test Connection" to auto-detect available models)
• Optionally set an API key for authenticated remote servers
• Toggle dark/light mode
• Click Save Settings

2. Data Input tab

• Click Select File to load a CSV or Excel file
• For Excel files, select the sheet from the dropdown
• Select the text column that contains the text to classify

3. Schema tab

Define your classification labels:

• Manual Labels: click "Add Label" to add labels with optional descriptions
• Load from File: select a CSV/Excel file containing labels (and optionally descriptions)

Then choose:

• Classification Method: Classify (fast) or Score (accurate probabilities)
• Output Format: Top label only or All labels (each as a column)
• Optionally override the system prompt

4. Results tab

• Click Run Classification to start
• Monitor progress in real-time
• Click Save Results to export to Excel (original data + classification columns)

5. Info tab

• Shows the application version and a link to the GitHub repository (both read from the package metadata defined in pyproject.toml)
• Click "Open in Browser" to open the repository in your default browser

Configuration

Settings are stored in a JSON config file:

API keys are stored separately using the OS native secure storage (Keychain, Credential Manager, libsecret).

Default Endpoints

License

MIT

Change Log

• 0.6.0 — Version bump
  • Bump project version to 0.6.0 across pyproject.toml, __init__.py, and uv.lock
  • Align ollama-classifier dependency to >=0.6.0
• 0.2.1 — Fix launch & runtime bugs
  • Fix the Schema tab "Classification Labels" card that failed to render (Tabs/TabBarView now have bounded height; tab switching wired via on_change)
  • Fix "Save Results" Excel export: remove invalid engine argument and add xlsxwriter (the polars 1.40 writer backend)
  • Rename entry module app.py → main.py (run directly with python -m ollama_classifier_gui.main)
  • Resolve all Flet deprecation warnings (ElevatedButton→Button, border.all→Border.all, margin.only→Margin.only) to prepare for Flet 1.0
  • Add tests/ smoke tests; fix CI Python matrix (3.11/3.12)
• 0.2.2 — Add Info section
  • New "Info" navigation tab showing the application version and GitHub repository link (sourced from pyproject.toml via importlib.metadata)