ahmedak/defluff

What it does

Defluff is a specialized tool designed to identify and flag common filler phrases, clichés, and low-quality patterns typically found in AI-generated content. Unlike model-based detectors, it uses a curated phrase list to ensure deterministic results with zero false positives by design. The tool analyzes text and returns exact character spans where problematic phrases occur, along with a confidence score (0-1) for each detection.

How to set up

Defluff can be deployed in multiple ways: as a CLI tool for direct command-line usage, as an MCP server for integration with AI agents, or as a CI/CD component with exit codes for automated workflows. Installation involves cloning the repository and running with Python 3. No API keys or external services are required as it operates entirely offline.

Key features

• Deterministic detection using curated phrase lists (not ML-based)
• Zero false positives by design
• Returns exact character spans for detected issues
• Confidence scoring (0-1 scale)
• Completely offline operation - no API calls needed
• CLI interface for direct usage
• CI/CD integration with exit codes
• MCP server mode for agent integration
• Cross-platform support (Windows, macOS, Linux)