Discover and explore top open-source AI tools and projects—updated daily.
devswhaAI-writing humanizer for multiple languages
Top 98.3% on SourcePulse
Summary Patina is a deterministic, pattern-based AI text humanizer for KO/EN/ZH/JA. It targets users of AI-assisted drafting, aiming to strip detectable "AI packaging" while preserving core meaning, facts, numbers, polarity, and causation. The benefit is a cleaner authorial voice and an audit trail without altering claims.
How It Works The system employs semantic anchor extraction, stylometry, and an AI-lexicon scan to identify AI-like phrasing. A pattern-guided rewrite engine then modifies the text, followed by self-audits using Mean Opinion Score (MPS) and fidelity checks. If meaning drifts, changes are retried or rolled back. This deterministic approach ensures meaning preservation, differentiating it from black-box paraphrasers.
Quick Start & Requirements
Immediate use is available via a browser playground (patina.vibetip.help). Installation options include an agent-friendly script (install.sh) or a Claude Code plugin (/plugin marketplace add devswha/patina). A Node.js CLI (requires Node.js >= 18) is accessible via npx patina-cli. Free usage is supported by logging into local CLIs like codex, claude, or gemini without API keys. Key documentation links are provided for Cookbook, CLI, and Authentication.
Highlighted Details
rewrite, verify, audit, score, and diff.--persona (including custom personas from samples), --tone, and --profile.devswha/patina-action@v1) and supports batch file processing.Maintenance & Community
Hosted on GitHub (devswha/patina), the project's maintenance and community activity can be tracked via its repository. It draws inspiration from established architectures like oh-my-zsh and AI writing research, suggesting potential areas for contribution or collaboration.
Licensing & Compatibility Released under the permissive MIT license, Patina offers broad compatibility for commercial use and integration within closed-source projects without significant restrictions.
Limitations & Caveats Scoring functions act as editing signals, not definitive authorship proof, and include known false positives/negatives. The free tier is rate-limited, and local CLI backends have conservative caps for batch safety.
22 hours ago
Inactive
blader