byakugan

byakugan is a private, open-source AI text detection tool that runs entirely on the user's own machine, designed to help users determine whether text reads like it was AI-generated and to improve its readability. The project is free and open source.

Purpose

byakugan is used to detect the AI similarity of text and provide improvement suggestions. It emphasizes that AI text detection is a signal rather than a verdict, and that it is biased against non-native English writers. Detection results are presented as likelihood and confidence, rather than a simple yes/no judgment. Stiff, uniform writing full of clichés usually scores higher; diverse, personalized writing scores lower. The suggestions are only for use on one's own work.

Main capabilities

• AI detection: Assesses the degree to which text reads like AI, and provides an explanation in plain language.
• AI-characteristic phrase identification: Points out specifically which phrases read like AI.
• Grammar checking: Performs grammar checking on the text.
• Reuse checking: Compares the text against the user's own documents to check for reuse.
• Local operation: All processing is done locally, protecting privacy.

How to use

byakugan offers two ways to use it:

1. Use in the browser

This is a static web application. After the user pastes text, they can get an AI similarity assessment, a plain-language explanation, grammar checking, and a reuse check against their own documents. All processing is done within the page, and the pasted content does not leave the browser.

How to run locally: (No specific commands are provided in the materials, so none are provided for now)

2. Use in an existing AI assistant

This is an MCP server, suitable for AI assistants (such as Claude Desktop, an IDE, or any MCP client). It provides the AI with detection results as well as specific AI-characteristic phrases, letting the user's own AI check and rewrite text locally.

For Claude Desktop, add the following to claude_desktop_config.json and restart: (No specific configuration content is provided in the materials, so none is provided for now)

Claude Code users can install it as a plugin:

/plugin marketplace add JayOfemi/byakugan
/plugin install ai-checker@ai-checker

Then tell the AI "use ai-checker to check this text" and paste the text to be checked. For complete setup, tools, and verification methods, see mcp/README.md.

Project structure

• site/: browser checker, a static web application. See site/README.md for details.
• mcp/: MCP server and TypeScript library. See mcp/README.md for details.

Each folder is an independent package with its own build. Requires Node 22 (.nvmrc).

Use cases

• The user wants to privately check whether their own writing reads like AI.
• The user wants to integrate detection functionality into an AI assistant so they can check and rewrite text locally.
• The user needs to check whether text has reuse with their own documents.

Limitations

• AI text detection is a signal rather than a verdict, and it is biased against non-native English writers.
• Results are presented as likelihood and confidence, rather than a simple yes/no judgment.
• The suggestions are only for use on one's own work.
• The specific local run commands and Claude Desktop configuration content are not provided in the materials.

License

MIT. Copyright (c) 2026 Jay Ofemi.