Clever AI Humanizer

Clever AI Humanizer (formerly hold your voice, abbreviated hyv) is an AI text de-tracing and personal style reinforcement tool for writers. Its core goal is to make written text sound like the author themselves, rather than templated AI output. According to the official description, hyv scans drafts, flags more than 250 AI writing traces, and rewrites them line by line in the user's own tone.

Purpose

hyv mainly solves the problem of "readers recognizing AI writing within seconds and stopping reading." The official view is that readers do not need to run a detector; they will leave simply because the text does not feel human-written, and hyv is responsible for finding and correcting the parts that sound generalized, mechanical, or fake before the content is published.

Main capabilities

• Local scanning: hyv runs on the user's own machine, and drafts are not uploaded. The official description explicitly states that each scan has zero API calls and zero uploads, and only the voice profile and learning signals are stored server-side.
• 250+ AI trace detection: Covers multiple dimensions such as vocabulary, rhythm, tone, and structure, flagging line by line the places that sound generalized.
• Personal voice profile: Builds a profile from the user's real writing samples, recording vocabulary, sentence rhythm, emotional tone, signature phrases, and banned patterns. Every accepted or rejected rewrite makes the profile more accurate, gradually forming a "never use" list.
• Usable across applications: One profile can be used in AI applications such as Cursor, Claude, Codex, and Windsurf, and can also be used in the terminal.
• Local fixes and agent rewrites: hyv fix can locally replace the parts that can be replaced; the remaining parts are rewritten by the user's own AI agent combined with the voice profile. The user accepts, rejects, or reinforces each suggestion.
• Multi-scenario integration: Supports scanning as you write in the editor (rescanning on every save), batch scanning Markdown files in an entire folder, and integration into automated workflows such as CI/CD, Cloudflare, and n8n. The official examples mention blocking blog drafts or marketing copy containing AI traces from entering the main branch before merging.

How to use

The basic process given by the official description is:

1. Register and create an account.
2. Build a voice profile from your own actual writing.
3. Save the output of ChatGPT or Claude as a file, or paste it directly into the terminal. Any format is acceptable.
4. hyv scans instantly and marks every line that sounds generalized.
5. Use hyv fix for local replacements, and let the user's own AI agent rewrite the rest according to the voice profile.
6. Accept, reject, or reinforce each rewrite choice to continuously train the profile.

hyv does not call its own large model; instead, it tells the user's AI application how to rewrite, so users complete the work with their own ChatGPT or Claude account without incurring additional subscription costs. The official description mentions that MCP can be automatically configured for Cursor, Claude, Windsurf, etc. during installation, and can also be set up manually via hyv mcp --setup .

Use cases

The official description states that the tool is used daily by founders and writers, and is suitable for:

• Batch scanning of Markdown files such as blogs, documentation, and changelogs.
• Automatically scanning every draft in CI/CD to prevent AI traces from entering production content.
• Scanning while writing in an editor (Cursor, VSCode, or any editor).
• Any AI application where text can be pasted: Claude Code, ChatGPT desktop, Codex, Cursor, Windsurf, etc.

Platforms and limitations

• Runtime environment: The official description mentions that it "runs on any Mac," and it can be completed by opening Terminal and pasting a single command once. Existing information does not clearly list support for Windows or Linux.
• Data boundary: Drafts remain on the user's laptop, with zero API calls and zero uploads during scanning; voice profiles and learning signals are stored server-side.
• Relies on the user's own model: hyv itself does not provide an LLM; rewriting depends on the user's existing AI application and tokens.
• Current status: hold your voice has become part of Ellora. The official page states that "new hyv subscriptions are closed—your writing workflow, voice profile, and drafts continue to work on Ellora." For related questions, the official guidance is to inquire via the contact page.
• Pricing information: The official description mentions a "find your voice" plan at "$1 for the first month," but also states that new subscriptions are closed. Specific billing, renewal pricing, and currently available plans are not fully provided in the existing information.
• Cancellation method: You can cancel from the dashboard, the CLI, or by replying to any billing email; the voice profile always belongs to the user and can be reactivated at any time.

Information not yet provided

The official text does not specify the complete list of supported operating systems, the specific subscription pricing structure, the scope of the free trial, team collaboration features, whether non-Mac desktop environments are supported, or the specific migration arrangements on the Ellora platform. This content is not yet provided in the existing information.