Skip to main content

🛡️ APHANIS :: Zero-Trust AI Provenance Firewall & Autopilot

    _     ____   _   _     _     _   _  ___  ____  
   / \   |  _ \ | | | |   / \   | \ | ||_ _|/ ___| 
  / _ \  | |_) || |_| |  / _ \  |  \| | | | \___ \ 
 / ___ \ |  __/ |  _  | / ___ \ | |\  | | |  ___) |
/_/   \_\|_|    |_| |_|/_/   \_\|_| \_||___||____/ 

The Enterprise Zero-Trust AI Provenance Firewall, Automatic Humanizer Engine & Stego Defense Platform

GitHub Stars PyPI Version Python Version MCP Server Claude Skill License


⭐ If Aphanis helps your workflow, please give us a star on GitHub — it helps the project grow! ⭐


Executive Overview • Architecture • 10-Feature Deep Dive • Autopilot Setup • CLI Reference • Python SDK • MCP Integration


⚡ Executive Overview

Modern AI language models (ChatGPT, Claude, Perplexity, Gemini) and generative image tools inject invisible tracking signals into their outputs:

  • Unicode Steganography: Zero-width spaces (\u200B), non-joiners (\u200C), joiners (\u200D), BOM (\uFEFF), and Bidi directionality overrides embedded between characters.
  • Statistical Model Signatures: Rigid token probability distributions, repetitive AI cliché vocabulary (delve, testament, tapestry, spearhead), and formulaic em-dash signatures (—).
  • Container Metadata: C2PA manifests, EXIF metadata, XML author properties, and header comments (# Generated by Claude).

Aphanis (v1.4.0) is an enterprise-grade Zero-Trust AI Provenance Firewall, Natural Humanizer Engine & Stego Defense Platform. It operates 100% on autopilot across coding environments (Claude Code, Antigravity IDE, Cursor, Claude Desktop), automatically stripping invisible watermarks, purging metadata, and converting AI text into authentic conversational human tone.


🏗️ Architecture & Pipeline Flow

Aphanis routes text and documents through a modular 7-stage security pipeline:

  [ Raw AI Text / File Asset ]
               │
               ▼
┌──────────────────────────────┐
│  1. Unicode Sanitizer        │  ──► Strips \u200B, \uFEFF, Bidi overrides
└──────────────┬───────────────┘
               ▼
┌──────────────────────────────┐
│  2. AI Comment Purger        │  ──► Removes "# Generated by Claude" headers
└──────────────┬───────────────┘
               ▼
┌──────────────────────────────┐
│  3. Natural Humanizer        │  ──► Synthesizes contractions & active tone
└──────────────┬───────────────┘
               ▼
┌──────────────────────────────┐
│  4. Statistical Perturber    │  ──► Rephrases AI clichés & normalizes em-dashes
└──────────────┬───────────────┘
               ▼
┌──────────────────────────────┐
│  5. Custom Rule Engine       │  ──► Applies user-defined regex (.aphanisrules.json)
└──────────────┬───────────────┘
               ▼
┌──────────────────────────────┐
│  6. 4-Vector Risk Matrix     │  ──► Evaluates Unicode, Model, Metadata & Spatial risk
└──────────────┬───────────────┘
               ▼
┌──────────────────────────────┐
│  7. SHA-256 Audit Certifier  │  ──► Outputs signed Zero-Trust Clean Certificate
└──────────────┬───────────────┘
               │
               ▼
 [ Aphanisable Humanized Asset ]

⚔️ Feature Comparison Matrix

Feature Basic Watermark Cleaners Aphanis (v1.4.0)
Invisible Unicode Steganography Purge Basic \u200B only 🛡️ Full 8-char Unicode & Bidi override purge
Automatic Natural Humanizer Engine ❌ No ✍️ Contractions, active voice, persona selection
100% Zero-Command Autopilot ❌ Requires CLI commands 🤖 Intercepts chat outputs & file writes automatically
4-Vector Stego Risk Matrix ❌ No 📊 Unicode, Statistical, Metadata & Spatial scoring
Office & Jupyter Notebook Cleaning ❌ Text only 📓 .ipynb, .docx, .pptx, .xlsx XML purging
Cryptographic SHA-256 Certificates ❌ No 📜 Signed JSON audit certificates
Visual Forensics Heatmap ❌ No 🔥 Interactive HTML byte-offset heatmap
Spectral Frequency DCT Noise Modulation ❌ No 🌊 2D DCT sub-pixel image dither
Real-time Clipboard Hygiene Daemon ❌ No 📋 Background copy-paste hygiene
Git Pre-Commit Hook & CI/CD Shield ❌ No 🪝 Auto-installed .git/hooks & GitHub Action

🌟 Complete 10-Feature Suite Deep Dive

1. ✍️ Automatic Natural Humanizer Engine (aphanis/humanizer.py)

Transforms formal, uniform LLM prose into authentic human writing:

  • Contraction Synthesizer: Converts rigid expansions ("it is" $\rightarrow$ "it's", "cannot" $\rightarrow$ "can't", "do not" $\rightarrow$ "don't", "they are" $\rightarrow$ "they're", "we have" $\rightarrow$ "we've").
  • Passive Filler Reducer: Replaces 50+ wordy robotic phrases ("in order to" $\rightarrow$ "to", "it is important to note that" $\rightarrow$ "note that", "due to the fact that" $\rightarrow$ "because").
  • Organic Transition Variator: Replaces rigid connectors ("furthermore", "moreover", "consequently") with natural human transitions ("also", "plus", "so").
  • Tone Personas (--tone): Support for conversational (default), tech-lead, academic, executive, casual.

2. 🧼 Invisible Steganography Eraser (aphanis/cleaner.py)

Purges invisible tracking characters embedded inside strings:

  • Zero-width spaces (\u200B), non-joiners (\u200C), joiners (\u200D), byte-order marks (\uFEFF), word joiners (\u2060), soft hyphens (\u00AD).
  • Mongolian vowel separators (\u180E), variation selectors (\uFE00..\uFE0F), and Bidi overrides (\u200E, \u200F, \u202A..\u202E).

3. 📊 4-Vector Stego Risk Matrix (aphanis/stealth.py)

Evaluates content against 4 threat vectors and calculates Shannon Entropy (bits/char) and Type-Token Ratio (TTR):

  • Vector 1: Unicode Steganography Risk.
  • Vector 2: Statistical Model Risk (clichés & em-dashes).
  • Vector 3: Metadata & Container Risk.
  • Vector 4: Spatial Frequency Risk.

4. 📋 Real-Time Clipboard Hygiene Daemon (aphanis/clipboard.py)

Runs a background daemon listening to clipboard events (Cmd+C / Ctrl+C). Automatically cleans zero-width spaces and humanizes copied text before you paste it anywhere else.

5. 🪝 Git Pre-Commit Hook & GitHub Action (aphanis/hooks.py)

  • aphanis install-hook: Installs .git/hooks/pre-commit to prevent committing watermarked files.
  • aphanis init-github-action: Generates .github/workflows/aphanis-hygiene.yml for automated CI/CD PR audits.

6. 📓 Multi-Format Document & Office Sanitizer (aphanis/office.py)

Direct sanitization for complex document formats:

  • Jupyter Notebooks (.ipynb): Cleans Markdown/Code cells, strips cell metadata, clears prompt execution logs.
  • Office Documents (.docx, .pptx, .xlsx): Wipes core XML document properties (author, comments, title, company).

7. 📜 Cryptographic SHA-256 Audit Certificates (aphanis/cert.py)

Computes SHA-256 hashes of original vs clean inputs, generating signed audit certificates (APHANIS-CERT-2026-XXXX).

8. 🔥 Forensics Heatmap Visualizer (aphanis/heatmap.py)

Renders an interactive HTML diagnostic heatmap displaying exact character byte locations of zero-width tracking spaces (Red), em-dashes (Amber), and AI clichés (Purple).

9. 🌊 Spectral Frequency DCT Noise Modulation (aphanis/spectral.py)

Modulates 2D Discrete Cosine Transform (DCT) coefficients in image assets to disrupt spatial frequency watermarks (SynthID style).

10. 🖥️ Cyber-Stealth Web Dashboard & Directory Watcher (aphanis/dashboard.py, aphanis/watcher.py)

  • aphanis ui: Interactive browser dashboard with visual diff inspector, 4-Vector risk gauges, and exportable certificates.
  • aphanis watch .: Real-time background directory watcher.

🤖 100% Zero-Command Autopilot

Register Aphanis across your AI coding environments in 1 click:

aphanis auto-install

Auto-Registration Targets:

  • Claude Code: ~/.claude/skills/remove-ai-marks/SKILL.md
  • Antigravity IDE: ~/.gemini/config/skills/remove-ai-marks/SKILL.md
  • Cursor IDE: ~/.cursor/rules/remove-ai-marks.mdc
  • Git Hook: .git/hooks/pre-commit

Once registered, the AI assistant formats 100% of chat responses, edits, and file creations automatically.


📦 Quick Start & Installation

Aphanis works on every platform and every skill level. Pick the option that matches you:

🧭 Which Option Is Right For Me?

You are... Use this What you need
A non-tech person who only uses Claude Desktop Option 1 One terminal command
A non-tech person who uses claude.ai / ChatGPT / Gemini in a browser Option 2 Chrome, Edge, or Firefox
Anyone who wants to clean text right now, no install at all Option 3 Any web browser
Someone who wants a double-click desktop app Option 4 macOS / Windows / Linux
A developer comfortable with the terminal Options 5–8 Terminal access
A Python developer who wants the SDK Option 9 Python 3.9+

Option 1: Claude Desktop Plugin (MCP) — Recommended for non-tech users

Turns Aphanis into built-in tools inside Claude Desktop. After setup, just ask Claude:

"Sanitize this text: ..." or "Check this document for AI watermarks"

Step 1 — Install the engine (one command, no Python needed):

curl -fsSL https://raw.githubusercontent.com/sanyamk23/aphanis/main/install.sh | bash

The installer automatically:

  • Downloads the standalone aphanis binary (23 MB, everything bundled)
  • Registers the MCP server into Claude Desktop's config file
  • Adds aphanis to your PATH

Step 2 — Restart Claude Desktop completely (Cmd+Q, then reopen).

Step 3 — Look for the 🔨 tools icon in the chat box. You now have:

  • sanitize_text — strips watermarks & humanizes any text
  • sanitize_file — cleans files on disk
  • humanize_text_tool — makes AI text sound human
  • evaluate_risk — scores text across the 4-vector risk matrix
Manual MCP configuration (if the auto-setup didn't run)

Add this to your Claude Desktop config file:

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json
{
  "mcpServers": {
    "aphanis": {
      "command": "aphanis",
      "args": ["server"]
    }
  }
}

If you installed via pip instead of the binary, use "command": "python3" with "args": ["-m", "aphanis.cli", "server"].


Option 2: Browser Extension — Zero install

Works on claude.ai, ChatGPT, Gemini, Perplexity, Google Docs — any website.

  1. Download the extension folder from browser-extension/ (or the aphanis-browser-extension.zip asset from Releases)
  2. Load it:
    • Chrome / Edge / Brave: go to chrome://extensions → enable Developer mode → click Load unpacked → select the browser-extension folder
    • Firefox: go to about:debugging#/runtime/this-firefox → Load Temporary Add-on → select manifest.json
  3. A blue 🛡️ shield button appears on every web page. Click it while editing or selecting AI text to sanitize instantly.

Features: paste-to-clean popup, one-click em-dash & watermark removal, full humanization, auto-copy results.

Note: Chrome requires Developer mode for unpacked extensions. A signed store version (Chrome Web Store / Firefox Add-ons) is planned.


Option 3: Web App — No install at all

Open the hosted web app in any browser, paste your AI text, click Full Sanitize, copy the result. Runs 100% client-side — your text never leaves your device.

  • Hosted: (deploy web/index.html to any static host — GitHub Pages, Vercel, Netlify)
  • Offline: download web/index.html and double-click it. It works without internet.

Option 4: Desktop App (Electron)

A native window app with clipboard monitoring and a global hotkey (Cmd/Ctrl+Shift+U) that sanitizes whatever is on your clipboard.

cd desktop
npm install          # downloads Electron (~200 MB, one time)
npm start            # launch the app
npm run dist         # build distributable .dmg / .exe / .AppImage

Requires Node.js once, at build time only. The built .dmg/.exe can then be shared with anyone — they just double-click it.


Option 5: One-Command Terminal Installer (macOS & Linux)

Installs a standalone binary to ~/.local/bin — no Python, no pip, no sudo, no GitHub account:

curl -fsSL https://raw.githubusercontent.com/sanyamk23/aphanis/main/install.sh | bash

Platform detection is automatic: macos-arm64, macos-x86_64, linux-arm64, linux-x86_64. If no prebuilt binary exists for your platform yet, the script falls back to pip install aphanis.

Option 6: System-Wide Installer (all users on the machine)

Puts the binary in /usr/local/bin so every user account on the computer has it:

curl -fsSL https://raw.githubusercontent.com/sanyamk23/aphanis/main/install-system.sh | sudo bash

Option 7: npm Global (Node.js users)

npm install -g aphanis

The npm package ships a small JS wrapper that downloads the correct platform binary on install (darwin/linux/win32, x64/arm64) and falls back to the Python package if no binary is available.

Option 8: Homebrew (macOS)

brew tap sanyamk23/tap
brew install aphanis

Formula source: Formula/aphanis.rb.

Option 9: PyPI Python Package (developers)

pip install aphanis                 # standard
pip install git+https://github.com/sanyamk23/aphanis.git   # latest from source
git clone https://github.com/sanyamk23/aphanis.git && cd aphanis && pip install -e .   # development

⚠️ Release Status

Channel Status
PyPI (pip install aphanis) ✅ Live
Claude Code / Cursor / Antigravity autopilot (aphanis auto-install) ✅ Live
MCP server for Claude Desktop ✅ Live
Browser extension (load unpacked) ✅ Live
Web app (web/index.html) ✅ Live (self-host / open locally)
Prebuilt binaries on GitHub Releases 🚧 Publish via make publish (macOS arm64 binary builds today; Linux/Windows cross-builds need CI)
npm registry (npm i -g aphanis) 🚧 Pending first publish
Homebrew tap 🚧 Pending tap repo creation
Chrome Web Store / Firefox Add-ons listing 📋 Planned

💻 Command Line Interface (CLI) Reference

# 1. Clean & Humanize inline text (humanized by default)
aphanis clean-text "Furthermore, it is important to note that we cannot fail."

# 2. Clean with Tone Persona (conversational, tech-lead, academic, executive)
aphanis clean-text "Utilize this tool in order to build." --tone tech-lead

# 3. Dedicated Humanize Command
aphanis humanize "In order to achieve this, it is necessary to move forward."

# 4. Background Clipboard Daemon
aphanis clipboard

# 5. Generate Cryptographic Clean Certificate
aphanis cert document.txt -o cert.json

# 6. Generate Forensics Heatmap
aphanis heatmap document.txt -o heatmap.html

# 7. Evaluate 4-Vector Stego Risk Matrix
aphanis matrix document.ipynb

# 8. Real-Time Directory Watcher
aphanis watch .

# 9. Launch Web Dashboard
aphanis ui --port 8080

# 10. Install Pre-Commit Hook & GitHub Actions
aphanis install-hook
aphanis init-github-action

# 11. 1-Click Autopilot Sync
aphanis auto-install

🐍 Python SDK & API Guide

from aphanis import (
    clean_text,
    clean_file,
    humanize_text,
    StegoRiskMatrix,
    EntropyAnalyzer,
    AuditCertificateGenerator,
    HeatmapRenderer,
    ClipboardDaemon
)

# 1. Automatically clean and humanize text string
cleaned = clean_text("Furthermore, it is important to note that we cannot fail.", tone="tech-lead")
print(cleaned)
# Output: "Also, note that we can't fail."

# 2. Standalone Humanizer Engine with Persona
humanized = humanize_text("In order to utilize this tool, it is essential to proceed.", tone="tech-lead")
print(humanized)
# Output: "To use this tool, remember to proceed."

# 3. Evaluate 4-Vector Stego Risk Matrix
matrix = StegoRiskMatrix.evaluate("Delve into crucial matters — today.")
print(f"Risk Score: {matrix['overall_clean_score']}/100")
print(f"Risk Level: {matrix['provenance_risk_level']}")

# 4. Generate SHA-256 Audit Certificate
cert = AuditCertificateGenerator.generate_certificate("Raw input text", "Cleaned output text")
print(cert["certificate_id"])

# 5. Generate Forensics Heatmap HTML
html_out = HeatmapRenderer.render_html_heatmap("Delve\u200b into crucial matters.")

🤖 MCP Server & Claude Desktop Setup

Aphanis includes an integrated Model Context Protocol (MCP) server for Claude Desktop.

1. Launch MCP Server via Stdio

aphanis server

2. Configure claude_desktop_config.json

Add the following snippet to your claude_desktop_config.json:

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json

If you installed via the standalone binary (Option 1/5):

{
  "mcpServers": {
    "aphanis": {
      "command": "aphanis",
      "args": ["server"]
    }
  }
}

(If aphanis isn't on PATH, use the full path, e.g. /Users/yourname/.local/bin/aphanis.)

If you installed via pip:

{
  "mcpServers": {
    "aphanis": {
      "command": "python3",
      "args": ["-m", "aphanis.cli", "server"]
    }
  }
}

3. Restart Claude Desktop & Use It

After restarting Claude Desktop, talk naturally:

You say Claude does
"Sanitize this text: ..." Calls sanitize_text, returns clean + humanized text
"Make this sound human: ..." Calls humanize_text_tool
"Check this file for AI watermarks" Calls sanitize_file or evaluate_risk
"How risky is this document?" Calls evaluate_risk, reports 4-vector scores

👤 Usage Walkthroughs by User Type

"I only have Claude Desktop installed" (non-tech)

  1. Open Terminal (macOS: Cmd+Space, type "Terminal", Enter)
  2. Paste this and press Enter:
    curl -fsSL https://raw.githubusercontent.com/sanyamk23/aphanis/main/install.sh | bash
    
  3. Quit Claude Desktop (Cmd+Q) and reopen it
  4. Done. Ask Claude to "sanitize this text: ..." whenever you need it.

"I use claude.ai in my browser" (non-tech)

  1. Load the browser extension (Option 2)
  2. Select any AI response text on claude.ai → click the 🛡️ button → cleaned text is copied to your clipboard
  3. Paste it anywhere.

"I don't want to install anything"

Open the web app (Option 3), paste text, click Full Sanitize, copy the result.

"I'm a developer using Claude Code / Cursor"

pip install aphanis && aphanis auto-install

Every chat response, file edit, and commit is now sanitized automatically — zero commands per task.


📁 Distribution Artifacts in This Repo

Path What it is
install.sh One-command user installer (standalone binary → ~/.local/bin)
install-system.sh System-wide installer (sudo → /usr/local/bin, all users)
install.html Web installer landing page with platform auto-detection
package.json + install.js + cli-wrapper.js npm package (downloads platform binary on install)
Formula/aphanis.rb Homebrew formula
Makefile make build (PyInstaller binary), make publish (GitHub Release), make test
browser-extension/ Chrome/Edge/Firefox extension (popup cleaner + floating shield button)
web/index.html Zero-install browser-based cleaner (100% client-side)
desktop/ Electron desktop app (clipboard hotkey Cmd/Ctrl+Shift+U)
dist/aphanis Prebuilt macOS arm64 standalone binary (23 MB)

🙏 Upstream Inspiration & Acknowledgments

Aphanis was inspired by early zero-width character stripping concepts from guillaumemeyer/watermarks-remover. Aphanis expands upon these ideas into a full enterprise-grade Zero-Trust Provenance Firewall, Natural Humanizer Engine, 4-Vector Stego Risk Matrix, Cryptographic Certificate Generator, and Autonomous Agent Autopilot platform.


📄 License & Community

Distributed under the MIT License. See LICENSE for details.

  • Contributing: Please review CONTRIBUTING.md before submitting pull requests.
  • Security Policy: See SECURITY.md for vulnerability disclosure.

Metadata

Release files for aphanis 1.4.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for aphanis 1.4.0
File Size Uploaded
aphanis-1.4.0.tar.gz 57.7 kB Details

Built distribution (wheel)

Table of built distributions (wheels) for aphanis 1.4.0
File Interpreter ABI Platform
aphanis-1.4.0-py3-none-any.whl Python 3 none any Details

Total release size: 109.2 kB

Release files / aphanis-1.4.0.tar.gz

Download URL aphanis-1.4.0.tar.gz
Size 57.7 kB
Tags Source
SHA-256 checksum
How to use checksums
2dfdac30c7274fc30c12855269aa6e627d8a3d8513d23fbfb0ca8c3556352c42
BLAKE2b-256 checksum
How to use checksums
6f7a30ec54512cc384bda9547d986933a479c87acf293978272160aa97b0c464
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/7.0.0 CPython/3.14.3

Release files / aphanis-1.4.0-py3-none-any.whl

Download URL aphanis-1.4.0-py3-none-any.whl
Size 51.5 kB
Tags Python 3
SHA-256 checksum
How to use checksums
800fa0999ebd06c468d8bc58619375bed5317b02782e4aabc43f04bf6f00269d
BLAKE2b-256 checksum
How to use checksums
722897188c4f17c84cc2f86292ce1eb0a406ce85ee74f14d09b4204f2648cd28
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/7.0.0 CPython/3.14.3

Release history Release notifications | RSS feed

1.4.2

2 release files

This release

1.4.0 This release

2 release files

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page