GhostCode — Privacy Proxy for AI

Hide your business logic before sharing code with LLMs. One click to anonymize, one click to restore.

GhostCode is a local-first privacy proxy that replaces your private symbols (variable names, functions, strings, comments) with opaque tokens before you share code with AI. The AI sees gv_001 instead of customer_revenue. When you get your answer back, GhostCode restores everything.

Features

🔒 One-Click Hide

Open any Python or C/C++ file and click the eye-closed icon in the editor toolbar — GhostCode strips all business context and generates ghost code safe for any LLM.

👁️ One-Click Reveal

After the AI responds, click the eye icon to restore every symbol to its original name — including in the AI's explanation.

📊 Pre-Send Risk Report

Before you paste, GhostCode shows you exactly what's protected:

Symbol and literal counts by category
Structural patterns still visible to the AI
Estimated domain exposure: LOW / MEDIUM / HIGH

🎚️ 4 Privacy Levels

Level	What it does
1	Rename symbols + strip comments
2	+ Scrub string/numeric literals
3	+ Function isolation with stubs
4	+ Dimension generalization

Browse every symbol mapping grouped by kind (variables, functions, types). Hover over ghost tokens in the editor to see the original name.

🔐 Encrypted Maps

Optionally encrypt your mapping files so even your ghost maps are safe at rest.

Requirements

GhostCode requires Python 3.10+ with the ghostcode package installed:

pip install ghostcode

Set the Python path in settings if needed:

Settings → ghostcode.pythonPath

Usage

Open a .py, .cpp, .c, or .h file
Click the 👁️‍🗨️ (eye-closed) icon in the editor toolbar — or use Cmd+Shift+G H
Pick a privacy level (1–4)
Review the Risk Report — check your exposure level
Copy the ghost output and paste it into ChatGPT, Claude, Copilot, etc.
Paste the AI's response back and click the 👁️ (eye) icon to reveal

Keyboard Shortcuts

Shortcut	Command
`Cmd+Shift+G H`	Hide current file
`Cmd+Shift+G P`	Hide project (multi-file)

Extension Settings

Setting	Description	Default
`ghostcode.pythonPath`	Path to Python interpreter	`python3`

How It Works

Your Code          →  GhostCode Hide  →  Ghost Code       →  LLM
customer_revenue      gv_001             gv_001 = gv_002      "gv_001 stores..."
calculate_tax()       gf_001()           gf_001(gv_003)

LLM Response       →  GhostCode Reveal  →  Your Answer
"gv_001 stores..."    customer_revenue      "customer_revenue stores..."

Everything runs locally. No data leaves your machine. No telemetry. No cloud.

Repository Configuration

Security teams can add a .ghostcode.yaml to enforce policies:

min_scrub_level: 3
block_level_1: true
encrypt_maps: true
banned_patterns:
  - "*.key"
  - "*credentials*"

License

MIT

GhostCode

Smit Vaishanav