Introduction

LLM Secrets protects your .env files from AI coding assistants like Claude Code. Your secrets are encrypted with Windows Hello and only decrypted at runtime in isolated subprocesses.

Installation

Prerequisites

macOS:

• macOS with Touch ID (Apple Silicon M1/M2/M3 recommended)

Windows:

• Windows 10 or Windows 11
• Windows Hello enabled (PIN, fingerprint, or face recognition)

WSL/Linux:

• Windows 10/11 with WSL2 installed
• Ubuntu or Debian recommended

Download

Download LLM Secrets from the downloads page:

• macOS: LLM Secrets 3.1.0 for macOS (Apple Silicon)
• Windows: LLM Secrets 3.1.1 Setup (Windows)
• WSL/Linux: Scrt WSL 0.3.1

Quick Start

1. Download & Install

Download LLM Secrets for your platform from the downloads page and run the installer.

2. Encrypt Your Existing .env File

Navigate to your project and run setup:

cd your-project
scrt setup

Or specify the path to your .env:

scrt setup C:\path\to\your\.env

3. View Your Secrets

# View all secrets (authenticates automatically)
scrt view

# View a specific secret
scrt view STRIPE_SECRET_KEY

4. Run Commands with Secrets

scrt run -- npm run deploy

Encrypting & Managing Secrets

Encrypt Existing Secrets (First Time)

# In your project directory with a .env file:
scrt setup

# Or specify path:
scrt setup /path/to/.env

This creates:

View Secrets

# Authenticates with Windows Hello and shows all secrets
scrt view

# Show only a specific secret
scrt view API_KEY

Add or Edit Secrets

# Step 1: Decrypt to .env for editing
scrt decrypt

# Step 2: Edit .env in your text editor (opens automatically)

# Step 3: Re-encrypt when done
scrt encrypt --force

Add to .gitignore

.env
.env.backup

Only commit .env.encrypted to your repository.

How It Works

The injection mechanism is simple:

┌─────────────────┐     ┌─────────────────┐     ┌─────────────────┐
│  Your Command   │ --> │  Scrt injects   │ --> │  Command runs   │
│  (no secrets)   │     │  $env:VARS      │     │  with secrets   │
└─────────────────┘     └─────────────────┘     └─────────────────┘

1. Encryption at rest - Your .env is encrypted with AES-256-CBC
2. Windows Hello unlock - Master key protected by DPAPI + Windows Hello
3. Runtime injection - Secrets decrypted only in subprocess environment
4. Isolation - Values never return to parent process or Claude's context

Security Model

Simple vs Advanced Mode

Using with Claude Code

Claude Code automatically has access to use your secrets via the $env:SECRET_NAME syntax.

Example: Deploy a Smart Contract

# Claude writes this command:
forge script script/Deploy.s.sol \
  --rpc-url $env:ALCHEMY_RPC_URL \
  --private-key $env:PRIVATE_KEY \
  --broadcast

# Scrt injects the actual values at runtime
# Claude never sees your private key

Example: Push to GitHub

git push https://$env:GITHUB_PAT@github.com/user/repo.git

Example: Deploy to Vercel

vercel --token $env:VERCEL_TOKEN --prod

Commands Reference

Environment Variables

Add secrets to your .env file using standard format:

PRIVATE_KEY=0x...
ALCHEMY_RPC_URL=https://eth-mainnet.g.alchemy.com/v2/...
GITHUB_PAT=ghp_...
VERCEL_TOKEN=...
GODADDY_API_KEY=...
GODADDY_API_SECRET=...

Reference them in commands as $env:SECRET_NAME.

Backup & Restore

LLM Secrets supports encrypted cloud backups:

• Automatic backups - Configurable frequency (daily, weekly, monthly)
• Recovery password - Separate from Windows Hello, for disaster recovery
• Cross-machine restore - Restore secrets on a new Windows machine

Licensing

LLM Secrets is free for Windows and WSL. macOS requires a one-time $10 license.

Why Free for Windows?

We believe security tools should be accessible. Windows and WSL users get full functionality at no cost. The macOS license helps support cross-platform development.

Source Available

The source code is available for security auditing:

• You may: View code, audit for security, build for personal use, report vulnerabilities
• You may not: Redistribute, sell, or provide as hosted service

Security FAQ

The following answers are from DeepWiki, an independent third-party AI analysis of this codebase. Use their chatbot to ask your own security questions.

Can LLM Secrets, the developers, or Claude ever see my decrypted secrets?

No. Here's why:

• Client-side only - All encryption/decryption happens locally on your machine. There is no server component.
• Isolated subprocesses - Secrets are decrypted only in isolated subprocesses. Claude sees $env:SECRET_NAME, never the actual value.
• DPAPI protection - Master keys are encrypted with Windows DPAPI, bound to your Windows user account and machine.
• In-memory only - Commands like scrt view decrypt to memory, never writing plaintext to disk.
• Fully auditable - The crypto core is Apache 2.0 open source. The desktop app is source available. Verify yourself.

Even if the developers wanted to see your secrets (which they can't), the architecture makes it impossible since all operations are local and keys never leave your machine.

What happens if I lose my Windows account or master key?

You can recover IF you set up backups beforehand.

Recovery options:

1. Master Key Backup - During setup, a 44-character key is shown once. Save it in your password manager.
2. Recovery Password - Set a recovery password that encrypts your master key for cloud backup.
3. Disaster Recovery Scripts - Decrypt-MasterKeyBackup.ps1 and Decrypt-EnvFile.ps1 work without the app installed.

Troubleshooting

"Windows Hello authentication required"

A Windows Hello prompt will appear when accessing secrets. This is expected behavior.

".env.encrypted not found"

Run setup in your project first:

scrt setup

"Secret not found"

Check available secrets:

scrt list

Then add the missing secret by decrypting, editing, and re-encrypting:

scrt decrypt
# Edit .env file
scrt encrypt --force

Need to update a secret?

# Decrypt, edit, re-encrypt
scrt decrypt
# Edit .env in your editor
scrt encrypt --force

Install

One command to get started. Run this inside your WSL2 terminal:

The installer checks and installs everything you need. If packages are missing, it offers to install them automatically.

Setup

The setup wizard walks you through everything interactively:

Usage

Once your secrets are added, inject them into any command:

Claude Code writes commands using $SECRET_NAME syntax. The actual values are injected at runtime and never appear in the terminal output or Claude's context.

Commands