Prompt-Vault

Vault API Documentation

Server-side prompt proxy with AES-256-GCM encryption. Your prompts never leave the server. Clients only interact through the API.

Quick start

1. Install

cd prompt-vault
npm install

2. Configure

cp .env.example .env

# Generate a secure API token:
node -e "console.log(require('crypto').randomBytes(32).toString('hex'))"

# Edit .env:
VAULT_SECRET=your-strong-random-secret
ANTHROPIC_API_KEY=sk-ant-api03-...
API_TOKENS=your-generated-token

3. Create prompts

Add .md files to prompts/raw/ with frontmatter:

---
id: my-prompt
name: My Prompt
author: Your Name
copyright: (c) 2024 Your Name
version: 1.0.0
description: What this prompt does
---
Your prompt content here.
Use {{variable}} for dynamic values.

4. Encrypt and start

npm run encrypt   # Encrypts prompts into vault.enc
npm start         # Starts the API server on port 3700

API Endpoints

POST /api/run Auth required

Execute a prompt through the Anthropic API. The prompt content is never returned to the client. The AI response is watermarked with invisible zero-width Unicode characters.

Request

curl -X POST http://localhost:3700/api/run \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt_id": "seo-article",
    "variables": {
      "topic": "AI Security",
      "keyword": "prompt protection",
      "word_count": "800"
    }
  }'

Request body

Field	Type	Required	Description
`prompt_id`	string	yes	ID of the prompt to execute
`variables`	object	no	Key-value pairs to interpolate into the prompt template
`model`	string	no	Anthropic model override (default: claude-sonnet-4-20250514)

Response

{
  "response": "...(watermarked AI response)...",
  "prompt_id": "seo-article",
  "model": "claude-sonnet-4-20250514",
  "watermarked": true,
  "usage": { "input_tokens": 150, "output_tokens": 800 }
}

GET /api/prompts Auth required

List all available prompts. Returns metadata only -- prompt content is never exposed.

Request

curl http://localhost:3700/api/prompts \
  -H "Authorization: Bearer YOUR_TOKEN"

Response

{
  "prompts": [
    {
      "id": "seo-article",
      "name": "SEO Article Writer",
      "author": "Belkis Aslani",
      "copyright": "(c) 2024 Belkis Aslani",
      "version": "1.0.0",
      "description": "Generates SEO-optimized articles"
    }
  ]
}

GET /api/health Public

Health check endpoint. No authentication required.

{ "status": "ok", "prompts_loaded": 3, "uptime": 1234 }

Security details

Encryption

Prompts are encrypted with AES-256-GCM. Key derivation uses scrypt with a 32-byte random salt. Each encryption uses a fresh IV. The GCM authentication tag ensures both confidentiality and integrity.

Vault format: [salt:32][iv:16][tag:16][ciphertext]

Authentication

API tokens are hashed with SHA-256 before comparison. Raw tokens are never stored or logged. Request logs only show the first 8 characters of the hash for debugging.

Header: Authorization: Bearer <token>

Watermarking

Every AI response is embedded with invisible zero-width Unicode characters that encode the prompt ID, timestamp, and client token hash. The watermark survives copy-paste.

Characters: U+200B U+200C U+200D

Response headers

Every response includes legal copyright headers, content-type protection, frame denial, and no-cache directives. Rate limiting is set to 30 requests per minute per IP.

X-Prompt-License / X-Prompt-Copyright

Error codes

Status	Meaning	Resolution
`400`	Missing prompt_id	Include prompt_id in request body
`401`	Missing Authorization header	Add Bearer token to request header
`403`	Invalid API token	Check that your token matches one in API_TOKENS
`404`	Prompt not found	Verify prompt_id exists via GET /api/prompts
`429`	Rate limit exceeded	Wait and retry (30 req/min limit)
`502`	Anthropic API auth failed	Check ANTHROPIC_API_KEY in .env

← Back to Prompt-Armor