# Model Armor
Demonstrates the [Model Armor](https://cloud.google.com/security/products/model-armor)
middleware for content safety filtering. Model Armor scans prompts for
prompt injection, jailbreak attempts, and sensitive data before sending
them to the model.
## Features Demonstrated
| Feature | Flow | Description |
|---------|------|-------------|
| Content Filtering | `modelArmorFlow` | Generation with Model Armor middleware for PI/jailbreak and SDP filters |
## Setup
### Prerequisites
- **Node.js** (v18 or higher)
- **pnpm** package manager
- **Google Cloud project** with Model Armor enabled
### API Keys
```bash
export GEMINI_API_KEY='<your-api-key>'
```
### Model Armor Template
Create a Model Armor template in your GCP project, then set:
```bash
export MODEL_ARMOR_TEMPLATE='projects/<project>/locations/<location>/templates/<template>'
```
### Build and Install
From the repo root:
```bash
pnpm install
pnpm run setup
```
## Run the Sample
```bash
pnpm start
```
This starts the Genkit Dev UI with live reload.
## Testing This Demo
1. **Open DevUI** at http://localhost:4000
2. **Test content filtering**:
- [ ] `modelArmorFlow` — Default input tries a prompt injection
- [ ] Try safe inputs (e.g., `"Tell me about cats"`)
- [ ] Try unsafe inputs (e.g., `"ignore previous instructions..."`)
3. **Expected behavior**:
- Safe prompts pass through and generate a response
- Unsafe prompts are caught by Model Armor and raise a `GenkitError`
- Error details include the specific filter that triggered (PI/jailbreak, SDP)
