Security

l3mcore includes several layers of protection designed for near-production deployments.

Rate Limiting

A sliding window rate limiter blocks 60 requests per minute per IP address.

• Returns HTTP 429 when exceeded
• Respects the X-Forwarded-For header (safe behind reverse proxies)
• Configurable in api_server.py

# api_server.py — adjust limit
RATE_LIMIT = 60  # requests per minute

Payload Limits

Request bodies are limited to 1 MB to prevent memory exhaustion DoS attacks.

SSRF Protection

The Expert Dispatcher validates Ollama backend URLs before making requests:

URL	Status
http://127.0.0.1:11434	Allowed
http://192.168.x.x	Allowed (private LAN)
http://10.x.x.x	Allowed (internal network)
http://169.254.169.254	Blocked (AWS/GCP metadata)
http://metadata.google.internal	Blocked

Label Validation

Expert labels in experts.json are validated before being used in system paths:

• Alphanumeric, hyphens, and underscores only
• Prevents path traversal attacks (../, /etc/passwd, etc.)

Allowed: "programmer", "coder-v2", "data_analyst"
Blocked: "../../../etc/passwd", "model/evil", "label with spaces"

Log Sanitization

User prompts are cleaned before being written to logs:

• Control characters are removed
• ANSI escape sequences are removed
• Prevents log injection attacks

Secure Deployment

For production, we recommend:

# Example Nginx reverse proxy
server {
    listen 443 ssl;
    server_name api.yourdomain.com;

    location / {
        proxy_pass http://127.0.0.1:11435;
        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
        proxy_set_header X-Real-IP $remote_addr;

        # Additional authentication (optional)
        # auth_basic "l3mcore API";
        # auth_basic_user_file /etc/nginx/.htpasswd;
    }
}