infergo supports API key authentication via Bearer tokens.

# Set API key on server
infergo serve --model models/llama3-8b-q4.gguf --api-key YOUR_SECRET_KEY

# Or via environment variable
export INFERGO_API_KEY=YOUR_SECRET_KEY
infergo serve --model models/llama3-8b-q4.gguf

curl http://localhost:9090/v1/chat/completions \
  -H "Authorization: Bearer YOUR_SECRET_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"llm","messages":[{"role":"user","content":"Hello"}]}'

apiVersion: v1
kind: Secret
metadata:
  name: infergo-api-key
type: Opaque
stringData:
  api-key: "your-production-key-here"
---
apiVersion: apps/v1
kind: Deployment
spec:
  template:
    spec:
      containers:
        - name: infergo
          env:
            - name: INFERGO_API_KEY
              valueFrom:
                secretKeyRef:
                  name: infergo-api-key
                  key: api-key

Prevent abuse with per-key and global rate limits.

infergo serve --model models/llama3-8b-q4.gguf \
  --api-key YOUR_KEY \
  --rate-limit 100        # 100 requests/minute global
  --rate-limit-per-key 20 # 20 requests/minute per API key

Rate-limited responses return HTTP 429 with Retry-After header.

Three built-in roles control endpoint access:

# In server config
infergo serve --model llm:model.gguf \
  --api-key admin:ADMIN_KEY \
  --api-key user:USER_KEY \
  --api-key readonly:READONLY_KEY

Unauthenticated requests (no API key set) default to user role.

Restrict access to specific IP addresses or CIDR ranges.

infergo serve --model models/llama3-8b-q4.gguf \
  --allow-ip 10.0.0.0/8 \
  --allow-ip 192.168.1.0/24 \
  --allow-ip 203.0.113.42

When enabled, requests from non-allowed IPs receive HTTP 403.

infergo can detect and redact personally identifiable information in inputs and outputs:

• Email addresses
• Phone numbers
• Social Security Numbers
• Credit card numbers
• IP addresses

Enable via server config or per-request with "redact_pii": true.

Block harmful content categories:

• Violence
• Hate speech
• Self-harm
• Explicit content

All requests are logged in JSONL format with cryptographic hash chains for tamper evidence.

infergo serve --model models/llama3-8b-q4.gguf \
  --audit-log /var/log/infergo/audit.jsonl

Each log entry contains:

• Timestamp, request ID, client IP
• Model used, endpoint, response status
• SHA-256 hash linking to previous entry (tamper-evident chain)

Hash-only mode (no request/response bodies) available for sensitive environments:

--audit-log-mode hash-only

Delete all stored data for a specific user:

curl -X DELETE http://localhost:9090/v1/admin/gdpr/user123 \
  -H "Authorization: Bearer ADMIN_KEY"

This purges:

• Conversation memory
• Vector DB entries with matching user metadata
• Feedback records
• Audit log entries (marked as deleted)

Encrypt model files at rest with AES-256-GCM:

# Encrypt
infergo convert --input model.gguf --encrypt --key-file model.key

# Serve encrypted model
infergo serve --model models/model.gguf.enc --key-file model.key

Use a reverse proxy (nginx, Caddy, Traefik) for TLS termination:

server {
    listen 443 ssl;
    ssl_certificate /etc/ssl/cert.pem;
    ssl_certificate_key /etc/ssl/key.pem;

    location / {
        proxy_pass http://127.0.0.1:9090;
        proxy_set_header X-Real-IP $remote_addr;
    }
}

For same-machine clients, use UDS to avoid network exposure entirely:

infergo serve --model models/llama3-8b-q4.gguf --uds /tmp/infergo.sock

For maximum security and performance on same machine:

infergo serve --model models/llama3-8b-q4.gguf --shm infergo_shm

No network interface exposed. Communication via kernel-managed shared memory.

Webhooks are signed with HMAC-SHA256:

infergo serve --model models/llama3-8b-q4.gguf \
  --webhook-url https://your-service.com/hook \
  --webhook-secret YOUR_WEBHOOK_SECRET

Verify webhook signatures in your handler:

import hmac, hashlib

def verify(body, signature, secret):
    expected = hmac.new(secret.encode(), body, hashlib.sha256).hexdigest()
    return hmac.compare_digest(signature, expected)