Secure Gateway Documentation

Roboflow Secure Gateway is a hardened proxy that sits between your inference servers and Roboflow Cloud. It caches model weights, enforces allow/deny rules, and streams structured audit logs to your SIEM — so your inference servers never need direct internet access.

The gateway runs as a single Docker container configured entirely through environment variables. It is a drop-in replacement for existing Roboflow License Server deployments.

Quick Start

Pull and run the container:

docker run -d \
  --name roboflow-secure-gateway \
  -p 443:443 \
  -e ROBOFLOW_API_KEY=your_api_key \
  -e CACHE_DIR=/var/cache/gateway \
  -v gateway-cache:/var/cache/gateway \
  roboflow/secure-gateway:latest

Point your inference servers at the gateway instead of https://api.roboflow.com:

export ROBOFLOW_API_URL=https://gateway.internal:443

Configuration

All configuration is done through environment variables.

Model Weight Caching

The gateway caches model weights on first download. Subsequent requests from any inference server in your fleet are served from the local cache, eliminating redundant transfers.

Cache eviction uses LRU (least recently used). Set CACHE_MAX_SIZE_GB to control the maximum disk usage. Mount a persistent volume to CACHE_DIR so the cache survives container restarts.

docker run -d \
  -e CACHE_DIR=/var/cache/gateway \
  -e CACHE_MAX_SIZE_GB=100 \
  -v /mnt/fast-storage:/var/cache/gateway \
  roboflow/secure-gateway:latest

Allow / Deny Rules

Rules control which models, versions, and API endpoints your inference servers can access through the gateway. Rules are defined in a YAML file.

Example: allow specific models only

mode: deny-all
allow:
  - model: "my-workspace/my-model"
    versions: ["3", "4"]
  - model: "my-workspace/safety-model"
    versions: ["*"]

Example: block specific endpoints

mode: allow-all
deny:
  - endpoint: "/dataset/*"
  - endpoint: "/workspace/keys"

Mount the rules file and set RULES_FILE to point to it. The gateway watches the file for changes and reloads rules without downtime.

SIEM Integration

The gateway streams structured JSON audit logs to your SIEM collector. Every proxied request generates a log entry.

Supported targets

• Splunk HEC (HTTPS)
• Datadog Logs API (HTTPS)
• Microsoft Sentinel (HTTPS)
• Any syslog collector (TCP or UDP)

Log schema

{
  "timestamp": "2026-04-14T12:00:00Z",
  "source_ip": "10.0.1.42",
  "method": "GET",
  "path": "/my-workspace/my-model/3",
  "action": "allowed",
  "rule_matched": "allow:my-workspace/my-model:3",
  "cache_hit": true,
  "response_code": 200,
  "bytes_transferred": 104857600,
  "duration_ms": 42
}

Example: Splunk HEC

docker run -d \
  -e SIEM_ENDPOINT=https://splunk.example.com:8088/services/collector \
  -e SIEM_PROTOCOL=https \
  -e SIEM_AUTH_TOKEN=your-hec-token \
  roboflow/secure-gateway:latest

TLS Configuration

By default the gateway generates a self-signed certificate on first start. For production, provide your own certificate:

docker run -d \
  -e TLS_CERT=/certs/gateway.crt \
  -e TLS_KEY=/certs/gateway.key \
  -v /path/to/certs:/certs:ro \
  roboflow/secure-gateway:latest

The gateway supports TLS 1.2 and 1.3. You can provide a custom CA certificate so that upstream traffic inspection appliances (DPI/IDS) can decrypt and inspect gateway traffic before it leaves your network.

Upgrading from License Server

Secure Gateway is a drop-in replacement for the Roboflow License Server. The upgrade path requires no changes to your inference server configuration.

1. Pull the new image: docker pull roboflow/secure-gateway:latest
2. Stop the existing License Server container.
3. Start Secure Gateway with the same ROBOFLOW_API_KEY and port mapping.
4. Add optional configuration for caching, rules, and SIEM as needed.

Your inference servers will continue to connect on the same address and port with no reconfiguration required.

Troubleshooting