Skip to main content
What you need to run a self-hosted Guardway gateway in development or production. The gateway ships as a Docker image; you don’t install any language runtimes on the host.
Using the SaaS Gateway? Nothing on this page applies. Guardway runs it, and the only network rule is outbound HTTPS to gateway.guardway.ai. See Network Requirements.

Hardware

Development

  • CPU: 2 cores
  • RAM: 4 GB
  • Disk: 2 GB free
  • Network: standard internet connection

Production (up to 100k requests/day)

  • CPU: 4 vCPU
  • RAM: 8 GB
  • Disk: 200 GB SSD — the local Postgres database (request history), container images, logs
  • Network: 1 Gbps interface recommended
The production card is the 100k requests/day tier for gateway v0.5.46 or later. For 1M and 10M requests/day, the per-request cost model, and the retention settings that keep the database bounded, see Sizing.

Software

1

Docker 24.0+ with Compose v2

2

Postgres 15 (bundled)

The gateway keeps request history, guardrail events and traces in Postgres. The onboarding bundle runs it as the database service on the same host; point the gateway at an external Postgres with the DB_* variables on Environment.
3

Redis 7 (bundled, off by default)

The bundle ships a cache service, but the gateway uses in-memory storage until REDIS_ENABLED=true. Turn it on for shared cache and rate-limit state when running more than one gateway replica.
You do not need Node.js, pnpm, or any language runtime on the host. The gateway container bundles everything.

API keys

At least one LLM provider API key is required for the gateway to serve traffic. You configure keys in the dashboard after the gateway is online — see Connect a provider. The full list of supported providers lives there; common ones include OpenAI, Anthropic, Google Gemini, Groq, and Mistral.

Network

The full list of hosts, ports and what each connection carries is on Network Requirements. In short:

Outbound (from the gateway host)

  • HTTPS (443) to api.us.guardway.ai and logs.us.guardway.ai: the Guardway control plane and event upload.
  • HTTPS (443) to registry.us.guardway.ai and prod-us-east-1-starport-layer-bucket.s3.us-east-1.amazonaws.com: image pulls on install and upgrade. Docker Compose installs also need 563763293556.dkr.ecr.us-east-1.amazonaws.com for the base images.
  • HTTPS (443) to id.guardway.ai: keys that verify dashboard sign-ins.
  • HTTPS (443) to each LLM provider you connect, and to any integration you turn on (SIEM, alerts, webhooks, MCP servers).
The gateway needs direct outbound HTTPS. Exempt the Guardway hosts from TLS inspection.

Inbound (to the gateway host)

  • Port 8080 (default): the gateway’s HTTP API, for your applications and SDKs. Allow it from your own networks only.
  • Nothing needs to reach the gateway from the internet or from Guardway. The dashboard at app.guardway.ai reaches the gateway through the control plane over the gateway’s own outbound connection.
  • The bundle also publishes Postgres on port 5432 on all interfaces. Block it from outside the host.

Security expectations

The container image is hardened (read-only root filesystem, non-root user, dropped capabilities, restricted seccomp profile). Your container runtime needs to support those features — any recent Docker, containerd, or Kubernetes release does.
Secrets you’ll need to store somewhere (env vars, Docker secrets, K8s secrets, Vault, Secrets Manager, Key Vault, etc.):
  • The one-time registration token (used once at boot — see Deployment)
  • LLM provider API keys
  • TLS certificates if you terminate TLS at the gateway

Platform-specific

  • Kubernetes 1.24+
  • Ingress controller (nginx, Traefik, or equivalent)
  • Persistent volume for /var/lib/guardway if you want cache + audit logs to survive restarts