Opsta AI Gateway

One self-hosted gateway for all your LLM traffic

Route work-type to the right model, cap tokens and spend per team, enforce model allow-lists and guardrails, and see every token — built on Higress, fully Infrastructure-as-Code, on-premise, no cloud lock-in.

What is it? ->

Getting started

Work-type → model routing

Header/tag-based routing to logical model routes (coding-default, bulk) with fallback — native Gateway API config, no plugin.

Per-group/user limits

Redis-backed token rate limits and per-group model allow-lists, keyed by the {org, project, group, user} identity tuple.

Observability

Grafana LGTM (Loki/Mimir/Tempo) + Alloy — tokens, latency and rate-limits per model & route, per-user via logs.

In-cluster TLS + remote access

Let's Encrypt certificates terminated in-cluster (cert-manager, DNS-01); Cloudflare Tunnel for dev, direct in prod — same config.