Cangyun AI Gateway
Enterprise AI traffic governance: unified multi-vendor model access with multi-tenant key auth, rate limiting, usage metering & billing, and smart routing.
Request Trial →Why Cangyun AI Gateway
Pain point: Teams connect to model vendors directly — keys sprawl, accountability blurs
One gateway entry with a multi-tenant key system: govern everything in one place
Pain point: AI spend is a black box; month-end bills cannot be attributed
Request-level metering pipeline with per-tenant reconciliation (0% variance in acceptance)
Pain point: One team's burst traffic exhausts the shared model quota
Per-tenant rate limits and quotas with fault isolation
Pain point: Switching models or multi-vendor failover requires code changes
Gateway-side model routing and protocol translation — zero business-code changes
Features
Multi-tenancy
Tenant / key / quota model with a branded management console.
Key Authentication
In-house wasm auth plugin; dynamic credential source takes effect in minutes without restarts.
Rate Limiting & Quotas
Per-tenant QPS limits and periodic quotas with standard error responses.
Metering & Billing
Request-level token metering pipeline with bill export and reconciliation.
Model Routing
Route by model name, tenant, or weight, with canary and fallback support.
Cloud-native Architecture
Built on Envoy/Higress with wasm plugins; Kubernetes-native deployment.
Deployment & Delivery
- ✓K8s / k3s cluster deployment with one-command Helm install
- ✓Single-node Docker Compose for PoC and small-scale use
- ✓China-network ready: registry mirrors and offline install packages
- ✓Passed real-cluster K8s E2E acceptance (multi-tenancy / rate limiting / billing)
FAQ
Which upstream model services are supported?
Any OpenAI-protocol-compatible service, including public model APIs and self-hosted inference.
How does it relate to open-source Higress?
Built on the Higress core, adding Cangyun's auth, metering, and routing wasm plugins plus a branded console, with commercial support and compliance statements.
How accurate is the billing data?
The metering pipeline passed reconciliation acceptance with 0% variance against upstream bills; per-tenant detail export is supported.