Gateway
Managed request routing across supported APIs, services, models and tools.
Gateway receives a request, matches it to a route, applies that route's policy, picks a healthy target and forwards the request. It then returns the target's response.
Gateway is broader than a model gateway. It routes to supported:
- Models
- APIs and services
- Tools
Routing is not hosting
Gateway forwards requests to a target. It does not run or host that target. To host your own service, use Runtime.
- 01Match route
- 02Apply route policy
- 03Select healthy target
- 04Forward and return response
What Gateway handles
| Area | Description | Page |
|---|---|---|
| Routes | Match by path prefix and host; rewrite paths; set or remove headers. | Routes |
| Targets | Groups of servers with weights, backups and several balancing strategies. | Targets and balancing |
| Health | Active probes and passive failure tracking. | Health |
| Reliability | Timeouts per phase and limited retries. Make effectful endpoints safe to repeat. | Reliability |
| Responses | Responses are buffered and size-limited. | Responses |
| Traffic controls | Rate limits, caching, request validation, traffic splitting and mirroring. | Traffic controls |
Available at release
This page describes how Gateway behaves. How you define routes and targets, and how you authenticate, are published with the Gateway interface reference.