- create an isolated Relay project,
- issue and safely store a Relay key,
- verify a provider credential,
- send one OpenAI-compatible routed request, and
- confirm the routing decision in Traffic.
1
Create a project
Open Relay → Projects, select New project, and give it a name that identifies the workload or environment.Project creation is idempotent in the guided quickstart: returning to setup reuses the project instead of creating duplicates.
2
Create a Relay key
Create a key for the application. Relay shows the full
ox_... value once; store it in your secret manager before closing the dialog.3
Connect a provider
Add the provider credential Relay should use to execute requests. Verify it before enabling automatic routing.Observe-only projects do not need a provider credential unless Relay will also execute shadow comparisons.
4
Point your client at Relay
Set the OpenAI-compatible base URL and use the Relay key as the API key:Send a request directly to the endpoint:An existing OpenAI-compatible client can use the same base URL and Relay
key after the direct request succeeds.
5
Confirm the request
Open Relay → Traffic. Confirm the request source, status, requested and resolved models, cost, latency, and routing reason.
Project operations
Project settings act as defaults. Individual Relay keys can carry narrower request-rate, monthly-budget, expiry, or routing overrides. Archiving removes a project from normal active use without erasing its historical records. Unarchive it when the same boundary should become active again.Choose another integration
Observe existing calls or request decisions without proxying provider traffic.
Configure models and routing
Move from the first request to a production routing policy.