Full reference
Deployment and routing choices
Choose where inference runs and which account supplies provider access. Treat model routing, compute ownership and application authentication as separate decisions.
Updated
Reviewed against Console v0.2.2. Available actions depend on project access and runtime configuration; check your installed CLI version for command differences.
Bring your own provider credential
Register a supported upstream credential with BYOK. Applications still call the Stimulir gateway with their own Stimulir key. Verify the model route and the upstream account’s limits.
Use managed inference
Use a managed route enabled for your workspace and plan. Inspect the current catalogue and usage evidence. Managed availability and credits are account configuration, not fixed guarantees in this reference.
Operate a configured runtime
A customer-operated runtime needs compatible hardware, model artifacts, service configuration and access controls. Direct runtime calls use that endpoint’s contract; they do not automatically inherit the hosted gateway’s authentication, metering or available models.
Choose from operational requirements
| Question | Check |
|---|---|
| Where will the request execute? | Hosted provider route or the exact configured runtime |
| Who controls credentials? | Workspace BYOK, managed admission or runtime operator |
| What is billed? | Current plan, route usage and any separately provisioned compute |
| What evidence is available? | Project traces, usage records and runtime job status |
| What changes production? | Routing configuration, labels, deployments and explicit promotion |
Use Inference for application requests and Compute for resource lifecycles.