Configuration

Manage the providers this factory can reach, the approved models available to workflows, and the protected credentials that connectors resolve at runtime.

Console appearanceone choice for every browser and every person
Card headers can carry a colour for their subject: spend green, approvals amber, live activity blue, controls red, knowledge violet, identity teal, traces indigo. Or the console can stay on its monochrome base.

Providers

…saved

Where model calls can go. The Foundry gateway comes from the environment; add a Vertex or Bedrock endpoint, an OpenAI-compatible service, or a vLLM, Ollama or LM Studio server on your own hardware. Discover what each serves and offer the models you want; the runtime routes every call to the provider its model names, with that provider’s credential from the vault. Local providers are priced by the hardware hour, not per token, and the FinOps page uses that rate.

Models

…saved

The deployments on the Foundry resource. Tick what this factory offers; every model dropdown in the builder shows exactly this list, and a spec names a deployment, never a provider. Record the price per million tokens beside each offered model: it is what cost figures and model comparisons are computed from.

Sort

Keys

referenced from specs as ${secret:name}
What specs will reference.
What this credential is for.
Any connector — including ones that do not exist yet — reads it as ${secret:name}