Connections

Owners/admins open Connections to add customer-owned inference endpoints.

1. Add a named connection and choose the provider type.
2. Supply credentials and an endpoint where needed.
3. Classify its execution environment based on the actual deployment.
4. Review enabled state, provider options, routing policy and budgets.
5. Save changes, then run the small request from Overview.

The connection list shows enabled state and environment. Changing selection with unsaved edits prompts you to save or discard. Deletion requires confirmation and may affect keys/workloads using that connection.

The header’s history link opens configuration changes for connections. API request success is verified through Activity; a configured connection alone does not prove reachability.

A customer-owned Vercel AI Gateway is a BYOK connection. Steinkauz does not supply Gateway credits or resell inference.

OpenAI-compatible local services

Use an OpenAI Compatible connection for services such as TEI. Enter the provider endpoint and actual model ID; a key is optional when the service permits keyless access. You can paste an origin or its /v1 API root. Private network access depends on where Steinkauz runs and the deployment’s egress rules.

Model listing is optional. If /v1/models is unavailable, enter the raw model ID in the consuming application. Capability hints are advisory and do not gate use. No custom /info or /health endpoint is required.

Standard text embeddings use /v1/embeddings; prepare model-specific prefixes and extracted text in the consuming application. Standard multipart speech uses /v1/audio/transcriptions on OpenAI Compatible/OpenAI connections where the upstream supports it. Choose the connection and model in the calling application, and configure D/E permissions in Policies.