AI providers

AI providers are the model backends agent work runs against: the credentials that power model routing, voice, and image/video for the workspace. Provider surfaces live at /ai/providers and /ai/agents/providers, with Models, Providers, and Gateways tabs. One credential per capability: model routing, voice, and image/video. Secret values are stored in GCP Secret Manager, never in this wiki. Cards read Ready (Manage) or Not connected (Connect); per-provider behavior beyond that is not documented in source yet.

Provider wiring is platform work, not tenant config. Tenants feel providers through session behavior and availability, not through keys they hold.
No password, API key, token, connection string, or private key goes in this wiki. Names of variables where source names them, values never.
Session trouble that looks like a provider issue (slow, down, or empty externals) is handled the same way: output stays sane, the session degrades instead of wedging, and the pattern gets reported with examples.
If provider-shaped trouble repeats, treat it as design feedback: narrow the scope, check the environment tool list, escalate with session references.

AI Providers: model routing, voice, and image/video credentials with ready and not-connected states

Grounding on the gateway-key step: Provisioning and internals. Model selection per workload: Model registry.