Cloud administration
Platform operators, service configuration, and AI provider operations.
Cloud operators use the administration workspace to view platform users, projects, billing accounts, and audited configuration changes. Customer project owners manage their own projects in the regular console.
Service settings
Each service has its own settings under Cloud administration → Services. Operators can pause availability, record an operator note, and configure rate, burst, and daily-quota policy. Billing tariffs remain code-reviewed. AI traffic settings tighten existing project policies.
Settings propagate to gateway replicas through background refreshes. A service pause rejects new requests; already submitted AI operations continue through billing settlement.
AI operations
The AI service adds Providers, Models, Usage, Pricing, and Settings sections. Providers use built-in adapters; credentials are encrypted, write-only, and stored on the server. Public model IDs follow provider/model-slug, with a separate upstream model ID.
Set Canonical model to link routes that serve the same model, such as direct OpenAI and OpenRouter. The public route ID remains unique and is what clients send for generation. Price cards, credentials, capabilities, publication state, and usage remain independent for each route. Pausing a provider affects its routes; it does not pause other providers of the canonical model. Published status is catalog readiness, not measured upstream uptime.
Save models as drafts while reviewing pricing. Publication requires a valid price card and configured provider credentials. Price meters use exact decimal USD per individual unit. Review all input, output, caching, modality, and tool costs before publishing.
OpenAI and Google defaults
Configure your provider keys, then choose Install defaults in AI administration. The October 11, 2026 pack adds 18 missing reviewed models across GPT-6, GPT Image 2.5, Gemini 3.x, Nano Banana, and Gemini TTS. Existing models and pricing policies are preserved; connections without keys receive drafts. Publish those drafts after configuring credentials. Customer projects still need AI enabled and a model policy that permits the selected IDs.
Cards use Standard paid-tier pricing, include cache/reasoning and applicable long-context rates, and expire before January 2027. Native Google tasks price each reported modality separately. Use native tasks for Google multimodal input. Default OpenAI GPT cards accept text input, and OpenAI image cards expose generation. The installer lists remaining video/Interactions, Live/WebSocket, and additional usage-adapter limitations; those models are not published with guessed billing rules.
The default retail policy is 10% markup on provider cost. Gross margin is a separate explicit policy. Existing requests retain their original pricing and credential reference when configuration changes.
Usage reports show exact ledger costs and customer charges. Unsettled requests remain visible because their final costs may still change.
Operator access
Only server-allowlisted Cloud accounts can access administration. Signup, project ownership, and OAuth permissions cannot grant this role. Platform configuration writes are audited and use revision checks to prevent concurrent editors from silently overwriting each other.
Deployment settings and the migration procedure are documented in the repository's cloud_project/17-SUPERADMIN.md. Preserve the existing AI data encryption key when enabling administration.