Cloud Connectors and Model Providers
Runtime manages cloud Connectors, provider catalogs, and low-level protocol execution drivers. Users configure API keys and endpoints in Nimi Home under Cloud Services, then route application capabilities to those connectors inside their AIConfigs.
Architecture and Layering
| Layer | Responsibility | Security & Scope |
|---|---|---|
| Runtime Connector | Third-party API keys, custom endpoints, retry and timeout policies | Credentials stored securely in protected local storage; hidden from renderers |
| AIConfig Cloud Route | References to active connectors (connectorRef) and targets (providerModelTarget) | Persisted as application configuration using safe symbolic references |
| Application Layer | High-level prompts, parameters, and business logic | Fully decoupled from wire protocols, HTTP headers, and credential handling |
Centralizing sensitive credentials within Runtime prevents API key leaks and eliminates complex networking plumbing from client applications.
Driver Adaptation and Deterministic Scheduling
- Protocol Translation: Runtime's
CapabilityImplementationDrivertranslates uniform platform operations into provider-specific REST, WebSocket, or gRPC requests. - Deterministic Routing: Once submitted, Runtime routes calls strictly to the designated target. If a remote provider experiences an outage or rate limit, Runtime returns the exact upstream failure rather than falling back to an unverified model that could cause hallucinations or unexpected billing.
- Normalized Outputs: Applications consume standardized response streams and structured errors; proprietary headers and vendor-specific envelopes are stripped cleanly.
Troubleshooting Cloud Failures
If a cloud inference call fails, follow this diagnostic workflow:
- Verify Route Configuration: Ensure the active AIConfig references a valid, enabled Connector.
- Test Connector Health: Use the "Test Connection" button in Cloud Services to verify API key validity and network connectivity.
- Check Provider Quotas: Verify that your provider account (e.g., OpenAI, Anthropic) has sufficient credit balance and has not hit rate limits.
- Update Targets: If switching models, update the capability target in your application's AI configuration and save changes.