Provider Registry

The Simple RPC AI Backend integrates with @anolilab/ai-model-registry to deliver curated provider and model metadata with resilient fallbacks.

Architecture

  1. External Registry Package – ships curated provider/model metadata via @anolilab/ai-model-registry.
  2. ProviderRegistryService – filters providers, applies overrides, and exposes metadata via tRPC.
  3. Fallback Data – ensures availability when the external registry package cannot be resolved.

Quick Start 

pnpm install
# Registry data is available immediately with sensible defaults

Programmatic Usage 

import { ProviderRegistryService } from 'simple-rpc-ai-backend';

const registry = new ProviderRegistryService(
  ['anthropic', 'openai'],
  ['anthropic', 'openai', 'custom-ai'],
  ['free']
);

tRPC Methods

  • ai.listProviders – service-managed providers with metadata and pricing.
  • ai.listProvidersBYOK – BYOK providers that expect user-supplied keys.
const { providers, source } = await client.ai.listProviders.query();

Configuration 

export AI_SERVICE_PROVIDERS=anthropic,openai,google
export AI_BYOK_PROVIDERS=anthropic,openai,google,custom-ai

Pricing Overrides 

registry.addPricingOverride({
  provider: 'openai',
  model: 'gpt-4o',
  pricing: {
    input: 2.5,
    output: 10
  },
  reason: 'Enterprise contract pricing'
});

Apply provider-level overrides by omitting the model property.

Monitoring & Resilience

The service relies on @anolilab/ai-model-registry at runtime. When the library is unavailable the backend serves the bundled fallback metadata and getHealthStatus() returns status: 'unhealthy' with the error message from the import. Hook that method into your own health endpoint or scheduler if you need alerts when the live registry is unreachable. Because data is loaded directly from the package there is no explicit cache directory to manage; customize pricing or model entries through addPricingOverride / addModelOverride as shown above.