LLM Providers

Provider overview

Amplify's provider system is designed for maximum flexibility — you can swap between providers freely, add custom OpenAI-compatible endpoints, and use both cloud and local models simultaneously.

Providers are classified into three categories:

All 22 registered providers

BaseProvider architecture

Every provider extends the BaseProvider abstract class which provides:

Required abstract fields:

• name: string — Unique provider identifier (e.g., "OpenAI", "Anthropic")
• staticModels: ModelInfo[] — Hardcoded fallback model list
• config: ProviderConfig — { baseUrlKey?, baseUrl?, apiTokenKey? }

Required abstract method:

• getModelInstance(options) — Returns an AI SDK LanguageModel instance

Optional fields:

• getDynamicModels() — Fetches live model list from provider API at runtime
• getApiKeyLink — URL for obtaining an API key
• labelForGetApiKey — Label text for the "Get API Key" button
• icon — Provider icon SVG

Helper methods from BaseProvider:

• getProviderBaseUrlAndKey() — Resolves API key via the priority chain
• resolveDockerUrl() — Rewrites localhost → host.docker.internal for Docker
• createTimeoutSignal() — 5s timeout for model fetch API calls
• getModelsFromCache() / storeDynamicModels() — Caching keyed by apiKeys+settings+env

• API key resolution priority chain

  When a provider needs an API key, it resolves through this priority chain:

  1. providerSettings.baseUrl — From UI settings (cookie providers)
  2. serverEnv[baseUrlKey] — Cloudflare environment bindings
  3. process.env[baseUrlKey] — Node.js environment variables (.env.local)
  4. LLMManager.env[baseUrlKey] — Vite-parsed env vars
  5. config.baseUrl — Hardcoded default per provider

  This means you can configure providers at any level — env vars for server-side defaults, cookies for per-browser customization.

Custom providers

You can add custom OpenAI-compatible endpoints via the "Add Provider" popup in the model selector.

CustomProvider interface:

interface CustomProvider {
  id: string;         // UUID
  name: string;       // User-chosen display name
  baseUrl: string;    // OpenAI-compatible endpoint URL
  apiKey: string;     // Bearer token
  createdAt: string;  // ISO timestamp
}

Custom providers are stored in localStorage key amplify_custom_providers with cross-tab sync via the storage event listener. Unlike built-in providers, custom provider API keys are NOT stored in cookies.

The AddProviderPopup has two modes:

• "Pick" mode : Select an existing built-in provider to configure
• "Custom" mode : Add any OpenAI-compatible endpoint with a test-connection feature

Model Selector UI

The ModelSelector component (app/components/chat/ModelSelector.tsx) provides:

• Fuzzy search : Levenshtein distance matching (threshold 0.6 similarity)
• Grouped display : Cloud providers first, then local providers (Ollama, LMStudio, OpenAILike)
• Custom providers : Dedicated "Custom" group section
• Free model filter : Toggle to show only free-tier models
• Local status detection : Shows connectivity status for local providers
• Inline API key popup : Add/manage keys without leaving the selector
• Context window display : Shows sizes like "128K", "1.0M"
• Keyboard navigation : Arrow keys, enter, escape support

Dynamic model fetching

Providers with getDynamicModels() fetch live model lists from provider-specific API endpoints:

Dynamic models are cached keyed by JSON(apiKeys+settings+relevantEnv). When API keys change, the cache invalidates and models are re-fetched.