Skip to main content

Interface: ModelConfig

Defined in: packages/core/src/providers/types.ts:245

Properties​

cacheControl?​

optional cacheControl?: CacheControlOption

Defined in: packages/core/src/providers/types.ts:301

Construction-time prompt-cache plan. true is the chat shorthand (last-tool + system + last four, then capped). An object is an explicit plan — { scope: 'prefix' } stamps last-tool + system. false turns it off. Per-call cache / cacheControl / enableCaching override this.

config?​

optional config?: ProviderConfig

Defined in: packages/core/src/providers/types.ts:260

Provider configuration (API key, custom fetch, etc.)

id​

id: string

Defined in: packages/core/src/providers/types.ts:247

Model identifier (e.g., 'gpt-5.6-luna', 'claude-3-5-sonnet-20241022')

provider?​

optional provider?: keyof ModelProviderRegistry

Defined in: packages/core/src/providers/types.ts:257

Provider name (auto-detected from id if not specified)

Auto-detection rules:

  • 'gpt-*' → 'openai'
  • 'claude-*' → 'anthropic'
  • Other models require explicit provider specification

reasoning?​

optional reasoning?: object

Defined in: packages/core/src/providers/types.ts:276

Reasoning configuration (for reasoning-capable models: gpt-5, o1, o3, o4)

budgetTokens?​

optional budgetTokens?: number

Token budget for extended thinking (Anthropic models)

effort?​

optional effort?: string & object | "low" | "medium" | "high" | "xhigh" | "max"

Reasoning effort level - higher effort = more thorough reasoning.

The named levels are the common ones and are what autocomplete offers, but the set is the provider's, not ours: Anthropic advertises xhigh and max for claude-opus-5, and publishes the list per model under capabilities.effort — which listModels reports as ModelInfo.effort. A closed union here would reject levels the provider documents, so the type suggests without forbidding.

summary?​

optional summary?: "concise" | "detailed" | "auto"

Reasoning summary style - concise, detailed, or auto

Example​

model: {
  id: 'gpt-5',
  reasoning: {
    effort: 'high',
    summary: true
  }
}