@opengeni/config 0.10.0 → 0.10.3

package/dist/index.d.ts

@@ -163,6 +163,7 @@ declare const SettingsSchema: z.ZodObject<{
     openaiBaseUrl: z.ZodOptional<z.ZodString>;
     openaiModel: z.ZodDefault<z.ZodString>;
     openaiAllowedModels: z.ZodDefault<z.ZodString>;
+    vercelAiGatewayApiKey: z.ZodOptional<z.ZodString>;
     voiceInputMaxDurationSeconds: z.ZodDefault<z.ZodCoercedNumber<unknown>>;
     voiceInputMaxSizeBytes: z.ZodDefault<z.ZodCoercedNumber<unknown>>;
     voiceInputProviderOrder: z.ZodDefault<z.ZodString>;
@@ -189,6 +190,7 @@ declare const SettingsSchema: z.ZodObject<{
     openaiReasoningEffort: z.ZodDefault<z.ZodEnum<{
         high: "high";
         low: "low";
+        max: "max";
         medium: "medium";
         minimal: "minimal";
         none: "none";
@@ -527,6 +529,7 @@ export declare const ModelCapabilitiesV1Schema: z.ZodObject<{
         efforts: z.ZodArray<z.ZodEnum<{
             high: "high";
             low: "low";
+            max: "max";
             medium: "medium";
             minimal: "minimal";
             none: "none";
@@ -535,6 +538,7 @@ export declare const ModelCapabilitiesV1Schema: z.ZodObject<{
         defaultEffort: z.ZodNullable<z.ZodEnum<{
             high: "high";
             low: "low";
+            max: "max";
             medium: "medium";
             minimal: "minimal";
             none: "none";
@@ -620,6 +624,19 @@ export declare const ModelCapabilitiesV1Schema: z.ZodObject<{
             runnable: z.ZodBoolean;
         }, z.core.$strip>;
     }, z.core.$strip>;
+    promptCaching: z.ZodOptional<z.ZodObject<{
+        upstream: z.ZodEnum<{
+            supported: "supported";
+            unknown: "unknown";
+            unsupported: "unsupported";
+        }>;
+        runnable: z.ZodBoolean;
+        mode: z.ZodEnum<{
+            automatic: "automatic";
+            implicit: "implicit";
+            none: "none";
+        }>;
+    }, z.core.$strip>>;
     latencyModes: z.ZodArray<z.ZodObject<{
         id: z.ZodEnum<{
             fast: "fast";
@@ -680,6 +697,8 @@ export type ModelProviderApi = z.infer<typeof ModelProviderApi>;
 export declare const RegistryProviderKind: z.ZodEnum<{
     "api-key": "api-key";
     "codex-subscription": "codex-subscription";
+    "vercel-gateway-managed": "vercel-gateway-managed";
+    "vercel-gateway-workspace": "vercel-gateway-workspace";
 }>;
 export type RegistryProviderKind = z.infer<typeof RegistryProviderKind>;
 /** A non-built-in provider declared by the host via OPENGENI_MODEL_PROVIDERS_JSON. */
@@ -687,6 +706,8 @@ declare const RegistryProviderSchema: z.ZodObject<{
     kind: z.ZodDefault<z.ZodEnum<{
         "api-key": "api-key";
         "codex-subscription": "codex-subscription";
+        "vercel-gateway-managed": "vercel-gateway-managed";
+        "vercel-gateway-workspace": "vercel-gateway-workspace";
     }>>;
     id: z.ZodString;
     label: z.ZodOptional<z.ZodString>;
@@ -725,6 +746,7 @@ declare const RegistryProviderSchema: z.ZodObject<{
                 efforts: z.ZodArray<z.ZodEnum<{
                     high: "high";
                     low: "low";
+                    max: "max";
                     medium: "medium";
                     minimal: "minimal";
                     none: "none";
@@ -733,6 +755,7 @@ declare const RegistryProviderSchema: z.ZodObject<{
                 defaultEffort: z.ZodNullable<z.ZodEnum<{
                     high: "high";
                     low: "low";
+                    max: "max";
                     medium: "medium";
                     minimal: "minimal";
                     none: "none";
@@ -818,6 +841,19 @@ declare const RegistryProviderSchema: z.ZodObject<{
                     runnable: z.ZodBoolean;
                 }, z.core.$strip>;
             }, z.core.$strip>;
+            promptCaching: z.ZodOptional<z.ZodObject<{
+                upstream: z.ZodEnum<{
+                    supported: "supported";
+                    unknown: "unknown";
+                    unsupported: "unsupported";
+                }>;
+                runnable: z.ZodBoolean;
+                mode: z.ZodEnum<{
+                    automatic: "automatic";
+                    implicit: "implicit";
+                    none: "none";
+                }>;
+            }, z.core.$strip>>;
             latencyModes: z.ZodArray<z.ZodObject<{
                 id: z.ZodEnum<{
                     fast: "fast";
@@ -907,6 +943,12 @@ export interface ConfiguredModel {
     credentialSource: CredentialSourceV1;
     billing: BillingAttributionV1;
     capabilities: ModelCapabilitiesV1;
+    requestPolicy?: {
+        gateway: {
+            only: [string, ...string[]];
+            caching: "auto" | "none";
+        };
+    };
     pricing?: ModelPricingScheduleV1 | undefined;
     definitionVersion: string;
     contextWindowTokens?: number | undefined;
@@ -916,6 +958,63 @@ export interface ConfiguredModel {
     reasoningEffort: boolean;
     hostedWebSearch: boolean;
 }
+export declare const VERCEL_AI_GATEWAY_BASE_URL: "https://ai-gateway.vercel.sh/v1";
+export declare const VERCEL_AI_GATEWAY_AI_SDK_BASE_URL: "https://ai-gateway.vercel.sh/v4/ai";
+export declare const OPENGENI_GATEWAY_PROVIDER_ID: "opengeni-gateway";
+export declare const WORKSPACE_GATEWAY_PROVIDER_ID: "workspace-gateway";
+export declare const WORKSPACE_GATEWAY_MODEL_ID_PREFIX: "workspace-gateway/";
+export declare const VERCEL_AI_GATEWAY_CONNECTION_DOMAIN: "ai-gateway.vercel.sh";
+export declare const VERCEL_AI_GATEWAY_CONNECTION_ROLE: "vercel_ai_gateway";
+export declare const CODEX_REALTIME_MODEL_ID: "gpt-live-1-boulder-alpha";
+export declare const OPENGENI_REALTIME_MODEL_ID_PREFIX: "opengeni-gateway/";
+export declare const WORKSPACE_REALTIME_MODEL_ID_PREFIX: "workspace-gateway/";
+/** Curated voice models exposed through AI Gateway's normalized realtime API. */
+export declare const AI_GATEWAY_REALTIME_MODELS: {
+    readonly openaiRealtime21: {
+        readonly upstreamModelId: "openai/gpt-realtime-2.1";
+        readonly managedModelId: "opengeni-gateway/openai/gpt-realtime-2.1";
+        readonly workspaceModelId: "workspace-gateway/openai/gpt-realtime-2.1";
+        readonly label: "GPT Realtime 2.1";
+        readonly description: "Best overall voice intelligence";
+    };
+    readonly openaiRealtimeMini: {
+        readonly upstreamModelId: "openai/gpt-realtime-mini";
+        readonly managedModelId: "opengeni-gateway/openai/gpt-realtime-mini";
+        readonly workspaceModelId: "workspace-gateway/openai/gpt-realtime-mini";
+        readonly label: "GPT Realtime Mini";
+        readonly description: "Faster, lighter live voice";
+    };
+    readonly grokVoiceThinkFast20: {
+        readonly upstreamModelId: "xai/grok-voice-think-fast-2.0";
+        readonly managedModelId: "opengeni-gateway/xai/grok-voice-think-fast-2.0";
+        readonly workspaceModelId: "workspace-gateway/xai/grok-voice-think-fast-2.0";
+        readonly label: "Grok Voice Think Fast 2.0";
+        readonly description: "Fast, natural xAI voice";
+    };
+};
+export type AiGatewayRealtimeModel = (typeof AI_GATEWAY_REALTIME_MODELS)[keyof typeof AI_GATEWAY_REALTIME_MODELS];
+export declare function resolveAiGatewayRealtimeModel(modelId: string): {
+    source: "managed" | "workspace";
+    upstreamModelId: string;
+} | null;
+export declare const OPENGENI_GATEWAY_MODELS: {
+    readonly deepseek: {
+        readonly productId: "deepseek-v4-flash-0731";
+        readonly workspaceProductId: "workspace-gateway/deepseek-v4-flash-0731";
+        readonly upstreamModelId: "deepseek/deepseek-v4-flash-0731";
+        readonly label: "DeepSeek V4 Flash 0731";
+        readonly providers: readonly ["baseten", "novita", "deepinfra"];
+        readonly implicitCaching: true;
+    };
+    readonly kimi: {
+        readonly productId: "kimi-k3";
+        readonly workspaceProductId: "workspace-gateway/kimi-k3";
+        readonly upstreamModelId: "moonshotai/kimi-k3";
+        readonly label: "Kimi K3";
+        readonly providers: readonly ["baseten", "fireworks"];
+        readonly implicitCaching: true;
+    };
+};
 /**
  * Built-in OpenGeni credit pricing schedules.
  *
@@ -977,6 +1076,11 @@ export declare function collectSandboxEnvironment(settings: Settings, source?: N
  * explicit environment in tests.
  */
 export declare function resolveProviderApiKey(provider: Pick<RegistryProvider, "apiKey" | "apiKeyEnv">, source?: NodeJS.ProcessEnv): string | undefined;
+export declare function gatewayRequestPolicyForUpstreamModel(upstreamModelId: string): ConfiguredModel["requestPolicy"];
+/** Static catalog overlay; it contains no concrete workspace credential. */
+export declare function withWorkspaceGatewayCatalogProvider(settings: Settings): Settings;
+/** Runtime overlay after the worker resolves the workspace's encrypted key. */
+export declare function withWorkspaceGatewayCredential(settings: Settings, apiKey: string): Settings;
 /**
  * Product display label for catalog/picker UI.
  * Same string for OpenAI and Codex copies of a slug (`gpt-5.6-luna` and
@@ -1118,6 +1222,14 @@ export declare function configuredEntitlements(settings: Settings): Entitlements
 export declare function calculateModelUsageCostMicros(settings: Settings, model: string, usage: ModelUsageInput, options?: {
     latencyMode?: LatencyMode;
 }): number;
+/**
+ * Convert AI Gateway's exact USD inference cost to OpenGeni credit micros and
+ * apply the configured model margin. Decimal arithmetic is integer-only so a
+ * sub-micro provider charge cannot be lost to floating-point rounding.
+ */
+export declare function calculateGatewayReportedCostMicros(settings: Settings, model: string, inferenceCostUsd: string, options?: {
+    inputTokens?: number;
+}): number;
 export declare function configuredAllowedReasoningEfforts(settings: Settings): Array<z.infer<typeof ReasoningEffort>>;
 /**
  * Decodes OPENGENI_ENVIRONMENTS_ENCRYPTION_KEY (base64, exactly 32 bytes) for