@juspay/neurolink 10.8.4 → 10.8.6

package/dist/server/routes/openaiProxyRoutes.js

@@ -321,7 +321,7 @@ export function createOpenAIProxyRoutes(modelRouter, basePath = "", loopbackPort
                         model: targetModel,
                     }, requestModelRouter?.getFallbackChain() ?? [], body.model, 
                     // The classifier only reads fields present on both types.
-                    adapted);
+                    adapted, requestModelRouter?.isAutoFallbackEnabled?.() ?? false);
                     const attempts = plan.attempts;
                     // --- Optional tracing ---
                     let tracer;

package/dist/types/errors.d.ts

@@ -12,6 +12,11 @@ export declare class ProviderError extends BaseError {
     provider?: string | undefined;
     constructor(message: string, provider?: string | undefined);
 }
+export type OllamaHttpError = ProviderError & {
+    statusCode?: number;
+    statusText?: string;
+    responseBody?: string;
+};
 /**
  * Thrown for authentication-related errors, such as invalid or missing API keys.
  */

package/dist/types/proxy.d.ts

@@ -29,6 +29,8 @@ export type ModelRouterInterface = {
     resolve(requestedModel: string): RouteResult;
     isClaudeTarget(requestedModel: string): boolean;
     getFallbackChain(): FallbackEntry[];
+    isAutoFallbackEnabled?(): boolean;
+    getMaxInflightPerAccount?(): number;
     getModelMappings?: () => ModelMapping[];
     getPassthroughModels?: () => string[];
 };
@@ -639,6 +641,27 @@ export type AnthropicSuccessResult = {
     };
 } | {
     response: Response | unknown;
+    holdsAccountAdmission?: boolean;
+};
+/** A release handle for one in-flight request admitted to an OAuth account. */
+export type AccountAdmissionLease = {
+    release(): void;
+};
+/** A cancellable queued request for per-account admission capacity. */
+export type QueuedAccountAdmission = {
+    accountKey: string;
+    promise: Promise<AccountAdmissionLease>;
+    cancel(): void;
+};
+/** One queued request waiting for per-account admission capacity. */
+export type AccountAdmissionWaiter = {
+    capacity: number;
+    resolve: (lease: AccountAdmissionLease) => void;
+};
+/** In-process request admission state for an OAuth account. */
+export type AccountAdmissionState = {
+    active: number;
+    waiters: AccountAdmissionWaiter[];
 };
 /** Result of buffering only enough upstream SSE to make a retry-safe decision. */
 export type AnthropicStreamPreflightResult = {
@@ -669,6 +692,7 @@ export type ParsedSSEBuffer = {
 };
 export type AnthropicAuthRetryResult = {
     response?: Response | unknown;
+    holdsAccountAdmission?: boolean;
     continueLoop: boolean;
     lastError: unknown;
     authFailureMessage: string | null;

package/dist/types/subscription.d.ts

@@ -908,6 +908,10 @@ export type ProxyRoutingConfig = {
     strategy: "round-robin" | "fill-first";
     modelMappings: ModelMapping[];
     fallbackChain: FallbackEntry[];
+    /** Permit a last-resort provider chosen by the translation layer. Disabled by default. */
+    autoFallback?: boolean;
+    /** Maximum in-flight upstream requests per OAuth account. Defaults to two. */
+    maxInflightPerAccount?: number;
     passthroughModels?: string[];
     /** Enable quota-aware fill-first account ordering. Defaults to true. */
     quotaRouting?: boolean;

package/dist/utils/providerHealth.js

@@ -7,6 +7,7 @@ import { AIProviderName, OpenAIModels, GoogleAIModels, AnthropicModels, BedrockM
 import { API_KEY_LENGTHS, PROJECT_ID_FORMAT } from "./providerConfig.js";
 import { basename } from "path";
 import { createProxyFetch } from "../proxy/proxyFetch.js";
+import { DEFAULT_OLLAMA_MODEL } from "../providers/ollama/constants.js";
 export class ProviderHealthChecker {
     static healthCache = new Map();
     static DEFAULT_TIMEOUT = 5000; // 5 seconds
@@ -690,7 +691,7 @@ export class ProviderHealthChecker {
         return new URL("/api/tags", this.getOllamaBaseUrl()).toString();
     }
     static getConfiguredOllamaModel() {
-        return process.env.OLLAMA_MODEL || "llama3.1:8b";
+        return process.env.OLLAMA_MODEL || DEFAULT_OLLAMA_MODEL;
     }
     static async fetchJsonWithTimeout(url, options = {}) {
         const controller = new AbortController();

package/dist/utils/providerUtils.js

@@ -7,6 +7,7 @@ import { logger } from "./logger.js";
 import { AIProviderName } from "../constants/enums.js";
 import { ProviderHealthChecker } from "./providerHealth.js";
 import { API_KEY_FORMATS, API_KEY_LENGTHS, PROJECT_ID_FORMAT, } from "./providerConfig.js";
+import { DEFAULT_OLLAMA_MODEL } from "../providers/ollama/constants.js";
 /**
  * Get the best available provider based on real-time availability checks
  * Enhanced version consolidated from providerUtils-fixed.ts
@@ -107,7 +108,7 @@ async function isProviderAvailable(providerName) {
         return availability.available;
     }
     if (providerName === "ollama") {
-        const availability = await ProviderHealthChecker.checkFallbackProviderAvailability(AIProviderName.OLLAMA, process.env.OLLAMA_MODEL || "llama3.1:8b");
+        const availability = await ProviderHealthChecker.checkFallbackProviderAvailability(AIProviderName.OLLAMA, process.env.OLLAMA_MODEL || DEFAULT_OLLAMA_MODEL);
         return availability.available;
     }
     try {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@juspay/neurolink",
-  "version": "10.8.4",
+  "version": "10.8.6",
   "packageManager": "pnpm@10.15.1",
   "description": "TypeScript AI SDK with 24+ LLM providers behind one consistent API. MCP-native (connect any MCP server), voice TTS/STT/realtime, RAG, agents, memory, context compaction. OpenAI · Anthropic · Gemini · Bedrock · Azure · Ollama · DeepSeek · NVIDIA NIM and more.",
   "author": {
@@ -348,6 +348,8 @@
   },
   "dependencies": {
     "@ai-sdk/anthropic": "^3.0.50",
+    "@ai-sdk/google": "^4.0.29",
+    "@ai-sdk/google-vertex": "^5.0.36",
     "@ai-sdk/mistral": "^3.0.21",
     "@ai-sdk/openai": "^3.0.37",
     "@ai-sdk/provider": "^3.0.8",