@juspay/neurolink 12.47.12 → 12.47.14

package/dist/constants/contextWindows.js

@@ -278,7 +278,14 @@ export const MODEL_CONTEXT_WINDOWS = {
         "anthropic.claude-opus-4-5-20251101-v1:0": 200_000,
         "anthropic.claude-sonnet-4-5-20250929-v1:0": 200_000,
         "anthropic.claude-haiku-4-5-20251001-v1:0": 200_000,
-        // Claude legacy
+        // Claude 4.1 / 4 — the AWS Bedrock model cards state 200K tokens
+        "anthropic.claude-opus-4-1-20250805-v1:0": 200_000,
+        "anthropic.claude-sonnet-4-20250514-v1:0": 200_000,
+        // Claude legacy. AWS lists no model card for 3.7 Sonnet; 200K is the window
+        // of the same model in the Anthropic manifest. The `us.` id is its
+        // cross-region inference profile, the form tokens.ts keys its output cap by.
+        "anthropic.claude-3-7-sonnet-20250219-v1:0": 200_000,
+        "us.anthropic.claude-3-7-sonnet-20250219-v1:0": 200_000,
         "anthropic.claude-3-5-sonnet-20241022-v1:0": 200_000,
         "anthropic.claude-3-5-haiku-20241022-v1:0": 200_000,
         "anthropic.claude-3-opus-20240229-v1:0": 200_000,

package/dist/providers/googleAiStudio/client.js

@@ -2393,7 +2393,7 @@ export class GoogleAIStudioProvider extends BaseProvider {
                 contents: [text],
             });
             const embedding = result.embeddings?.[0]?.values;
-            if (!embedding) {
+            if (!embedding || embedding.length === 0) {
                 throw new ProviderError("No embedding returned from Google AI", this.providerName);
             }
             logger.debug("Embedding generated successfully", {
@@ -2425,10 +2425,11 @@ export class GoogleAIStudioProvider extends BaseProvider {
             model: embeddingModelName,
             count: texts.length,
         });
+        let rows;
         try {
             const apiKey = this.getApiKey();
             const client = await createGoogleGenAIClient(apiKey, this.getBaseURL());
-            const embeddings = [];
+            rows = [];
             if (embeddingModelName.includes("gemini-embedding-2")) {
                 // @google/genai 2.x reads an array of strings for these models as ONE
                 // content with several parts, so a batch would come back as a single
@@ -2441,7 +2442,7 @@ export class GoogleAIStudioProvider extends BaseProvider {
                         contents: [text],
                     })));
                     for (const result of results) {
-                        embeddings.push(result.embeddings?.[0]?.values || []);
+                        rows.push(result.embeddings?.[0]?.values);
                     }
                 }
             }
@@ -2450,20 +2451,13 @@ export class GoogleAIStudioProvider extends BaseProvider {
                     model: embeddingModelName,
                     contents: texts,
                 });
-                embeddings.push(...(result.embeddings || []).map((e) => e.values || []));
+                rows.push(...(result.embeddings || []).map((e) => e.values));
             }
             // One vector per text, or fail: a silent mismatch would pair texts with
             // the wrong vectors downstream.
-            if (embeddings.length !== texts.length) {
-                throw new Error(`Embedding response held ${embeddings.length} vectors for ${texts.length} texts`);
+            if (rows.length !== texts.length) {
+                throw new Error(`Embedding response held ${rows.length} vectors for ${texts.length} texts`);
             }
-            logger.debug("Batch embeddings generated successfully", {
-                provider: this.providerName,
-                model: embeddingModelName,
-                count: embeddings.length,
-                embeddingDimension: embeddings[0]?.length,
-            });
-            return embeddings;
         }
         catch (error) {
             logger.error("Batch embedding generation failed", {
@@ -2473,6 +2467,26 @@ export class GoogleAIStudioProvider extends BaseProvider {
             });
             throw this.handleProviderError(error);
         }
+        // Checked after the try on purpose: a thrown error is re-worded by
+        // handleProviderError from keywords in its text, so an index such as 429
+        // or 502 would turn this into a rate-limit or server error.
+        const embeddings = rows.map((values, index) => {
+            if (!values || values.length === 0) {
+                logger.error("Batch embedding returned no values", {
+                    model: embeddingModelName,
+                    index,
+                });
+                throw new ProviderError(`Embedding for the text at index ${index} came back without values`, this.providerName);
+            }
+            return values;
+        });
+        logger.debug("Batch embeddings generated successfully", {
+            provider: this.providerName,
+            model: embeddingModelName,
+            count: embeddings.length,
+            embeddingDimension: embeddings[0]?.length,
+        });
+        return embeddings;
     }
     getApiKey() {
         const apiKey = this.credentials?.apiKey ||

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@juspay/neurolink",
-  "version": "12.47.12",
+  "version": "12.47.14",
   "packageManager": "pnpm@10.15.1",
   "description": "The pipe layer of an AI nervous system: one interface connecting provider neurons to your application, across three inference types — generate, stream and decide. `decide` returns typed, calibrated judgements from a non-generative model (~400ms, ~$0.00002/call) for routing, tool selection and context budgeting. MCP-native (4 transports), voice TTS/STT/realtime, RAG, agents, memory, compaction, 9 observability exporters. OpenAI · Anthropic · Gemini · Bedrock · Azure · Ollama · TypeSafe Jev and more.",
   "author": {