Merge branch 'main' into feat/krutrim-integration

zhao-lun · web-flow · commit d7fd38e18901 · 2025-07-04T11:21:12.000+08:00
diff --git a/package-lock.json b/package-lock.json
diff --git a/package.json b/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@portkey-ai/gateway",
-  "version": "1.10.0",
+  "version": "1.10.1",
   "description": "A fast AI gateway by Portkey",
   "repository": {
     "type": "git",
diff --git a/src/providers/anthropic/chatComplete.ts b/src/providers/anthropic/chatComplete.ts
@@ -362,11 +362,15 @@ export const AnthropicChatCompleteConfig: ProviderConfig = {
                 cache_control: { type: 'ephemeral' },
               }),
             });
-          } else if (tool.computer) {
+          } else if (tool.type) {
+            const toolOptions = tool[tool.type];
             tools.push({
-              ...tool.computer,
-              name: 'computer',
-              type: tool.computer.name,
+              ...(toolOptions && { ...toolOptions }),
+              name: tool.type,
+              type: toolOptions?.name,
+              ...(tool.cache_control && {
+                cache_control: { type: 'ephemeral' },
+              }),
             });
           }
         });
diff --git a/src/providers/azure-openai/chatComplete.ts b/src/providers/azure-openai/chatComplete.ts
@@ -111,6 +111,9 @@ export const AzureOpenAIChatCompleteConfig: ProviderConfig = {
   stream_options: {
     param: 'stream_options',
   },
+  web_search_options: {
+    param: 'web_search_options',
+  },
 };
 
 interface AzureOpenAIChatCompleteResponse extends ChatCompletionResponse {}
diff --git a/src/providers/bedrock/chatComplete.ts b/src/providers/bedrock/chatComplete.ts
@@ -315,13 +315,15 @@ export const BedrockConverseChatCompleteConfig: ProviderConfig = {
         | { cachePoint: { type: string } }
       > = [];
       params.tools?.forEach((tool) => {
-        tools.push({
-          toolSpec: {
-            name: tool.function.name,
-            description: tool.function.description,
-            inputSchema: { json: tool.function.parameters },
-          },
-        });
+        if (tool.function) {
+          tools.push({
+            toolSpec: {
+              name: tool.function.name,
+              description: tool.function.description,
+              inputSchema: { json: tool.function.parameters },
+            },
+          });
+        }
         if (tool.cache_control && !canBeAmazonModel) {
           tools.push({
             cachePoint: {
@@ -353,7 +355,8 @@ export const BedrockConverseChatCompleteConfig: ProviderConfig = {
           }
         }
       }
-      return { ...toolConfig, toolChoice };
+      // TODO: split this into two provider options, one for tools and one for toolChoice
+      return tools.length ? { ...toolConfig, toolChoice } : null;
     },
   },
   guardrailConfig: {
diff --git a/src/providers/bedrock/embed.ts b/src/providers/bedrock/embed.ts
@@ -1,35 +1,129 @@
 import { BEDROCK } from '../../globals';
-import { EmbedResponse } from '../../types/embedRequestBody';
+import { EmbedParams, EmbedResponse } from '../../types/embedRequestBody';
 import { Params } from '../../types/requestBody';
 import { ErrorResponse, ProviderConfig } from '../types';
 import { generateInvalidProviderResponseError } from '../utils';
 import { BedrockErrorResponseTransform } from './chatComplete';
 
 export const BedrockCohereEmbedConfig: ProviderConfig = {
-  input: {
-    param: 'texts',
-    required: true,
-    transform: (params: any): string[] => {
-      if (Array.isArray(params.input)) {
-        return params.input;
-      } else {
-        return [params.input];
-      }
+  input: [
+    {
+      param: 'texts',
+      required: false,
+      transform: (params: EmbedParams): string[] | undefined => {
+        if (typeof params.input === 'string') return [params.input];
+        else if (Array.isArray(params.input) && params.input.length > 0) {
+          const texts: string[] = [];
+          params.input.forEach((item) => {
+            if (typeof item === 'string') {
+              texts.push(item);
+            } else if (item.text) {
+              texts.push(item.text);
+            }
+          });
+          return texts.length > 0 ? texts : undefined;
+        }
+      },
     },
-  },
+    {
+      param: 'images',
+      required: false,
+      transform: (params: EmbedParams): string[] | undefined => {
+        if (Array.isArray(params.input) && params.input.length > 0) {
+          const images: string[] = [];
+          params.input.forEach((item) => {
+            if (typeof item === 'object' && item.image?.base64) {
+              images.push(item.image.base64);
+            }
+          });
+          return images.length > 0 ? images : undefined;
+        }
+      },
+    },
+  ],
   input_type: {
     param: 'input_type',
     required: true,
   },
   truncate: {
     param: 'truncate',
+    required: false,
+  },
+  encoding_format: {
+    param: 'embedding_types',
+    required: false,
+    transform: (params: any): string[] => {
+      if (Array.isArray(params.encoding_format)) return params.encoding_format;
+      return [params.encoding_format];
+    },
   },
 };
 
 export const BedrockTitanEmbedConfig: ProviderConfig = {
-  input: {
-    param: 'inputText',
-    required: true,
+  input: [
+    {
+      param: 'inputText',
+      required: false,
+      transform: (params: EmbedParams): string | undefined => {
+        if (
+          Array.isArray(params.input) &&
+          typeof params.input[0] === 'object' &&
+          params.input[0].text
+        ) {
+          return params.input[0].text;
+        }
+        if (typeof params.input === 'string') return params.input;
+      },
+    },
+    {
+      param: 'inputImage',
+      required: false,
+      transform: (params: EmbedParams) => {
+        // Titan models only support one image per request
+        if (
+          Array.isArray(params.input) &&
+          typeof params.input[0] === 'object' &&
+          params.input[0].image?.base64
+        ) {
+          return params.input[0].image.base64;
+        }
+      },
+    },
+  ],
+  dimensions: [
+    {
+      param: 'dimensions',
+      required: false,
+      transform: (params: EmbedParams): number | undefined => {
+        if (typeof params.input === 'string') return params.dimensions;
+      },
+    },
+    {
+      param: 'embeddingConfig',
+      required: false,
+      transform: (
+        params: EmbedParams
+      ): { outputEmbeddingLength: number } | undefined => {
+        if (Array.isArray(params.input) && params.dimensions) {
+          return {
+            outputEmbeddingLength: params.dimensions,
+          };
+        }
+      },
+    },
+  ],
+  encoding_format: {
+    param: 'embeddingTypes',
+    required: false,
+    transform: (params: any): string[] => {
+      if (Array.isArray(params.encoding_format)) return params.encoding_format;
+      return [params.encoding_format];
+    },
+  },
+  // Titan specific parameters
+  normalize: {
+    param: 'normalize',
+    required: false,
   },
 };
 
diff --git a/src/providers/bedrock/utils.ts b/src/providers/bedrock/utils.ts
@@ -8,7 +8,7 @@ import {
   BedrockConverseAnthropicChatCompletionsParams,
   BedrockConverseCohereChatCompletionsParams,
 } from './chatComplete';
-import { Options } from '../../types/requestBody';
+import { Options, Tool } from '../../types/requestBody';
 import { GatewayError } from '../../errors/GatewayError';
 import { BedrockFinetuneRecord, BedrockInferenceProfile } from './types';
 import { FinetuneRequest } from '../types';
@@ -137,6 +137,25 @@ export const transformAnthropicAdditionalModelRequestFields = (
       additionalModelRequestFields['anthropic_beta'] = params['anthropic_beta'];
     }
   }
+  if (params.tools && params.tools.length) {
+    const anthropicTools: any[] = [];
+    params.tools.forEach((tool: Tool) => {
+      if (tool.type !== 'function') {
+        const toolOptions = tool[tool.type];
+        anthropicTools.push({
+          ...(toolOptions && { ...toolOptions }),
+          name: tool.type,
+          type: toolOptions?.name,
+          ...(tool.cache_control && {
+            cache_control: { type: 'ephemeral' },
+          }),
+        });
+      }
+    });
+    if (anthropicTools.length) {
+      additionalModelRequestFields['tools'] = anthropicTools;
+    }
+  }
   return additionalModelRequestFields;
 };
 
diff --git a/src/providers/cohere/embed.ts b/src/providers/cohere/embed.ts
@@ -4,31 +4,60 @@ import { generateErrorResponse } from '../utils';
 import { COHERE } from '../../globals';
 
 export const CohereEmbedConfig: ProviderConfig = {
-  input: {
-    param: 'texts',
-    required: true,
-    transform: (params: EmbedParams): string[] => {
-      if (Array.isArray(params.input)) {
-        return params.input as string[];
-      } else {
-        return [params.input];
-      }
+  input: [
+    {
+      param: 'texts',
+      required: false,
+      transform: (params: EmbedParams): string[] | undefined => {
+        if (typeof params.input === 'string') return [params.input];
+        else if (Array.isArray(params.input) && params.input.length > 0) {
+          const texts: string[] = [];
+          params.input.forEach((item) => {
+            if (typeof item === 'string') {
+              texts.push(item);
+            } else if (item.text) {
+              texts.push(item.text);
+            }
+          });
+          return texts.length > 0 ? texts : undefined;
+        }
+      },
     },
-  },
-  model: {
-    param: 'model',
-    default: 'embed-english-light-v2.0',
-  },
+    {
+      param: 'images',
+      required: false,
+      transform: (params: EmbedParams): string[] | undefined => {
+        if (Array.isArray(params.input) && params.input.length > 0) {
+          const images: string[] = [];
+          params.input.forEach((item) => {
+            if (typeof item === 'object' && item.image?.base64) {
+              images.push(item.image.base64);
+            }
+          });
+          return images.length > 0 ? images : undefined;
+        }
+      },
+    },
+  ],
   input_type: {
     param: 'input_type',
+    required: true,
+  },
+  truncate: {
+    param: 'truncate',
     required: false,
   },
-  embedding_types: {
+  encoding_format: {
     param: 'embedding_types',
     required: false,
+    transform: (params: any): string[] => {
+      if (Array.isArray(params.encoding_format)) return params.encoding_format;
+      return [params.encoding_format];
+    },
   },
-  truncate: {
-    param: 'truncate',
+  //backwards compatibility
+  embedding_types: {
+    param: 'embedding_types',
     required: false,
   },
 };
diff --git a/src/providers/google-vertex-ai/chatComplete.ts b/src/providers/google-vertex-ai/chatComplete.ts
@@ -832,6 +832,10 @@ export const VertexAnthropicChatCompleteStreamChunkTransform: (
 
   if (parsedChunk.type === 'message_start' && parsedChunk.message?.usage) {
     streamState.model = parsedChunk?.message?.model ?? '';
+
+    streamState.usage = {
+      prompt_tokens: parsedChunk.message.usage?.input_tokens,
+    };
     return (
       `data: ${JSON.stringify({
         id: fallbackId,
@@ -850,7 +854,7 @@ export const VertexAnthropicChatCompleteStreamChunkTransform: (
           },
         ],
         usage: {
-          prompt_tokens: parsedChunk.message?.usage?.input_tokens,
+          prompt_tokens: streamState.usage.prompt_tokens,
         },
       })}` + '\n\n'
     );
@@ -873,6 +877,10 @@ export const VertexAnthropicChatCompleteStreamChunkTransform: (
         ],
         usage: {
           completion_tokens: parsedChunk.usage?.output_tokens,
+          prompt_tokens: streamState.usage?.prompt_tokens,
+          total_tokens:
+            (streamState.usage?.prompt_tokens || 0) +
+            (parsedChunk.usage?.output_tokens || 0),
         },
       })}` + '\n\n'
     );
diff --git a/src/providers/open-ai-base/index.ts b/src/providers/open-ai-base/index.ts
@@ -132,6 +132,9 @@ export const chatCompleteParams = (
     stream_options: {
       param: 'stream_options',
     },
+    web_search_options: {
+      param: 'web_search_options',
+    },
   };
 
   // Exclude params that are not needed.
diff --git a/src/providers/openai/chatComplete.ts b/src/providers/openai/chatComplete.ts
@@ -118,6 +118,9 @@ export const OpenAIChatCompleteConfig: ProviderConfig = {
   reasoning_effort: {
     param: 'reasoning_effort',
   },
+  web_search_options: {
+    param: 'web_search_options',
+  },
 };
 
 export interface OpenAIChatCompleteResponse extends ChatCompletionResponse {
diff --git a/src/types/requestBody.ts b/src/types/requestBody.ts

Original file line number	Diff line number	Diff line change
`@@ -1,6 +1,6 @@`
`1`	`1`	`{`
`2`	`2`	`"name": "@portkey-ai/gateway",`
`3`		`- "version": "1.10.0",`
	`3`	`+ "version": "1.10.1",`
`4`	`4`	`"description": "A fast AI gateway by Portkey",`
`5`	`5`	`"repository": {`
`6`	`6`	`"type": "git",`