@librechat/agents 2.4.78 → 2.4.80

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@librechat/agents",
-  "version": "2.4.78",
+  "version": "2.4.80",
   "main": "./dist/cjs/main.cjs",
   "module": "./dist/esm/main.mjs",
   "types": "./dist/types/index.d.ts",
@@ -51,7 +51,7 @@
     "caching": "node -r dotenv/config --loader ./tsconfig-paths-bootstrap.mjs --experimental-specifier-resolution=node ./src/scripts/caching.ts --name 'Jo' --location 'New York, NY'",
     "thinking": "node -r dotenv/config --loader ./tsconfig-paths-bootstrap.mjs --experimental-specifier-resolution=node ./src/scripts/thinking.ts --name 'Jo' --location 'New York, NY'",
     "memory": "node -r dotenv/config --loader ./tsconfig-paths-bootstrap.mjs --experimental-specifier-resolution=node ./src/scripts/memory.ts --provider 'openAI' --name 'Jo' --location 'New York, NY'",
-    "tool-test": "node -r dotenv/config --loader ./tsconfig-paths-bootstrap.mjs --experimental-specifier-resolution=node ./src/scripts/tools.ts --provider 'bedrock' --name 'Jo' --location 'New York, NY'",
+    "tool-test": "node -r dotenv/config --loader ./tsconfig-paths-bootstrap.mjs --experimental-specifier-resolution=node ./src/scripts/tools.ts --provider 'openrouter' --name 'Jo' --location 'New York, NY'",
     "search": "node -r dotenv/config --loader ./tsconfig-paths-bootstrap.mjs --experimental-specifier-resolution=node ./src/scripts/search.ts --provider 'anthropic' --name 'Jo' --location 'New York, NY'",
     "ant_web_search": "node -r dotenv/config --loader ./tsconfig-paths-bootstrap.mjs --experimental-specifier-resolution=node ./src/scripts/ant_web_search.ts --name 'Jo' --location 'New York, NY'",
     "abort": "node -r dotenv/config --loader ./tsconfig-paths-bootstrap.mjs --experimental-specifier-resolution=node ./src/scripts/abort.ts --provider 'openAI' --name 'Jo' --location 'New York, NY'",

package/src/llm/anthropic/index.ts

@@ -141,6 +141,10 @@ export class CustomAnthropic extends ChatAnthropicMessages {
     this._lc_stream_delay = fields?._lc_stream_delay ?? 25;
   }
 
+  static lc_name(): 'LibreChatAnthropic' {
+    return 'LibreChatAnthropic';
+  }
+
   /**
    * Get the parameters used to invoke the model
    */

package/src/llm/google/index.ts

@@ -107,6 +107,10 @@ export class CustomChatGoogleGenerativeAI extends ChatGoogleGenerativeAI {
     this.streamUsage = fields.streamUsage ?? this.streamUsage;
   }
 
+  static lc_name(): 'LibreChatGoogleGenerativeAI' {
+    return 'LibreChatGoogleGenerativeAI';
+  }
+
   invocationParams(
     options?: this['ParsedCallOptions']
   ): Omit<GenerateContentRequest, 'contents'> {

package/src/llm/ollama/index.ts

@@ -13,6 +13,9 @@ import {
 } from './utils';
 
 export class ChatOllama extends BaseChatOllama {
+  static lc_name(): 'LibreChatOllama' {
+    return 'LibreChatOllama';
+  }
   async *_streamResponseChunks(
     messages: BaseMessage[],
     options: this['ParsedCallOptions'],

package/src/llm/openai/index.ts

@@ -198,6 +198,9 @@ export class ChatOpenAI extends OriginalChatOpenAI<t.ChatOpenAICallOptions> {
   public get exposedClient(): CustomOpenAIClient {
     return this.client;
   }
+  static lc_name(): string {
+    return 'LibreChatOpenAI';
+  }
   protected _getClientOptions(
     options?: OpenAICoreRequestOptions
   ): OpenAICoreRequestOptions {
@@ -233,7 +236,8 @@ export class ChatOpenAI extends OriginalChatOpenAI<t.ChatOpenAICallOptions> {
   getReasoningParams(
     options?: this['ParsedCallOptions']
   ): OpenAIClient.Reasoning | undefined {
-    if (!isReasoningModel(this.model)) {
+    const lc_name = (this.constructor as typeof ChatOpenAI).lc_name();
+    if (lc_name === 'LibreChatOpenAI' && !isReasoningModel(this.model)) {
       return;
     }
 
@@ -345,6 +349,10 @@ export class ChatOpenAI extends OriginalChatOpenAI<t.ChatOpenAICallOptions> {
       } else if ('reasoning' in delta) {
         chunk.additional_kwargs.reasoning_content = delta.reasoning;
       }
+      if ('provider_specific_fields' in delta) {
+        chunk.additional_kwargs.provider_specific_fields =
+          delta.provider_specific_fields;
+      }
       defaultRole = delta.role ?? defaultRole;
       const newTokenIndices = {
         prompt: options.promptIndex ?? 0,
@@ -435,6 +443,9 @@ export class AzureChatOpenAI extends OriginalAzureChatOpenAI {
   public get exposedClient(): CustomOpenAIClient {
     return this.client;
   }
+  static lc_name(): 'LibreChatAzureOpenAI' {
+    return 'LibreChatAzureOpenAI';
+  }
   /**
    * Returns backwards compatible reasoning parameters from constructor params and call options
    * @internal
@@ -576,6 +587,9 @@ export class ChatDeepSeek extends OriginalChatDeepSeek {
   public get exposedClient(): CustomOpenAIClient {
     return this.client;
   }
+  static lc_name(): 'LibreChatDeepSeek' {
+    return 'LibreChatDeepSeek';
+  }
   protected _getClientOptions(
     options?: OpenAICoreRequestOptions
   ): OpenAICoreRequestOptions {
@@ -644,6 +658,10 @@ export class ChatXAI extends OriginalChatXAI {
     }
   }
 
+  static lc_name(): 'LibreChatXAI' {
+    return 'LibreChatXAI';
+  }
+
   public get exposedClient(): CustomOpenAIClient {
     return this.client;
   }

package/src/llm/openai/utils/index.ts

@@ -298,10 +298,16 @@ export function _convertMessagesToOpenAIParams(
       role = 'developer';
     }
 
+    let hasAnthropicThinkingBlock: boolean = false;
+
     const content =
       typeof message.content === 'string'
         ? message.content
         : message.content.map((m) => {
+          if ('type' in m && m.type === 'thinking') {
+            hasAnthropicThinkingBlock = true;
+            return m;
+          }
           if (isDataContentBlock(m)) {
             return convertToProviderContentBlock(
               m,
@@ -326,7 +332,7 @@ export function _convertMessagesToOpenAIParams(
       completionParam.tool_calls = message.tool_calls.map(
         convertLangChainToolCallToOpenAI
       );
-      completionParam.content = '';
+      completionParam.content = hasAnthropicThinkingBlock ? content : '';
     } else {
       if (message.additional_kwargs.tool_calls != null) {
         completionParam.tool_calls = message.additional_kwargs.tool_calls;

package/src/llm/openrouter/index.ts

@@ -22,6 +22,9 @@ export class ChatOpenRouter extends ChatOpenAI {
       },
     });
   }
+  static lc_name(): 'LibreChatOpenRouter' {
+    return 'LibreChatOpenRouter';
+  }
   protected override _convertOpenAIDeltaToBaseMessageChunk(
     // eslint-disable-next-line @typescript-eslint/no-explicit-any
     delta: Record<string, any>,

package/src/llm/vertexai/index.ts

@@ -313,8 +313,8 @@ export class ChatVertexAI extends ChatGoogle {
   lc_namespace = ['langchain', 'chat_models', 'vertexai'];
   dynamicThinkingBudget = false;
 
-  static lc_name(): 'ChatVertexAI' {
-    return 'ChatVertexAI';
+  static lc_name(): 'LibreChatVertexAI' {
+    return 'LibreChatVertexAI';
   }
 
   constructor(fields?: VertexAIClientOptions) {

package/src/messages/prune.ts

@@ -389,6 +389,14 @@ export function checkValidNumber(value: unknown): value is number {
   return typeof value === 'number' && !isNaN(value) && value > 0;
 }
 
+type ThinkingBlocks = {
+  thinking_blocks?: Array<{
+    type: 'thinking';
+    thinking: string;
+    signature: string;
+  }>;
+};
+
 export function createPruneMessages(factoryParams: PruneMessagesFactoryParams) {
   const indexTokenCountMap = { ...factoryParams.indexTokenCountMap };
   let lastTurnStartIndex = factoryParams.startIndex;
@@ -402,6 +410,49 @@ export function createPruneMessages(factoryParams: PruneMessagesFactoryParams) {
     context: BaseMessage[];
     indexTokenCountMap: Record<string, number | undefined>;
   } {
+    if (
+      factoryParams.provider === Providers.OPENAI &&
+      factoryParams.thinkingEnabled === true
+    ) {
+      for (let i = lastTurnStartIndex; i < params.messages.length; i++) {
+        const m = params.messages[i];
+        if (
+          m.getType() === 'ai' &&
+          typeof m.additional_kwargs.reasoning_content === 'string' &&
+          Array.isArray(
+            (
+              m.additional_kwargs.provider_specific_fields as
+                | ThinkingBlocks
+                | undefined
+            )?.thinking_blocks
+          ) &&
+          (m as AIMessage).tool_calls &&
+          ((m as AIMessage).tool_calls?.length ?? 0) > 0
+        ) {
+          const message = m as AIMessage;
+          const thinkingBlocks = (
+            message.additional_kwargs.provider_specific_fields as ThinkingBlocks
+          ).thinking_blocks;
+          const signature =
+            thinkingBlocks?.[thinkingBlocks.length - 1].signature;
+          const thinkingBlock: ThinkingContentText = {
+            signature,
+            type: ContentTypes.THINKING,
+            thinking: message.additional_kwargs.reasoning_content as string,
+          };
+
+          params.messages[i] = new AIMessage({
+            ...message,
+            content: [thinkingBlock],
+            additional_kwargs: {
+              ...message.additional_kwargs,
+              reasoning_content: undefined,
+            },
+          });
+        }
+      }
+    }
+
     let currentUsage: UsageMetadata | undefined;
     if (
       params.usageMetadata &&

package/src/scripts/tools.ts

@@ -127,7 +127,10 @@ async function testStandardStreaming(): Promise<void> {
   const inputs = {
     messages: conversationHistory,
   };
-  const finalContentParts = await run.processStream(inputs, config);
+  const finalContentParts = await run.processStream(inputs, config, {
+    indexTokenCountMap: { 0: 35 },
+    maxContextTokens: 89000,
+  });
   const finalMessages = run.getRunMessages();
   if (finalMessages) {
     conversationHistory.push(...finalMessages);

package/src/utils/llmConfig.ts

@@ -12,6 +12,30 @@ export const llmConfigs: Record<string, t.LLMConfig | undefined> = {
     streamUsage: true,
     // disableStreaming: true,
   },
+  anthropicLITELLM: {
+    provider: Providers.OPENAI,
+    streaming: true,
+    streamUsage: false,
+    apiKey: 'sk-1234',
+    model: 'claude-sonnet-4',
+    maxTokens: 8192,
+    modelKwargs: {
+      metadata: {
+        user_id: 'some_user_id',
+      },
+      thinking: {
+        type: 'enabled',
+        budget_tokens: 2000,
+      },
+    },
+    configuration: {
+      baseURL: 'http://host.docker.internal:4000/v1',
+      defaultHeaders: {
+        'anthropic-beta': 'prompt-caching-2024-07-31,context-1m-2025-08-07',
+      },
+    },
+    // disableStreaming: true,
+  },
   [Providers.XAI]: {
     provider: Providers.XAI,
     model: 'grok-2-latest',