@librechat/agents 3.0.81 → 3.0.771

package/dist/types/llm/bedrock/index.d.ts

@@ -1,9 +1,5 @@
 /**
  * Optimized ChatBedrockConverse wrapper that fixes contentBlockIndex conflicts
- * and adds support for latest @langchain/aws features:
- *
- * - Application Inference Profiles (PR #9129)
- * - Service Tiers (Priority/Standard/Flex) (PR #9785) - requires AWS SDK 3.966.0+
  *
  * Bedrock sends the same contentBlockIndex for both text and tool_use content blocks,
  * causing LangChain's merge logic to fail with "field[contentBlockIndex] already exists"
@@ -15,94 +11,19 @@
  * the purpose of tracking tool call ordering.
  */
 import { ChatBedrockConverse } from '@langchain/aws';
-import { ChatGenerationChunk, ChatResult } from '@langchain/core/outputs';
-import type { CallbackManagerForLLMRun } from '@langchain/core/callbacks/manager';
 import type { ChatBedrockConverseInput } from '@langchain/aws';
 import type { BaseMessage } from '@langchain/core/messages';
-/**
- * Service tier type for Bedrock invocations.
- * Requires AWS SDK >= 3.966.0 to actually work.
- * @see https://docs.aws.amazon.com/bedrock/latest/userguide/service-tiers-inference.html
- */
-export type ServiceTierType = 'priority' | 'default' | 'flex' | 'reserved';
-/**
- * Extended input interface with additional features:
- * - applicationInferenceProfile: Use an inference profile ARN instead of model ID
- * - serviceTier: Specify service tier (Priority, Standard, Flex, Reserved)
- */
-export interface CustomChatBedrockConverseInput extends ChatBedrockConverseInput {
-    /**
-     * Application Inference Profile ARN to use for the model.
-     * For example, "arn:aws:bedrock:eu-west-1:123456789102:application-inference-profile/fm16bt65tzgx"
-     * When provided, this ARN will be used for the actual inference calls instead of the model ID.
-     * Must still provide `model` as normal modelId to benefit from all the metadata.
-     * @see https://docs.aws.amazon.com/bedrock/latest/userguide/inference-profiles-create.html
-     */
-    applicationInferenceProfile?: string;
-    /**
-     * Service tier for model invocation.
-     * Specifies the processing tier type used for serving the request.
-     * Supported values are 'priority', 'default', 'flex', and 'reserved'.
-     *
-     * - 'priority': Prioritized processing for lower latency
-     * - 'default': Standard processing tier
-     * - 'flex': Flexible processing tier with lower cost
-     * - 'reserved': Reserved capacity for consistent performance
-     *
-     * If not provided, AWS uses the default tier.
-     * Note: Requires AWS SDK >= 3.966.0 to work.
-     * @see https://docs.aws.amazon.com/bedrock/latest/userguide/service-tiers-inference.html
-     */
-    serviceTier?: ServiceTierType;
-}
-/**
- * Extended call options with serviceTier override support.
- */
-export interface CustomChatBedrockConverseCallOptions {
-    serviceTier?: ServiceTierType;
-}
+import { ChatGenerationChunk } from '@langchain/core/outputs';
+import type { CallbackManagerForLLMRun } from '@langchain/core/callbacks/manager';
 export declare class CustomChatBedrockConverse extends ChatBedrockConverse {
-    /**
-     * Application Inference Profile ARN to use instead of model ID.
-     */
-    applicationInferenceProfile?: string;
-    /**
-     * Service tier for model invocation.
-     */
-    serviceTier?: ServiceTierType;
-    constructor(fields?: CustomChatBedrockConverseInput);
+    constructor(fields?: ChatBedrockConverseInput);
     static lc_name(): string;
     /**
-     * Get the model ID to use for API calls.
-     * Returns applicationInferenceProfile if set, otherwise returns this.model.
-     */
-    protected getModelId(): string;
-    /**
-     * Override invocationParams to add serviceTier support.
-     */
-    invocationParams(options?: this['ParsedCallOptions'] & CustomChatBedrockConverseCallOptions): ReturnType<ChatBedrockConverse['invocationParams']> & {
-        serviceTier?: {
-            type: ServiceTierType;
-        };
-    };
-    /**
-     * Override _generateNonStreaming to use applicationInferenceProfile as modelId.
-     * Uses the same model-swapping pattern as streaming for consistency.
-     */
-    _generateNonStreaming(messages: BaseMessage[], options: this['ParsedCallOptions'] & CustomChatBedrockConverseCallOptions, runManager?: CallbackManagerForLLMRun): Promise<ChatResult>;
-    /**
-     * Override _streamResponseChunks to:
-     * 1. Use applicationInferenceProfile as modelId (by temporarily swapping this.model)
-     * 2. Strip contentBlockIndex from response_metadata to prevent merge conflicts
-     *
-     * Note: We delegate to super._streamResponseChunks() to preserve @langchain/aws's
-     * internal chunk handling which correctly preserves array content for reasoning blocks.
-     */
-    _streamResponseChunks(messages: BaseMessage[], options: this['ParsedCallOptions'] & CustomChatBedrockConverseCallOptions, runManager?: CallbackManagerForLLMRun): AsyncGenerator<ChatGenerationChunk>;
-    /**
-     * Clean a chunk by removing contentBlockIndex from response_metadata.
+     * Override _streamResponseChunks to strip contentBlockIndex from response_metadata
+     * This prevents LangChain's merge conflicts when the same index is used for
+     * different content types (text vs tool calls)
      */
-    private cleanChunk;
+    _streamResponseChunks(messages: BaseMessage[], options: this['ParsedCallOptions'], runManager?: CallbackManagerForLLMRun): AsyncGenerator<ChatGenerationChunk>;
     /**
      * Check if contentBlockIndex exists at any level in the object
      */

package/dist/types/types/tools.d.ts

@@ -57,8 +57,6 @@ export type FileRef = {
     id: string;
     name: string;
     path?: string;
-    /** Session ID this file belongs to (for multi-session file tracking) */
-    session_id?: string;
 };
 export type FileRefs = FileRef[];
 export type ExecuteResult = {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@librechat/agents",
-  "version": "3.0.81",
+  "version": "3.0.771",
   "main": "./dist/cjs/main.cjs",
   "module": "./dist/esm/main.mjs",
   "types": "./dist/types/index.d.ts",
@@ -51,14 +51,12 @@
     "image": "node -r dotenv/config --loader ./tsconfig-paths-bootstrap.mjs --experimental-specifier-resolution=node ./src/scripts/image.ts --provider 'google' --name 'Jo' --location 'New York, NY'",
     "code_exec_files": "node -r dotenv/config --loader ./tsconfig-paths-bootstrap.mjs --experimental-specifier-resolution=node ./src/scripts/code_exec_files.ts --provider 'openAI' --name 'Jo' --location 'New York, NY'",
     "code_exec_session": "node -r dotenv/config --loader ./tsconfig-paths-bootstrap.mjs --experimental-specifier-resolution=node ./src/scripts/code_exec_session.ts --provider 'openAI' --name 'Jo' --location 'New York, NY'",
-    "code_exec_multi_session": "node -r dotenv/config --loader ./tsconfig-paths-bootstrap.mjs --experimental-specifier-resolution=node ./src/scripts/code_exec_multi_session.ts --provider 'openAI' --name 'Jo' --location 'New York, NY'",
     "code_exec_simple": "node -r dotenv/config --loader ./tsconfig-paths-bootstrap.mjs --experimental-specifier-resolution=node ./src/scripts/code_exec_simple.ts --provider 'openAI' --name 'Jo' --location 'New York, NY'",
-    "simple": "node -r dotenv/config --loader ./tsconfig-paths-bootstrap.mjs --experimental-specifier-resolution=node ./src/scripts/simple.ts --provider 'bedrock' --name 'Jo' --location 'New York, NY'",
+    "simple": "node -r dotenv/config --loader ./tsconfig-paths-bootstrap.mjs --experimental-specifier-resolution=node ./src/scripts/simple.ts --provider 'openrouter' --name 'Jo' --location 'New York, NY'",
     "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'",
-    "thinking:bedrock": "node -r dotenv/config --loader ./tsconfig-paths-bootstrap.mjs --experimental-specifier-resolution=node ./src/scripts/thinking-bedrock.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": "node --trace-warnings -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": "node --trace-warnings -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 'bedrock' --name 'Jo' --location 'New York, NY'",
     "tool_search": "node -r dotenv/config --loader ./tsconfig-paths-bootstrap.mjs --experimental-specifier-resolution=node ./src/scripts/tool_search.ts",
     "programmatic_exec": "node -r dotenv/config --loader ./tsconfig-paths-bootstrap.mjs --experimental-specifier-resolution=node ./src/scripts/programmatic_exec.ts",
@@ -112,7 +110,6 @@
     }
   },
   "dependencies": {
-    "@aws-sdk/client-bedrock-runtime": "^3.970.0",
     "@langchain/anthropic": "^0.3.26",
     "@langchain/aws": "^0.1.15",
     "@langchain/core": "^0.3.80",

package/src/graphs/Graph.ts

@@ -1042,8 +1042,8 @@ export class StandardGraph extends Graph<t.BaseGraphState, t.GraphNode> {
 
     /**
      * Extract and store code execution session context from artifacts.
-     * Each file is stamped with its source session_id to support multi-session file tracking.
-     * When the same filename appears in a later execution, the newer version replaces the old.
+     * Only update session_id when files are generated - this ensures we don't
+     * lose the original session that contains the files.
      */
     const toolName = output.name;
     if (
@@ -1060,35 +1060,17 @@ export class StandardGraph extends Graph<t.BaseGraphState, t.GraphNode> {
         artifact.session_id !== ''
       ) {
         /**
-         * Stamp each new file with its source session_id.
-         * This enables files from different executions (parallel or sequential)
-         * to be tracked and passed to subsequent calls.
+         * Files were generated - update session with the new session_id.
+         * The new session_id is the one that contains these files.
          */
-        const filesWithSession: t.FileRefs = newFiles.map((file) => ({
-          ...file,
-          session_id: artifact.session_id,
-        }));
-
         const existingSession = this.sessions.get(Constants.EXECUTE_CODE) as
           | t.CodeSessionContext
           | undefined;
         const existingFiles = existingSession?.files ?? [];
 
-        /**
-         * Merge files, preferring latest versions by name.
-         * If a file with the same name exists, replace it with the new version.
-         * This handles cases where files are edited/recreated in subsequent executions.
-         */
-        const newFileNames = new Set(filesWithSession.map((f) => f.name));
-        const filteredExisting = existingFiles.filter(
-          (f) => !newFileNames.has(f.name)
-        );
-
         this.sessions.set(Constants.EXECUTE_CODE, {
-          /** Keep latest session_id for reference/fallback */
           session_id: artifact.session_id,
-          /** Accumulated files with latest versions preferred */
-          files: [...filteredExisting, ...filesWithSession],
+          files: [...existingFiles, ...newFiles],
           lastUpdated: Date.now(),
         });
       }

package/src/llm/bedrock/index.ts

@@ -1,9 +1,5 @@
 /**
  * Optimized ChatBedrockConverse wrapper that fixes contentBlockIndex conflicts
- * and adds support for latest @langchain/aws features:
- *
- * - Application Inference Profiles (PR #9129)
- * - Service Tiers (Priority/Standard/Flex) (PR #9785) - requires AWS SDK 3.966.0+
  *
  * Bedrock sends the same contentBlockIndex for both text and tool_use content blocks,
  * causing LangChain's merge logic to fail with "field[contentBlockIndex] already exists"
@@ -16,74 +12,15 @@
  */
 
 import { ChatBedrockConverse } from '@langchain/aws';
-import { AIMessageChunk } from '@langchain/core/messages';
-import { ChatGenerationChunk, ChatResult } from '@langchain/core/outputs';
-import type { CallbackManagerForLLMRun } from '@langchain/core/callbacks/manager';
 import type { ChatBedrockConverseInput } from '@langchain/aws';
+import { AIMessageChunk } from '@langchain/core/messages';
 import type { BaseMessage } from '@langchain/core/messages';
-
-/**
- * Service tier type for Bedrock invocations.
- * Requires AWS SDK >= 3.966.0 to actually work.
- * @see https://docs.aws.amazon.com/bedrock/latest/userguide/service-tiers-inference.html
- */
-export type ServiceTierType = 'priority' | 'default' | 'flex' | 'reserved';
-
-/**
- * Extended input interface with additional features:
- * - applicationInferenceProfile: Use an inference profile ARN instead of model ID
- * - serviceTier: Specify service tier (Priority, Standard, Flex, Reserved)
- */
-export interface CustomChatBedrockConverseInput
-  extends ChatBedrockConverseInput {
-  /**
-   * Application Inference Profile ARN to use for the model.
-   * For example, "arn:aws:bedrock:eu-west-1:123456789102:application-inference-profile/fm16bt65tzgx"
-   * When provided, this ARN will be used for the actual inference calls instead of the model ID.
-   * Must still provide `model` as normal modelId to benefit from all the metadata.
-   * @see https://docs.aws.amazon.com/bedrock/latest/userguide/inference-profiles-create.html
-   */
-  applicationInferenceProfile?: string;
-
-  /**
-   * Service tier for model invocation.
-   * Specifies the processing tier type used for serving the request.
-   * Supported values are 'priority', 'default', 'flex', and 'reserved'.
-   *
-   * - 'priority': Prioritized processing for lower latency
-   * - 'default': Standard processing tier
-   * - 'flex': Flexible processing tier with lower cost
-   * - 'reserved': Reserved capacity for consistent performance
-   *
-   * If not provided, AWS uses the default tier.
-   * Note: Requires AWS SDK >= 3.966.0 to work.
-   * @see https://docs.aws.amazon.com/bedrock/latest/userguide/service-tiers-inference.html
-   */
-  serviceTier?: ServiceTierType;
-}
-
-/**
- * Extended call options with serviceTier override support.
- */
-export interface CustomChatBedrockConverseCallOptions {
-  serviceTier?: ServiceTierType;
-}
+import { ChatGenerationChunk } from '@langchain/core/outputs';
+import type { CallbackManagerForLLMRun } from '@langchain/core/callbacks/manager';
 
 export class CustomChatBedrockConverse extends ChatBedrockConverse {
-  /**
-   * Application Inference Profile ARN to use instead of model ID.
-   */
-  applicationInferenceProfile?: string;
-
-  /**
-   * Service tier for model invocation.
-   */
-  serviceTier?: ServiceTierType;
-
-  constructor(fields?: CustomChatBedrockConverseInput) {
+  constructor(fields?: ChatBedrockConverseInput) {
     super(fields);
-    this.applicationInferenceProfile = fields?.applicationInferenceProfile;
-    this.serviceTier = fields?.serviceTier;
   }
 
   static lc_name(): string {
@@ -91,126 +28,52 @@ export class CustomChatBedrockConverse extends ChatBedrockConverse {
   }
 
   /**
-   * Get the model ID to use for API calls.
-   * Returns applicationInferenceProfile if set, otherwise returns this.model.
-   */
-  protected getModelId(): string {
-    return this.applicationInferenceProfile ?? this.model;
-  }
-
-  /**
-   * Override invocationParams to add serviceTier support.
-   */
-  override invocationParams(
-    options?: this['ParsedCallOptions'] & CustomChatBedrockConverseCallOptions
-  ): ReturnType<ChatBedrockConverse['invocationParams']> & {
-    serviceTier?: { type: ServiceTierType };
-  } {
-    const baseParams = super.invocationParams(options);
-
-    /** Service tier from options or fall back to class-level setting */
-    const serviceTierType = options?.serviceTier ?? this.serviceTier;
-
-    return {
-      ...baseParams,
-      serviceTier: serviceTierType ? { type: serviceTierType } : undefined,
-    };
-  }
-
-  /**
-   * Override _generateNonStreaming to use applicationInferenceProfile as modelId.
-   * Uses the same model-swapping pattern as streaming for consistency.
+   * Override _streamResponseChunks to strip contentBlockIndex from response_metadata
+   * This prevents LangChain's merge conflicts when the same index is used for
+   * different content types (text vs tool calls)
    */
-  override async _generateNonStreaming(
+  async *_streamResponseChunks(
     messages: BaseMessage[],
-    options: this['ParsedCallOptions'] & CustomChatBedrockConverseCallOptions,
-    runManager?: CallbackManagerForLLMRun
-  ): Promise<ChatResult> {
-    // Temporarily swap model for applicationInferenceProfile support
-    const originalModel = this.model;
-    if (
-      this.applicationInferenceProfile != null &&
-      this.applicationInferenceProfile !== ''
-    ) {
-      this.model = this.applicationInferenceProfile;
-    }
-
-    try {
-      return await super._generateNonStreaming(messages, options, runManager);
-    } finally {
-      // Restore original model
-      this.model = originalModel;
-    }
-  }
-
-  /**
-   * Override _streamResponseChunks to:
-   * 1. Use applicationInferenceProfile as modelId (by temporarily swapping this.model)
-   * 2. Strip contentBlockIndex from response_metadata to prevent merge conflicts
-   *
-   * Note: We delegate to super._streamResponseChunks() to preserve @langchain/aws's
-   * internal chunk handling which correctly preserves array content for reasoning blocks.
-   */
-  override async *_streamResponseChunks(
-    messages: BaseMessage[],
-    options: this['ParsedCallOptions'] & CustomChatBedrockConverseCallOptions,
+    options: this['ParsedCallOptions'],
     runManager?: CallbackManagerForLLMRun
   ): AsyncGenerator<ChatGenerationChunk> {
-    // Temporarily swap model for applicationInferenceProfile support
-    const originalModel = this.model;
-    if (
-      this.applicationInferenceProfile != null &&
-      this.applicationInferenceProfile !== ''
-    ) {
-      this.model = this.applicationInferenceProfile;
-    }
-
-    try {
-      // Use parent's streaming logic which correctly handles reasoning content
-      const baseStream = super._streamResponseChunks(
-        messages,
-        options,
-        runManager
-      );
-
-      for await (const chunk of baseStream) {
-        // Clean contentBlockIndex from response_metadata to prevent merge conflicts
-        yield this.cleanChunk(chunk);
+    const baseStream = super._streamResponseChunks(
+      messages,
+      options,
+      runManager
+    );
+
+    for await (const chunk of baseStream) {
+      // Only process if we have response_metadata
+      if (
+        chunk.message instanceof AIMessageChunk &&
+        (chunk.message as Partial<AIMessageChunk>).response_metadata &&
+        typeof chunk.message.response_metadata === 'object'
+      ) {
+        // Check if contentBlockIndex exists anywhere in response_metadata (top level or nested)
+        const hasContentBlockIndex = this.hasContentBlockIndex(
+          chunk.message.response_metadata
+        );
+
+        if (hasContentBlockIndex) {
+          const cleanedMetadata = this.removeContentBlockIndex(
+            chunk.message.response_metadata
+          ) as Record<string, unknown>;
+
+          yield new ChatGenerationChunk({
+            text: chunk.text,
+            message: new AIMessageChunk({
+              ...chunk.message,
+              response_metadata: cleanedMetadata,
+            }),
+            generationInfo: chunk.generationInfo,
+          });
+          continue;
+        }
       }
-    } finally {
-      // Restore original model
-      this.model = originalModel;
-    }
-  }
 
-  /**
-   * Clean a chunk by removing contentBlockIndex from response_metadata.
-   */
-  private cleanChunk(chunk: ChatGenerationChunk): ChatGenerationChunk {
-    const message = chunk.message;
-    if (!(message instanceof AIMessageChunk)) {
-      return chunk;
+      yield chunk;
     }
-
-    const metadata = message.response_metadata as Record<string, unknown>;
-    const hasContentBlockIndex = this.hasContentBlockIndex(metadata);
-    if (!hasContentBlockIndex) {
-      return chunk;
-    }
-
-    const cleanedMetadata = this.removeContentBlockIndex(metadata) as Record<
-      string,
-      unknown
-    >;
-
-    return new ChatGenerationChunk({
-      text: chunk.text,
-      message: new AIMessageChunk({
-        ...message,
-        response_metadata: cleanedMetadata,
-      }),
-      generationInfo: chunk.generationInfo,
-    });
   }
 
   /**

package/src/messages/core.ts

@@ -41,7 +41,7 @@ User: ${userMessage[1]}
 const _allowedTypes = ['image_url', 'text', 'tool_use', 'tool_result'];
 const allowedTypesByProvider: Record<string, string[]> = {
   default: _allowedTypes,
-  [Providers.ANTHROPIC]: [..._allowedTypes, 'thinking', 'redacted_thinking'],
+  [Providers.ANTHROPIC]: [..._allowedTypes, 'thinking'],
   [Providers.BEDROCK]: [..._allowedTypes, 'reasoning_content'],
   [Providers.OPENAI]: _allowedTypes,
 };

package/src/scripts/thinking.ts

@@ -1,11 +1,7 @@
 // src/scripts/test-thinking.ts
 import { config } from 'dotenv';
 config();
-import {
-  HumanMessage,
-  SystemMessage,
-  BaseMessage,
-} from '@langchain/core/messages';
+import { HumanMessage, SystemMessage, BaseMessage } from '@langchain/core/messages';
 import type { UsageMetadata } from '@langchain/core/messages';
 import * as t from '@/types';
 import { ChatModelStreamHandler, createContentAggregator } from '@/stream';
@@ -25,23 +21,17 @@ async function testThinking(): Promise<void> {
   const instructions = `You are a helpful AI assistant for ${userName}. When answering questions, be thorough in your reasoning.`;
   const { contentParts, aggregateContent } = createContentAggregator();
   _contentParts = contentParts as t.MessageContentComplex[];
-
+  
   // Set up event handlers
   const customHandlers = {
     [GraphEvents.TOOL_END]: new ToolEndHandler(),
     [GraphEvents.CHAT_MODEL_END]: new ModelEndHandler(collectedUsage),
     [GraphEvents.CHAT_MODEL_STREAM]: new ChatModelStreamHandler(),
     [GraphEvents.ON_RUN_STEP_COMPLETED]: {
-      handle: (
-        event: GraphEvents.ON_RUN_STEP_COMPLETED,
-        data: t.StreamEventData
-      ): void => {
+      handle: (event: GraphEvents.ON_RUN_STEP_COMPLETED, data: t.StreamEventData): void => {
         console.log('====== ON_RUN_STEP_COMPLETED ======');
-        aggregateContent({
-          event,
-          data: data as unknown as { result: t.ToolEndEvent },
-        });
-      },
+        aggregateContent({ event, data: data as unknown as { result: t.ToolEndEvent } });
+      }
     },
     [GraphEvents.ON_RUN_STEP]: {
       handle: (event: GraphEvents.ON_RUN_STEP, data: t.RunStep) => {
@@ -49,38 +39,29 @@ async function testThinking(): Promise<void> {
       },
     },
     [GraphEvents.ON_RUN_STEP_DELTA]: {
-      handle: (
-        event: GraphEvents.ON_RUN_STEP_DELTA,
-        data: t.RunStepDeltaEvent
-      ) => {
+      handle: (event: GraphEvents.ON_RUN_STEP_DELTA, data: t.RunStepDeltaEvent) => {
         aggregateContent({ event, data });
       },
     },
     [GraphEvents.ON_MESSAGE_DELTA]: {
-      handle: (
-        event: GraphEvents.ON_MESSAGE_DELTA,
-        data: t.MessageDeltaEvent
-      ) => {
+      handle: (event: GraphEvents.ON_MESSAGE_DELTA, data: t.MessageDeltaEvent) => {
         aggregateContent({ event, data });
       },
     },
     [GraphEvents.ON_REASONING_DELTA]: {
-      handle: (
-        event: GraphEvents.ON_REASONING_DELTA,
-        data: t.ReasoningDeltaEvent
-      ) => {
+      handle: (event: GraphEvents.ON_REASONING_DELTA, data: t.ReasoningDeltaEvent) => {
         aggregateContent({ event, data });
       },
     },
   };
 
   const baseLlmConfig: t.LLMConfig = getLLMConfig(Providers.ANTHROPIC);
-
+  
   // Enable thinking with token budget
   const llmConfig = {
     ...baseLlmConfig,
     model: 'claude-3-7-sonnet-latest',
-    thinking: { type: 'enabled', budget_tokens: 2000 },
+    thinking: { type: "enabled", budget_tokens: 2000 }
   };
 
   const run = await Run.create<t.IState>({
@@ -112,7 +93,7 @@ async function testThinking(): Promise<void> {
   console.log('Running first query with thinking enabled...');
   const firstInputs = { messages: [...conversationHistory] };
   await run.processStream(firstInputs, config);
-
+  
   // Extract and display thinking blocks
   const finalMessages = run.getRunMessages();
 
@@ -120,32 +101,30 @@ async function testThinking(): Promise<void> {
   console.log('\n\nTest 2: Multi-turn conversation with thinking enabled');
   const userMessage2 = `Given your previous analysis, what would be the most significant technical challenges in making this transition?`;
   conversationHistory.push(new HumanMessage(userMessage2));
-
+  
   console.log('Running second query with thinking enabled...');
   const secondInputs = { messages: [...conversationHistory] };
   await run.processStream(secondInputs, config);
-
+  
   // Display thinking blocks for second response
   const finalMessages2 = run.getRunMessages();
 
   // Test 3: Redacted thinking mode
   console.log('\n\nTest 3: Redacted thinking mode');
-  const magicString =
-    'ANTHROPIC_MAGIC_STRING_TRIGGER_REDACTED_THINKING_46C9A13E193C177646C7398A98432ECCCE4C1253D5E2D82641AC0E52CC2876CB';
+  const magicString = "ANTHROPIC_MAGIC_STRING_TRIGGER_REDACTED_THINKING_46C9A13E193C177646C7398A98432ECCCE4C1253D5E2D82641AC0E52CC2876CB";
   const userMessage3 = `${magicString}\n\nExplain how quantum computing works in simple terms.`;
-
+  
   // Reset conversation for clean test
   conversationHistory.length = 0;
   conversationHistory.push(new HumanMessage(userMessage3));
-
+  
   console.log('Running query with redacted thinking...');
   const thirdInputs = { messages: [...conversationHistory] };
   await run.processStream(thirdInputs, config);
-
+  
   // Display redacted thinking blocks
   const finalMessages3 = run.getRunMessages();
   console.log('\n\nThinking feature test completed!');
-  console.dir(finalMessages3, { depth: null });
 }
 
 process.on('unhandledRejection', (reason, promise) => {
@@ -168,4 +147,4 @@ testThinking().catch((err) => {
   console.log('Content parts:');
   console.dir(_contentParts, { depth: null });
   process.exit(1);
-});
+});

package/src/scripts/tools.ts

@@ -18,13 +18,9 @@ async function testStandardStreaming(): Promise<void> {
   const { userName, location, provider, currentDate } = await getArgs();
   const { contentParts, aggregateContent } = createContentAggregator();
   const customHandlers = {
-    [GraphEvents.TOOL_END]: new ToolEndHandler(
-      undefined,
-      undefined,
-      (name?: string) => {
-        return true;
-      }
-    ),
+    [GraphEvents.TOOL_END]: new ToolEndHandler(undefined, (name?: string) => {
+      return true;
+    }),
     [GraphEvents.CHAT_MODEL_END]: {
       handle: (
         _event: string,

package/src/stream.ts

@@ -339,8 +339,7 @@ hasToolCallChunks: ${hasToolCallChunks}
         (c) =>
           (c.type?.startsWith(ContentTypes.THINKING) ?? false) ||
           (c.type?.startsWith(ContentTypes.REASONING) ?? false) ||
-          (c.type?.startsWith(ContentTypes.REASONING_CONTENT) ?? false) ||
-          c.type === 'redacted_thinking'
+          (c.type?.startsWith(ContentTypes.REASONING_CONTENT) ?? false)
       )
     ) {
       await graph.dispatchReasoningDelta(stepId, {
@@ -366,8 +365,7 @@ hasToolCallChunks: ${hasToolCallChunks}
       Array.isArray(chunk.content) &&
       (chunk.content[0]?.type === ContentTypes.THINKING ||
         chunk.content[0]?.type === ContentTypes.REASONING ||
-        chunk.content[0]?.type === ContentTypes.REASONING_CONTENT ||
-        chunk.content[0]?.type === 'redacted_thinking')
+        chunk.content[0]?.type === ContentTypes.REASONING_CONTENT)
     ) {
       reasoning_content = 'valid';
     } else if (