langchain 1.0.0-alpha.8 → 1.0.1

package/dist/agents/tests/utils.d.ts

@@ -10,14 +10,14 @@ import { CallbackManagerForLLMRun } from "@langchain/core/callbacks/manager";
 
 //#region src/agents/tests/utils.d.ts
 
-interface ToolCall {
+interface ToolCall$1 {
   name: string;
   args: Record<string, any>;
   id: string;
   type?: "tool_call";
 }
 interface FakeToolCallingModelFields {
-  toolCalls?: ToolCall[][];
+  toolCalls?: ToolCall$1[][];
   toolStyle?: "openai" | "anthropic";
   index?: number;
   structuredResponse?: any;
@@ -28,9 +28,10 @@ interface FakeToolCallingModelFields {
  * Fake chat model for testing tool calling functionality
  */
 declare class FakeToolCallingModel extends BaseChatModel {
-  toolCalls: ToolCall[][];
+  toolCalls: ToolCall$1[][];
   toolStyle: "openai" | "anthropic";
-  index: number;
+  // Use a shared reference object so the index persists across bindTools calls
+  private indexRef;
   structuredResponse?: any;
   private tools;
   constructor({
@@ -38,8 +39,16 @@ declare class FakeToolCallingModel extends BaseChatModel {
     toolStyle,
     index,
     structuredResponse,
+    indexRef,
     ...rest
-  }?: FakeToolCallingModelFields);
+  }?: FakeToolCallingModelFields & {
+    indexRef?: {
+      current: number;
+    };
+  });
+  // Getter/setter for backwards compatibility
+  get index(): number;
+  set index(value: number);
   _llmType(): string;
   _combineLLMOutput(): never[];
   bindTools(tools: StructuredTool[]): FakeToolCallingModel | RunnableBinding<any, any, any & {

package/dist/agents/tests/utils.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"utils.d.ts","names":["CallbackManagerForLLMRun","BaseChatModel","BaseChatModelParams","BindToolsInput","ToolChoice","StructuredTool","BaseMessage","AIMessage","HumanMessage","BaseMessageFields","AIMessageFields","ToolMessage","ToolMessageFields","ChatResult","Runnable","RunnableConfig","RunnableLambda","RunnableBinding","MemorySaver","Checkpoint","CheckpointMetadata","BaseCheckpointSaver","LanguageModelLike","z","_AnyIdAIMessage","_AnyIdHumanMessage","_AnyIdToolMessage","FakeConfigurableModel","Record","Promise","FakeToolCallingChatModel","MemorySaverAssertImmutable","Uint8Array","ToolCall","FakeToolCallingModelFields","createCheckpointer","FakeToolCallingModel","toolCalls","toolStyle","index","structuredResponse","SearchAPI","ZodString","ZodTypeAny","ZodObject","this","schema","infer"],"sources":["../../../src/agents/tests/utils.d.ts"],"sourcesContent":["import { CallbackManagerForLLMRun } from \"@langchain/core/callbacks/manager\";\nimport { BaseChatModel, BaseChatModelParams, BindToolsInput, ToolChoice } from \"@langchain/core/language_models/chat_models\";\nimport { StructuredTool } from \"@langchain/core/tools\";\nimport { BaseMessage, AIMessage, HumanMessage, BaseMessageFields, AIMessageFields, ToolMessage, ToolMessageFields } from \"@langchain/core/messages\";\nimport { ChatResult } from \"@langchain/core/outputs\";\nimport { Runnable, RunnableConfig, RunnableLambda, RunnableBinding } from \"@langchain/core/runnables\";\nimport { MemorySaver, Checkpoint, CheckpointMetadata, type BaseCheckpointSaver } from \"@langchain/langgraph-checkpoint\";\nimport { LanguageModelLike } from \"@langchain/core/language_models/base\";\nimport { z } from \"zod/v3\";\nexport declare class _AnyIdAIMessage extends AIMessage {\n    get lc_id(): string[];\n    constructor(fields: AIMessageFields | string);\n}\nexport declare class _AnyIdHumanMessage extends HumanMessage {\n    get lc_id(): string[];\n    constructor(fields: BaseMessageFields | string);\n}\nexport declare class _AnyIdToolMessage extends ToolMessage {\n    get lc_id(): string[];\n    constructor(fields: ToolMessageFields);\n}\nexport declare class FakeConfigurableModel extends BaseChatModel {\n    _queuedMethodOperations: Record<string, any>;\n    _chatModel: LanguageModelLike;\n    constructor(fields: {\n        model: LanguageModelLike;\n    } & BaseChatModelParams);\n    _llmType(): string;\n    _generate(_messages: BaseMessage[], _options: this[\"ParsedCallOptions\"], _runManager?: CallbackManagerForLLMRun): Promise<ChatResult>;\n    _model(): Promise<LanguageModelLike>;\n    bindTools(tools: BindToolsInput[]): FakeConfigurableModel;\n}\nexport declare class FakeToolCallingChatModel extends BaseChatModel {\n    sleep?: number;\n    responses?: BaseMessage[];\n    thrownErrorString?: string;\n    idx: number;\n    toolStyle: \"openai\" | \"anthropic\" | \"bedrock\" | \"google\";\n    structuredResponse?: Record<string, unknown>;\n    // Track messages passed to structured output calls\n    structuredOutputMessages: BaseMessage[][];\n    constructor(fields: {\n        sleep?: number;\n        responses?: BaseMessage[];\n        thrownErrorString?: string;\n        toolStyle?: \"openai\" | \"anthropic\" | \"bedrock\" | \"google\";\n        structuredResponse?: Record<string, unknown>;\n    } & BaseChatModelParams);\n    _llmType(): string;\n    _generate(messages: BaseMessage[], _options: this[\"ParsedCallOptions\"], runManager?: CallbackManagerForLLMRun): Promise<ChatResult>;\n    bindTools(tools: BindToolsInput[]): Runnable<any>;\n    withStructuredOutput<RunOutput extends Record<string, any> = Record<string, any>>(_: unknown): Runnable<any>;\n}\nexport declare class MemorySaverAssertImmutable extends MemorySaver {\n    storageForCopies: Record<string, Record<string, Uint8Array>>;\n    constructor();\n    put(config: RunnableConfig, checkpoint: Checkpoint, metadata: CheckpointMetadata): Promise<RunnableConfig>;\n}\ninterface ToolCall {\n    name: string;\n    args: Record<string, any>;\n    id: string;\n    type?: \"tool_call\";\n}\ninterface FakeToolCallingModelFields {\n    toolCalls?: ToolCall[][];\n    toolStyle?: \"openai\" | \"anthropic\";\n    index?: number;\n    structuredResponse?: any;\n}\n// Helper function to create checkpointer\nexport declare function createCheckpointer(): BaseCheckpointSaver;\n/**\n * Fake chat model for testing tool calling functionality\n */\nexport declare class FakeToolCallingModel extends BaseChatModel {\n    toolCalls: ToolCall[][];\n    toolStyle: \"openai\" | \"anthropic\";\n    index: number;\n    structuredResponse?: any;\n    private tools;\n    constructor({ toolCalls, toolStyle, index, structuredResponse, ...rest }?: FakeToolCallingModelFields);\n    _llmType(): string;\n    _combineLLMOutput(): never[];\n    bindTools(tools: StructuredTool[]): FakeToolCallingModel | RunnableBinding<any, any, any & {\n        tool_choice?: ToolChoice | undefined;\n    }>;\n    withStructuredOutput(_schema: any): RunnableLambda<unknown, any, RunnableConfig<Record<string, any>>>;\n    _generate(messages: BaseMessage[], _options?: this[\"ParsedCallOptions\"], _runManager?: CallbackManagerForLLMRun): Promise<ChatResult>;\n}\nexport declare class SearchAPI extends StructuredTool {\n    name: string;\n    description: string;\n    schema: z.ZodObject<{\n        query: z.ZodString;\n    }, \"strip\", z.ZodTypeAny, {\n        query: string;\n    }, {\n        query: string;\n    }>;\n    _call(input: z.infer<typeof this.schema>): Promise<string>;\n}\nexport {};\n"],"mappings":";;;;;;;;;;;;UA0DUiC,QAAAA;;QAEAL;;;;UAIAM,0BAAAA;cACMD;;;;;;;;;;cAUKG,oBAAAA,SAA6BnC,aAAAA;aACnCgC;;;;;;;;;;;MAKgEC;;;mBAG1D7B,mBAAmB+B,uBAAuBnB;kBACzCb;;sCAEkBY,6BAA6BD,eAAea;sBAC5DtB,mEAAmEN,2BAA2B6B,QAAQhB"}
+{"version":3,"file":"utils.d.ts","names":["CallbackManagerForLLMRun","BaseChatModel","BaseChatModelParams","BindToolsInput","ToolChoice","StructuredTool","BaseMessage","AIMessage","HumanMessage","BaseMessageFields","AIMessageFields","ToolMessage","ToolMessageFields","ChatResult","Runnable","RunnableConfig","RunnableLambda","RunnableBinding","MemorySaver","Checkpoint","CheckpointMetadata","BaseCheckpointSaver","LanguageModelLike","z","_AnyIdAIMessage","_AnyIdHumanMessage","_AnyIdToolMessage","FakeConfigurableModel","Record","Promise","FakeToolCallingChatModel","MemorySaverAssertImmutable","Uint8Array","ToolCall","FakeToolCallingModelFields","createCheckpointer","FakeToolCallingModel","toolCalls","toolStyle","index","structuredResponse","indexRef","SearchAPI","ZodString","ZodTypeAny","ZodObject","this","schema","infer"],"sources":["../../../src/agents/tests/utils.d.ts"],"sourcesContent":["import { CallbackManagerForLLMRun } from \"@langchain/core/callbacks/manager\";\nimport { BaseChatModel, BaseChatModelParams, BindToolsInput, ToolChoice } from \"@langchain/core/language_models/chat_models\";\nimport { StructuredTool } from \"@langchain/core/tools\";\nimport { BaseMessage, AIMessage, HumanMessage, BaseMessageFields, AIMessageFields, ToolMessage, ToolMessageFields } from \"@langchain/core/messages\";\nimport { ChatResult } from \"@langchain/core/outputs\";\nimport { Runnable, RunnableConfig, RunnableLambda, RunnableBinding } from \"@langchain/core/runnables\";\nimport { MemorySaver, Checkpoint, CheckpointMetadata, type BaseCheckpointSaver } from \"@langchain/langgraph-checkpoint\";\nimport { LanguageModelLike } from \"@langchain/core/language_models/base\";\nimport { z } from \"zod/v3\";\nexport declare class _AnyIdAIMessage extends AIMessage {\n    get lc_id(): string[];\n    constructor(fields: AIMessageFields | string);\n}\nexport declare class _AnyIdHumanMessage extends HumanMessage {\n    get lc_id(): string[];\n    constructor(fields: BaseMessageFields | string);\n}\nexport declare class _AnyIdToolMessage extends ToolMessage {\n    get lc_id(): string[];\n    constructor(fields: ToolMessageFields);\n}\nexport declare class FakeConfigurableModel extends BaseChatModel {\n    _queuedMethodOperations: Record<string, any>;\n    _chatModel: LanguageModelLike;\n    constructor(fields: {\n        model: LanguageModelLike;\n    } & BaseChatModelParams);\n    _llmType(): string;\n    _generate(_messages: BaseMessage[], _options: this[\"ParsedCallOptions\"], _runManager?: CallbackManagerForLLMRun): Promise<ChatResult>;\n    _model(): Promise<LanguageModelLike>;\n    bindTools(tools: BindToolsInput[]): FakeConfigurableModel;\n}\nexport declare class FakeToolCallingChatModel extends BaseChatModel {\n    sleep?: number;\n    responses?: BaseMessage[];\n    thrownErrorString?: string;\n    idx: number;\n    toolStyle: \"openai\" | \"anthropic\" | \"bedrock\" | \"google\";\n    structuredResponse?: Record<string, unknown>;\n    // Track messages passed to structured output calls\n    structuredOutputMessages: BaseMessage[][];\n    constructor(fields: {\n        sleep?: number;\n        responses?: BaseMessage[];\n        thrownErrorString?: string;\n        toolStyle?: \"openai\" | \"anthropic\" | \"bedrock\" | \"google\";\n        structuredResponse?: Record<string, unknown>;\n    } & BaseChatModelParams);\n    _llmType(): string;\n    _generate(messages: BaseMessage[], _options: this[\"ParsedCallOptions\"], runManager?: CallbackManagerForLLMRun): Promise<ChatResult>;\n    bindTools(tools: BindToolsInput[]): Runnable<any>;\n    withStructuredOutput<RunOutput extends Record<string, any> = Record<string, any>>(_: unknown): Runnable<any>;\n}\nexport declare class MemorySaverAssertImmutable extends MemorySaver {\n    storageForCopies: Record<string, Record<string, Uint8Array>>;\n    constructor();\n    put(config: RunnableConfig, checkpoint: Checkpoint, metadata: CheckpointMetadata): Promise<RunnableConfig>;\n}\ninterface ToolCall {\n    name: string;\n    args: Record<string, any>;\n    id: string;\n    type?: \"tool_call\";\n}\ninterface FakeToolCallingModelFields {\n    toolCalls?: ToolCall[][];\n    toolStyle?: \"openai\" | \"anthropic\";\n    index?: number;\n    structuredResponse?: any;\n}\n// Helper function to create checkpointer\nexport declare function createCheckpointer(): BaseCheckpointSaver;\n/**\n * Fake chat model for testing tool calling functionality\n */\nexport declare class FakeToolCallingModel extends BaseChatModel {\n    toolCalls: ToolCall[][];\n    toolStyle: \"openai\" | \"anthropic\";\n    // Use a shared reference object so the index persists across bindTools calls\n    private indexRef;\n    structuredResponse?: any;\n    private tools;\n    constructor({ toolCalls, toolStyle, index, structuredResponse, indexRef, ...rest }?: FakeToolCallingModelFields & {\n        indexRef?: {\n            current: number;\n        };\n    });\n    // Getter/setter for backwards compatibility\n    get index(): number;\n    set index(value: number);\n    _llmType(): string;\n    _combineLLMOutput(): never[];\n    bindTools(tools: StructuredTool[]): FakeToolCallingModel | RunnableBinding<any, any, any & {\n        tool_choice?: ToolChoice | undefined;\n    }>;\n    withStructuredOutput(_schema: any): RunnableLambda<unknown, any, RunnableConfig<Record<string, any>>>;\n    _generate(messages: BaseMessage[], _options?: this[\"ParsedCallOptions\"], _runManager?: CallbackManagerForLLMRun): Promise<ChatResult>;\n}\nexport declare class SearchAPI extends StructuredTool {\n    name: string;\n    description: string;\n    schema: z.ZodObject<{\n        query: z.ZodString;\n    }, \"strip\", z.ZodTypeAny, {\n        query: string;\n    }, {\n        query: string;\n    }>;\n    _call(input: z.infer<typeof this.schema>): Promise<string>;\n}\nexport {};\n"],"mappings":";;;;;;;;;;;;UA0DUiC,UAAAA;;QAEAL;;;;UAIAM,0BAAAA;cACMD;;;;;;;;;;cAUKG,oBAAAA,SAA6BnC,aAAAA;aACnCgC;;;;;;;;;;;;;MAM0EC;;;;;;;;;;mBAUpE7B,mBAAmB+B,uBAAuBnB;kBACzCb;;sCAEkBY,6BAA6BD,eAAea;sBAC5DtB,mEAAmEN,2BAA2B6B,QAAQhB"}

package/dist/agents/tests/utils.js

@@ -12,16 +12,22 @@ import "@langchain/langgraph-checkpoint";
 var FakeToolCallingModel = class FakeToolCallingModel extends BaseChatModel {
 	toolCalls;
 	toolStyle;
-	index;
+	indexRef;
 	structuredResponse;
 	tools = [];
-	constructor({ toolCalls = [], toolStyle = "openai", index = 0, structuredResponse,...rest } = {}) {
+	constructor({ toolCalls = [], toolStyle = "openai", index = 0, structuredResponse, indexRef,...rest } = {}) {
 		super(rest);
 		this.toolCalls = toolCalls;
 		this.toolStyle = toolStyle;
-		this.index = index;
+		this.indexRef = indexRef ?? { current: index };
 		this.structuredResponse = structuredResponse;
 	}
+	get index() {
+		return this.indexRef.current;
+	}
+	set index(value) {
+		this.indexRef.current = value;
+	}
 	_llmType() {
 		return "fake-tool-calling";
 	}
@@ -32,8 +38,8 @@ var FakeToolCallingModel = class FakeToolCallingModel extends BaseChatModel {
 		const newInstance = new FakeToolCallingModel({
 			toolCalls: this.toolCalls,
 			toolStyle: this.toolStyle,
-			index: this.index,
-			structuredResponse: this.structuredResponse
+			structuredResponse: this.structuredResponse,
+			indexRef: this.indexRef
 		});
 		newInstance.tools = [...this.tools, ...tools];
 		return newInstance;
@@ -50,6 +56,8 @@ var FakeToolCallingModel = class FakeToolCallingModel extends BaseChatModel {
 			const parts = messages.map((m) => m.content).filter(Boolean);
 			content = parts.join("-");
 		}
+		const isStartOfConversation = messages.length === 1 || messages.length === 2 && messages.every(HumanMessage.isInstance);
+		if (isStartOfConversation && this.index !== 0) this.index = 0;
 		const currentToolCalls = this.toolCalls[this.index] || [];
 		const messageId = this.index.toString();
 		this.index = (this.index + 1) % Math.max(1, this.toolCalls.length);

package/dist/agents/tests/utils.js.map

@@ -1 +1 @@
-{"version":3,"file":"utils.js","names":["tools: StructuredTool[]","_schema: any","messages: BaseMessage[]","_options?: this[\"ParsedCallOptions\"]","_runManager?: CallbackManagerForLLMRun"],"sources":["../../../src/agents/tests/utils.ts"],"sourcesContent":["/* eslint-disable @typescript-eslint/no-explicit-any */\n/* eslint-disable import/no-extraneous-dependencies */\nimport { expect } from \"vitest\";\nimport { CallbackManagerForLLMRun } from \"@langchain/core/callbacks/manager\";\nimport {\n  BaseChatModel,\n  BaseChatModelParams,\n  BaseChatModelCallOptions,\n  BindToolsInput,\n  ToolChoice,\n} from \"@langchain/core/language_models/chat_models\";\nimport { StructuredTool } from \"@langchain/core/tools\";\nimport {\n  BaseMessage,\n  AIMessage,\n  HumanMessage,\n  BaseMessageFields,\n  AIMessageFields,\n  ToolMessage,\n  ToolMessageFields,\n} from \"@langchain/core/messages\";\nimport { ChatResult } from \"@langchain/core/outputs\";\nimport {\n  Runnable,\n  RunnableConfig,\n  RunnableLambda,\n  RunnableBinding,\n} from \"@langchain/core/runnables\";\nimport {\n  MemorySaver,\n  Checkpoint,\n  CheckpointMetadata,\n  type BaseCheckpointSaver,\n} from \"@langchain/langgraph-checkpoint\";\nimport { LanguageModelLike } from \"@langchain/core/language_models/base\";\nimport { z } from \"zod/v3\";\n\nexport class _AnyIdAIMessage extends AIMessage {\n  get lc_id() {\n    return [\"langchain_core\", \"messages\", \"AIMessage\"];\n  }\n\n  constructor(fields: AIMessageFields | string) {\n    let fieldsWithJestMatcher: Partial<AIMessageFields> = {\n      id: expect.any(String) as unknown as string,\n    };\n    if (typeof fields === \"string\") {\n      fieldsWithJestMatcher = {\n        content: fields,\n        ...fieldsWithJestMatcher,\n      };\n    } else {\n      fieldsWithJestMatcher = {\n        ...fields,\n        ...fieldsWithJestMatcher,\n      };\n    }\n    super(fieldsWithJestMatcher as AIMessageFields);\n  }\n}\n\nexport class _AnyIdHumanMessage extends HumanMessage {\n  get lc_id() {\n    return [\"langchain_core\", \"messages\", \"HumanMessage\"];\n  }\n\n  constructor(fields: BaseMessageFields | string) {\n    let fieldsWithJestMatcher: Partial<BaseMessageFields> = {\n      id: expect.any(String) as unknown as string,\n    };\n    if (typeof fields === \"string\") {\n      fieldsWithJestMatcher = {\n        content: fields,\n        ...fieldsWithJestMatcher,\n      };\n    } else {\n      fieldsWithJestMatcher = {\n        ...fields,\n        ...fieldsWithJestMatcher,\n      };\n    }\n    super(fieldsWithJestMatcher as BaseMessageFields);\n  }\n}\n\nexport class _AnyIdToolMessage extends ToolMessage {\n  get lc_id() {\n    return [\"langchain_core\", \"messages\", \"ToolMessage\"];\n  }\n\n  constructor(fields: ToolMessageFields) {\n    const fieldsWithJestMatcher: Partial<ToolMessageFields> = {\n      id: expect.any(String) as unknown as string,\n      ...fields,\n    };\n    super(fieldsWithJestMatcher as ToolMessageFields);\n  }\n}\n\nexport class FakeConfigurableModel extends BaseChatModel {\n  _queuedMethodOperations: Record<string, any> = {};\n\n  _chatModel: LanguageModelLike;\n\n  constructor(\n    fields: {\n      model: LanguageModelLike;\n    } & BaseChatModelParams\n  ) {\n    super(fields);\n    this._chatModel = fields.model;\n  }\n\n  _llmType() {\n    return \"fake_configurable\";\n  }\n\n  async _generate(\n    _messages: BaseMessage[],\n    _options: this[\"ParsedCallOptions\"],\n    _runManager?: CallbackManagerForLLMRun\n  ): Promise<ChatResult> {\n    throw new Error(\"Not implemented\");\n  }\n\n  async _model() {\n    return this._chatModel;\n  }\n\n  bindTools(tools: BindToolsInput[]) {\n    const modelWithTools = new FakeConfigurableModel({\n      model: (this._chatModel as FakeToolCallingChatModel).bindTools(tools),\n    });\n    modelWithTools._queuedMethodOperations.bindTools = tools;\n    return modelWithTools;\n  }\n}\n\nexport class FakeToolCallingChatModel extends BaseChatModel {\n  sleep?: number = 50;\n\n  responses?: BaseMessage[];\n\n  thrownErrorString?: string;\n\n  idx: number;\n\n  toolStyle: \"openai\" | \"anthropic\" | \"bedrock\" | \"google\" = \"openai\";\n\n  structuredResponse?: Record<string, unknown>;\n\n  // Track messages passed to structured output calls\n  structuredOutputMessages: BaseMessage[][] = [];\n\n  constructor(\n    fields: {\n      sleep?: number;\n      responses?: BaseMessage[];\n      thrownErrorString?: string;\n      toolStyle?: \"openai\" | \"anthropic\" | \"bedrock\" | \"google\";\n      structuredResponse?: Record<string, unknown>;\n    } & BaseChatModelParams\n  ) {\n    super(fields);\n    this.sleep = fields.sleep ?? this.sleep;\n    this.responses = fields.responses;\n    this.thrownErrorString = fields.thrownErrorString;\n    this.idx = 0;\n    this.toolStyle = fields.toolStyle ?? this.toolStyle;\n    this.structuredResponse = fields.structuredResponse;\n    this.structuredOutputMessages = [];\n  }\n\n  _llmType() {\n    return \"fake\";\n  }\n\n  async _generate(\n    messages: BaseMessage[],\n    _options: this[\"ParsedCallOptions\"],\n    runManager?: CallbackManagerForLLMRun\n  ): Promise<ChatResult> {\n    if (this.thrownErrorString) {\n      throw new Error(this.thrownErrorString);\n    }\n    if (this.sleep !== undefined) {\n      await new Promise((resolve) => setTimeout(resolve, this.sleep));\n    }\n    const responses = this.responses?.length ? this.responses : messages;\n    const msg = responses[this.idx % responses.length];\n    const generation: ChatResult = {\n      generations: [\n        {\n          text: \"\",\n          message: msg,\n        },\n      ],\n    };\n    this.idx += 1;\n\n    if (typeof msg.content === \"string\") {\n      await runManager?.handleLLMNewToken(msg.content);\n    }\n    return generation;\n  }\n\n  bindTools(tools: BindToolsInput[]): Runnable<any> {\n    const toolDicts = [];\n    const serverTools = [];\n    for (const tool of tools) {\n      if (!(\"name\" in tool)) {\n        serverTools.push(tool);\n        continue;\n      }\n\n      // NOTE: this is a simplified tool spec for testing purposes only\n      if (this.toolStyle === \"openai\") {\n        toolDicts.push({\n          type: \"function\",\n          function: {\n            name: tool.name,\n          },\n        });\n      } else if ([\"anthropic\", \"google\"].includes(this.toolStyle)) {\n        toolDicts.push({\n          name: tool.name,\n        });\n      } else if (this.toolStyle === \"bedrock\") {\n        toolDicts.push({\n          toolSpec: {\n            name: tool.name,\n          },\n        });\n      }\n    }\n    let toolsToBind: BindToolsInput[] = toolDicts;\n    if (this.toolStyle === \"google\") {\n      toolsToBind = [{ functionDeclarations: toolDicts }];\n    }\n    return this.bind({\n      tools: [...toolsToBind, ...serverTools],\n    } as BaseChatModelCallOptions);\n  }\n\n  withStructuredOutput<\n    RunOutput extends Record<string, any> = Record<string, any>\n  >(_: unknown): Runnable<any> {\n    if (!this.structuredResponse) {\n      throw new Error(\"No structured response provided\");\n    }\n    // Create a runnable that returns the proper structured format\n    return RunnableLambda.from(async (messages: BaseMessage[]) => {\n      if (this.sleep) {\n        await new Promise((resolve) => setTimeout(resolve, this.sleep));\n      }\n\n      // Store the messages that were sent to generate structured output\n      this.structuredOutputMessages.push([...messages]);\n\n      // Return in the format expected: { raw: BaseMessage, parsed: RunOutput }\n      return this.structuredResponse as RunOutput;\n    });\n  }\n}\n\nexport class MemorySaverAssertImmutable extends MemorySaver {\n  storageForCopies: Record<string, Record<string, Uint8Array>> = {};\n\n  constructor() {\n    super();\n    this.storageForCopies = {};\n  }\n\n  async put(\n    config: RunnableConfig,\n    checkpoint: Checkpoint,\n    metadata: CheckpointMetadata\n  ): Promise<RunnableConfig> {\n    const thread_id = config.configurable?.thread_id;\n    this.storageForCopies[thread_id] ??= {};\n\n    // assert checkpoint hasn't been modified since last written\n    const saved = await this.get(config);\n    if (saved) {\n      const savedId = saved.id;\n      if (this.storageForCopies[thread_id][savedId]) {\n        const loaded = await this.serde.loadsTyped(\n          \"json\",\n          this.storageForCopies[thread_id][savedId]\n        );\n\n        expect(\n          saved,\n          `Checkpoint [${savedId}] has been modified since last written`\n        ).toEqual(loaded);\n      }\n    }\n    const [, serializedCheckpoint] = await this.serde.dumpsTyped(checkpoint);\n    // save a copy of the checkpoint\n    this.storageForCopies[thread_id][checkpoint.id] = serializedCheckpoint;\n\n    return super.put(config, checkpoint, metadata);\n  }\n}\n\ninterface ToolCall {\n  name: string;\n  args: Record<string, any>;\n  id: string;\n  type?: \"tool_call\";\n}\n\ninterface FakeToolCallingModelFields {\n  toolCalls?: ToolCall[][];\n  toolStyle?: \"openai\" | \"anthropic\";\n  index?: number;\n  structuredResponse?: any;\n}\n\n// Helper function to create checkpointer\nexport function createCheckpointer(): BaseCheckpointSaver {\n  return new MemorySaver();\n}\n\n/**\n * Fake chat model for testing tool calling functionality\n */\nexport class FakeToolCallingModel extends BaseChatModel {\n  toolCalls: ToolCall[][];\n\n  toolStyle: \"openai\" | \"anthropic\";\n\n  index: number;\n\n  structuredResponse?: any;\n\n  private tools: StructuredTool[] = [];\n\n  constructor({\n    toolCalls = [],\n    toolStyle = \"openai\",\n    index = 0,\n    structuredResponse,\n    ...rest\n  }: FakeToolCallingModelFields = {}) {\n    super(rest);\n    this.toolCalls = toolCalls;\n    this.toolStyle = toolStyle;\n    this.index = index;\n    this.structuredResponse = structuredResponse;\n  }\n\n  _llmType(): string {\n    return \"fake-tool-calling\";\n  }\n\n  _combineLLMOutput() {\n    return [];\n  }\n\n  bindTools(\n    tools: StructuredTool[]\n  ):\n    | FakeToolCallingModel\n    | RunnableBinding<\n        any,\n        any,\n        any & { tool_choice?: ToolChoice | undefined }\n      > {\n    const newInstance = new FakeToolCallingModel({\n      toolCalls: this.toolCalls,\n      toolStyle: this.toolStyle,\n      index: this.index,\n      structuredResponse: this.structuredResponse,\n    });\n    newInstance.tools = [...this.tools, ...tools];\n    return newInstance;\n  }\n\n  withStructuredOutput(_schema: any) {\n    return new RunnableLambda({\n      func: async () => {\n        return this.structuredResponse;\n      },\n    });\n  }\n\n  async _generate(\n    messages: BaseMessage[],\n    _options?: this[\"ParsedCallOptions\"],\n    _runManager?: CallbackManagerForLLMRun\n  ): Promise<ChatResult> {\n    const lastMessage = messages[messages.length - 1];\n    let content = lastMessage.content as string;\n\n    // Handle prompt concatenation\n    if (messages.length > 1) {\n      const parts = messages.map((m) => m.content as string).filter(Boolean);\n      content = parts.join(\"-\");\n    }\n\n    const currentToolCalls = this.toolCalls[this.index] || [];\n    const messageId = this.index.toString();\n\n    // Move to next set of tool calls for subsequent invocations\n    this.index = (this.index + 1) % Math.max(1, this.toolCalls.length);\n\n    const message = new AIMessage({\n      content,\n      id: messageId,\n      tool_calls:\n        currentToolCalls.length > 0\n          ? currentToolCalls.map((tc) => ({\n              ...tc,\n              type: \"tool_call\" as const,\n            }))\n          : undefined,\n    });\n\n    return {\n      generations: [\n        {\n          text: content,\n          message,\n        },\n      ],\n      llmOutput: {},\n    };\n  }\n}\n\nexport class SearchAPI extends StructuredTool {\n  name = \"search_api\";\n\n  description = \"A simple API that returns the input string.\";\n\n  schema = z.object({\n    query: z.string().describe(\"The query to search for.\"),\n  });\n\n  async _call(input: z.infer<typeof this.schema>) {\n    if (input?.query === \"error\") {\n      throw new Error(\"Error\");\n    }\n    return `result for ${input?.query}`;\n  }\n}\n"],"mappings":";;;;;;;;;;;AAuUA,IAAa,uBAAb,MAAa,6BAA6B,cAAc;CACtD;CAEA;CAEA;CAEA;CAEA,AAAQ,QAA0B,CAAE;CAEpC,YAAY,EACV,YAAY,CAAE,GACd,YAAY,UACZ,QAAQ,GACR,mBACA,GAAG,MACwB,GAAG,CAAE,GAAE;EAClC,MAAM,KAAK;EACX,KAAK,YAAY;EACjB,KAAK,YAAY;EACjB,KAAK,QAAQ;EACb,KAAK,qBAAqB;CAC3B;CAED,WAAmB;AACjB,SAAO;CACR;CAED,oBAAoB;AAClB,SAAO,CAAE;CACV;CAED,UACEA,OAOI;EACJ,MAAM,cAAc,IAAI,qBAAqB;GAC3C,WAAW,KAAK;GAChB,WAAW,KAAK;GAChB,OAAO,KAAK;GACZ,oBAAoB,KAAK;EAC1B;EACD,YAAY,QAAQ,CAAC,GAAG,KAAK,OAAO,GAAG,KAAM;AAC7C,SAAO;CACR;CAED,qBAAqBC,SAAc;AACjC,SAAO,IAAI,eAAe,EACxB,MAAM,YAAY;AAChB,UAAO,KAAK;EACb,EACF;CACF;CAED,MAAM,UACJC,UACAC,UACAC,aACqB;EACrB,MAAM,cAAc,SAAS,SAAS,SAAS;EAC/C,IAAI,UAAU,YAAY;AAG1B,MAAI,SAAS,SAAS,GAAG;GACvB,MAAM,QAAQ,SAAS,IAAI,CAAC,MAAM,EAAE,QAAkB,CAAC,OAAO,QAAQ;GACtE,UAAU,MAAM,KAAK,IAAI;EAC1B;EAED,MAAM,mBAAmB,KAAK,UAAU,KAAK,UAAU,CAAE;EACzD,MAAM,YAAY,KAAK,MAAM,UAAU;EAGvC,KAAK,SAAS,KAAK,QAAQ,KAAK,KAAK,IAAI,GAAG,KAAK,UAAU,OAAO;EAElE,MAAM,UAAU,IAAI,UAAU;GAC5B;GACA,IAAI;GACJ,YACE,iBAAiB,SAAS,IACtB,iBAAiB,IAAI,CAAC,QAAQ;IAC5B,GAAG;IACH,MAAM;GACP,GAAE,GACH;EACP;AAED,SAAO;GACL,aAAa,CACX;IACE,MAAM;IACN;GACD,CACF;GACD,WAAW,CAAE;EACd;CACF;AACF"}
+{"version":3,"file":"utils.js","names":["value: number","tools: StructuredTool[]","_schema: any","messages: BaseMessage[]","_options?: this[\"ParsedCallOptions\"]","_runManager?: CallbackManagerForLLMRun"],"sources":["../../../src/agents/tests/utils.ts"],"sourcesContent":["/* eslint-disable @typescript-eslint/no-explicit-any */\n/* eslint-disable import/no-extraneous-dependencies */\nimport { expect } from \"vitest\";\nimport { CallbackManagerForLLMRun } from \"@langchain/core/callbacks/manager\";\nimport {\n  BaseChatModel,\n  BaseChatModelParams,\n  BaseChatModelCallOptions,\n  BindToolsInput,\n  ToolChoice,\n} from \"@langchain/core/language_models/chat_models\";\nimport { StructuredTool } from \"@langchain/core/tools\";\nimport {\n  BaseMessage,\n  AIMessage,\n  HumanMessage,\n  BaseMessageFields,\n  AIMessageFields,\n  ToolMessage,\n  ToolMessageFields,\n} from \"@langchain/core/messages\";\nimport { ChatResult } from \"@langchain/core/outputs\";\nimport {\n  Runnable,\n  RunnableConfig,\n  RunnableLambda,\n  RunnableBinding,\n} from \"@langchain/core/runnables\";\nimport {\n  MemorySaver,\n  Checkpoint,\n  CheckpointMetadata,\n  type BaseCheckpointSaver,\n} from \"@langchain/langgraph-checkpoint\";\nimport { LanguageModelLike } from \"@langchain/core/language_models/base\";\nimport { z } from \"zod/v3\";\n\nexport class _AnyIdAIMessage extends AIMessage {\n  get lc_id() {\n    return [\"langchain_core\", \"messages\", \"AIMessage\"];\n  }\n\n  constructor(fields: AIMessageFields | string) {\n    let fieldsWithJestMatcher: Partial<AIMessageFields> = {\n      id: expect.any(String) as unknown as string,\n    };\n    if (typeof fields === \"string\") {\n      fieldsWithJestMatcher = {\n        content: fields,\n        ...fieldsWithJestMatcher,\n      };\n    } else {\n      fieldsWithJestMatcher = {\n        ...fields,\n        ...fieldsWithJestMatcher,\n      };\n    }\n    super(fieldsWithJestMatcher as AIMessageFields);\n  }\n}\n\nexport class _AnyIdHumanMessage extends HumanMessage {\n  get lc_id() {\n    return [\"langchain_core\", \"messages\", \"HumanMessage\"];\n  }\n\n  constructor(fields: BaseMessageFields | string) {\n    let fieldsWithJestMatcher: Partial<BaseMessageFields> = {\n      id: expect.any(String) as unknown as string,\n    };\n    if (typeof fields === \"string\") {\n      fieldsWithJestMatcher = {\n        content: fields,\n        ...fieldsWithJestMatcher,\n      };\n    } else {\n      fieldsWithJestMatcher = {\n        ...fields,\n        ...fieldsWithJestMatcher,\n      };\n    }\n    super(fieldsWithJestMatcher as BaseMessageFields);\n  }\n}\n\nexport class _AnyIdToolMessage extends ToolMessage {\n  get lc_id() {\n    return [\"langchain_core\", \"messages\", \"ToolMessage\"];\n  }\n\n  constructor(fields: ToolMessageFields) {\n    const fieldsWithJestMatcher: Partial<ToolMessageFields> = {\n      id: expect.any(String) as unknown as string,\n      ...fields,\n    };\n    super(fieldsWithJestMatcher as ToolMessageFields);\n  }\n}\n\nexport class FakeConfigurableModel extends BaseChatModel {\n  _queuedMethodOperations: Record<string, any> = {};\n\n  _chatModel: LanguageModelLike;\n\n  constructor(\n    fields: {\n      model: LanguageModelLike;\n    } & BaseChatModelParams\n  ) {\n    super(fields);\n    this._chatModel = fields.model;\n  }\n\n  _llmType() {\n    return \"fake_configurable\";\n  }\n\n  async _generate(\n    _messages: BaseMessage[],\n    _options: this[\"ParsedCallOptions\"],\n    _runManager?: CallbackManagerForLLMRun\n  ): Promise<ChatResult> {\n    throw new Error(\"Not implemented\");\n  }\n\n  async _model() {\n    return this._chatModel;\n  }\n\n  bindTools(tools: BindToolsInput[]) {\n    const modelWithTools = new FakeConfigurableModel({\n      model: (this._chatModel as FakeToolCallingChatModel).bindTools(tools),\n    });\n    modelWithTools._queuedMethodOperations.bindTools = tools;\n    return modelWithTools;\n  }\n}\n\nexport class FakeToolCallingChatModel extends BaseChatModel {\n  sleep?: number = 50;\n\n  responses?: BaseMessage[];\n\n  thrownErrorString?: string;\n\n  idx: number;\n\n  toolStyle: \"openai\" | \"anthropic\" | \"bedrock\" | \"google\" = \"openai\";\n\n  structuredResponse?: Record<string, unknown>;\n\n  // Track messages passed to structured output calls\n  structuredOutputMessages: BaseMessage[][] = [];\n\n  constructor(\n    fields: {\n      sleep?: number;\n      responses?: BaseMessage[];\n      thrownErrorString?: string;\n      toolStyle?: \"openai\" | \"anthropic\" | \"bedrock\" | \"google\";\n      structuredResponse?: Record<string, unknown>;\n    } & BaseChatModelParams\n  ) {\n    super(fields);\n    this.sleep = fields.sleep ?? this.sleep;\n    this.responses = fields.responses;\n    this.thrownErrorString = fields.thrownErrorString;\n    this.idx = 0;\n    this.toolStyle = fields.toolStyle ?? this.toolStyle;\n    this.structuredResponse = fields.structuredResponse;\n    this.structuredOutputMessages = [];\n  }\n\n  _llmType() {\n    return \"fake\";\n  }\n\n  async _generate(\n    messages: BaseMessage[],\n    _options: this[\"ParsedCallOptions\"],\n    runManager?: CallbackManagerForLLMRun\n  ): Promise<ChatResult> {\n    if (this.thrownErrorString) {\n      throw new Error(this.thrownErrorString);\n    }\n    if (this.sleep !== undefined) {\n      await new Promise((resolve) => setTimeout(resolve, this.sleep));\n    }\n    const responses = this.responses?.length ? this.responses : messages;\n    const msg = responses[this.idx % responses.length];\n    const generation: ChatResult = {\n      generations: [\n        {\n          text: \"\",\n          message: msg,\n        },\n      ],\n    };\n    this.idx += 1;\n\n    if (typeof msg.content === \"string\") {\n      await runManager?.handleLLMNewToken(msg.content);\n    }\n    return generation;\n  }\n\n  bindTools(tools: BindToolsInput[]): Runnable<any> {\n    const toolDicts = [];\n    const serverTools = [];\n    for (const tool of tools) {\n      if (!(\"name\" in tool)) {\n        serverTools.push(tool);\n        continue;\n      }\n\n      // NOTE: this is a simplified tool spec for testing purposes only\n      if (this.toolStyle === \"openai\") {\n        toolDicts.push({\n          type: \"function\",\n          function: {\n            name: tool.name,\n          },\n        });\n      } else if ([\"anthropic\", \"google\"].includes(this.toolStyle)) {\n        toolDicts.push({\n          name: tool.name,\n        });\n      } else if (this.toolStyle === \"bedrock\") {\n        toolDicts.push({\n          toolSpec: {\n            name: tool.name,\n          },\n        });\n      }\n    }\n    let toolsToBind: BindToolsInput[] = toolDicts;\n    if (this.toolStyle === \"google\") {\n      toolsToBind = [{ functionDeclarations: toolDicts }];\n    }\n    return this.withConfig({\n      tools: [...toolsToBind, ...serverTools],\n    } as BaseChatModelCallOptions);\n  }\n\n  withStructuredOutput<\n    RunOutput extends Record<string, any> = Record<string, any>\n  >(_: unknown): Runnable<any> {\n    if (!this.structuredResponse) {\n      throw new Error(\"No structured response provided\");\n    }\n    // Create a runnable that returns the proper structured format\n    return RunnableLambda.from(async (messages: BaseMessage[]) => {\n      if (this.sleep) {\n        await new Promise((resolve) => setTimeout(resolve, this.sleep));\n      }\n\n      // Store the messages that were sent to generate structured output\n      this.structuredOutputMessages.push([...messages]);\n\n      // Return in the format expected: { raw: BaseMessage, parsed: RunOutput }\n      return this.structuredResponse as RunOutput;\n    });\n  }\n}\n\nexport class MemorySaverAssertImmutable extends MemorySaver {\n  storageForCopies: Record<string, Record<string, Uint8Array>> = {};\n\n  constructor() {\n    super();\n    this.storageForCopies = {};\n  }\n\n  async put(\n    config: RunnableConfig,\n    checkpoint: Checkpoint,\n    metadata: CheckpointMetadata\n  ): Promise<RunnableConfig> {\n    const thread_id = config.configurable?.thread_id;\n    this.storageForCopies[thread_id] ??= {};\n\n    // assert checkpoint hasn't been modified since last written\n    const saved = await this.get(config);\n    if (saved) {\n      const savedId = saved.id;\n      if (this.storageForCopies[thread_id][savedId]) {\n        const loaded = await this.serde.loadsTyped(\n          \"json\",\n          this.storageForCopies[thread_id][savedId]\n        );\n\n        expect(\n          saved,\n          `Checkpoint [${savedId}] has been modified since last written`\n        ).toEqual(loaded);\n      }\n    }\n    const [, serializedCheckpoint] = await this.serde.dumpsTyped(checkpoint);\n    // save a copy of the checkpoint\n    this.storageForCopies[thread_id][checkpoint.id] = serializedCheckpoint;\n\n    return super.put(config, checkpoint, metadata);\n  }\n}\n\ninterface ToolCall {\n  name: string;\n  args: Record<string, any>;\n  id: string;\n  type?: \"tool_call\";\n}\n\ninterface FakeToolCallingModelFields {\n  toolCalls?: ToolCall[][];\n  toolStyle?: \"openai\" | \"anthropic\";\n  index?: number;\n  structuredResponse?: any;\n}\n\n// Helper function to create checkpointer\nexport function createCheckpointer(): BaseCheckpointSaver {\n  return new MemorySaver();\n}\n\n/**\n * Fake chat model for testing tool calling functionality\n */\nexport class FakeToolCallingModel extends BaseChatModel {\n  toolCalls: ToolCall[][];\n\n  toolStyle: \"openai\" | \"anthropic\";\n\n  // Use a shared reference object so the index persists across bindTools calls\n  private indexRef: { current: number };\n\n  structuredResponse?: any;\n\n  private tools: StructuredTool[] = [];\n\n  constructor({\n    toolCalls = [],\n    toolStyle = \"openai\",\n    index = 0,\n    structuredResponse,\n    indexRef,\n    ...rest\n  }: FakeToolCallingModelFields & { indexRef?: { current: number } } = {}) {\n    super(rest);\n    this.toolCalls = toolCalls;\n    this.toolStyle = toolStyle;\n    // Share the same index reference across instances\n    this.indexRef = indexRef ?? { current: index };\n    this.structuredResponse = structuredResponse;\n  }\n\n  // Getter/setter for backwards compatibility\n  get index(): number {\n    return this.indexRef.current;\n  }\n\n  set index(value: number) {\n    this.indexRef.current = value;\n  }\n\n  _llmType(): string {\n    return \"fake-tool-calling\";\n  }\n\n  _combineLLMOutput() {\n    return [];\n  }\n\n  bindTools(\n    tools: StructuredTool[]\n  ):\n    | FakeToolCallingModel\n    | RunnableBinding<\n        any,\n        any,\n        any & { tool_choice?: ToolChoice | undefined }\n      > {\n    const newInstance = new FakeToolCallingModel({\n      toolCalls: this.toolCalls,\n      toolStyle: this.toolStyle,\n      structuredResponse: this.structuredResponse,\n      // Pass the same indexRef so all instances share the same index state\n      indexRef: this.indexRef,\n    });\n    newInstance.tools = [...this.tools, ...tools];\n    return newInstance;\n  }\n\n  withStructuredOutput(_schema: any) {\n    return new RunnableLambda({\n      func: async () => {\n        return this.structuredResponse;\n      },\n    });\n  }\n\n  async _generate(\n    messages: BaseMessage[],\n    _options?: this[\"ParsedCallOptions\"],\n    _runManager?: CallbackManagerForLLMRun\n  ): Promise<ChatResult> {\n    const lastMessage = messages[messages.length - 1];\n    let content = lastMessage.content as string;\n\n    // Handle prompt concatenation\n    if (messages.length > 1) {\n      const parts = messages.map((m) => m.content).filter(Boolean);\n      content = parts.join(\"-\");\n    }\n\n    // Reset index at the start of a new conversation (only human message)\n    // This allows the model to be reused across multiple agent.invoke() calls\n    const isStartOfConversation =\n      messages.length === 1 ||\n      (messages.length === 2 && messages.every(HumanMessage.isInstance));\n    if (isStartOfConversation && this.index !== 0) {\n      this.index = 0;\n    }\n\n    const currentToolCalls = this.toolCalls[this.index] || [];\n    const messageId = this.index.toString();\n\n    // Move to next set of tool calls for subsequent invocations\n    this.index = (this.index + 1) % Math.max(1, this.toolCalls.length);\n\n    const message = new AIMessage({\n      content,\n      id: messageId,\n      tool_calls:\n        currentToolCalls.length > 0\n          ? currentToolCalls.map((tc) => ({\n              ...tc,\n              type: \"tool_call\" as const,\n            }))\n          : undefined,\n    });\n\n    return {\n      generations: [\n        {\n          text: content,\n          message,\n        },\n      ],\n      llmOutput: {},\n    };\n  }\n}\n\nexport class SearchAPI extends StructuredTool {\n  name = \"search_api\";\n\n  description = \"A simple API that returns the input string.\";\n\n  schema = z.object({\n    query: z.string().describe(\"The query to search for.\"),\n  });\n\n  async _call(input: z.infer<typeof this.schema>) {\n    if (input?.query === \"error\") {\n      throw new Error(\"Error\");\n    }\n    return `result for ${input?.query}`;\n  }\n}\n"],"mappings":";;;;;;;;;;;AAuUA,IAAa,uBAAb,MAAa,6BAA6B,cAAc;CACtD;CAEA;CAGA,AAAQ;CAER;CAEA,AAAQ,QAA0B,CAAE;CAEpC,YAAY,EACV,YAAY,CAAE,GACd,YAAY,UACZ,QAAQ,GACR,oBACA,SACA,GAAG,MAC6D,GAAG,CAAE,GAAE;EACvE,MAAM,KAAK;EACX,KAAK,YAAY;EACjB,KAAK,YAAY;EAEjB,KAAK,WAAW,YAAY,EAAE,SAAS,MAAO;EAC9C,KAAK,qBAAqB;CAC3B;CAGD,IAAI,QAAgB;AAClB,SAAO,KAAK,SAAS;CACtB;CAED,IAAI,MAAMA,OAAe;EACvB,KAAK,SAAS,UAAU;CACzB;CAED,WAAmB;AACjB,SAAO;CACR;CAED,oBAAoB;AAClB,SAAO,CAAE;CACV;CAED,UACEC,OAOI;EACJ,MAAM,cAAc,IAAI,qBAAqB;GAC3C,WAAW,KAAK;GAChB,WAAW,KAAK;GAChB,oBAAoB,KAAK;GAEzB,UAAU,KAAK;EAChB;EACD,YAAY,QAAQ,CAAC,GAAG,KAAK,OAAO,GAAG,KAAM;AAC7C,SAAO;CACR;CAED,qBAAqBC,SAAc;AACjC,SAAO,IAAI,eAAe,EACxB,MAAM,YAAY;AAChB,UAAO,KAAK;EACb,EACF;CACF;CAED,MAAM,UACJC,UACAC,UACAC,aACqB;EACrB,MAAM,cAAc,SAAS,SAAS,SAAS;EAC/C,IAAI,UAAU,YAAY;AAG1B,MAAI,SAAS,SAAS,GAAG;GACvB,MAAM,QAAQ,SAAS,IAAI,CAAC,MAAM,EAAE,QAAQ,CAAC,OAAO,QAAQ;GAC5D,UAAU,MAAM,KAAK,IAAI;EAC1B;EAID,MAAM,wBACJ,SAAS,WAAW,KACnB,SAAS,WAAW,KAAK,SAAS,MAAM,aAAa,WAAW;AACnE,MAAI,yBAAyB,KAAK,UAAU,GAC1C,KAAK,QAAQ;EAGf,MAAM,mBAAmB,KAAK,UAAU,KAAK,UAAU,CAAE;EACzD,MAAM,YAAY,KAAK,MAAM,UAAU;EAGvC,KAAK,SAAS,KAAK,QAAQ,KAAK,KAAK,IAAI,GAAG,KAAK,UAAU,OAAO;EAElE,MAAM,UAAU,IAAI,UAAU;GAC5B;GACA,IAAI;GACJ,YACE,iBAAiB,SAAS,IACtB,iBAAiB,IAAI,CAAC,QAAQ;IAC5B,GAAG;IACH,MAAM;GACP,GAAE,GACH;EACP;AAED,SAAO;GACL,aAAa,CACX;IACE,MAAM;IACN;GACD,CACF;GACD,WAAW,CAAE;EACd;CACF;AACF"}

package/dist/agents/tools.d.cts

@@ -0,0 +1,9 @@
+import { RunnableToolLike } from "@langchain/core/runnables";
+import { DynamicTool, StructuredToolInterface } from "@langchain/core/tools";
+
+//#region src/agents/tools.d.ts
+type ServerTool = Record<string, unknown>;
+type ClientTool = StructuredToolInterface | DynamicTool | RunnableToolLike;
+//#endregion
+export { ClientTool, ServerTool };
+//# sourceMappingURL=tools.d.cts.map

package/dist/agents/tools.d.cts.map

@@ -0,0 +1 @@
+{"version":3,"file":"tools.d.cts","names":["StructuredToolInterface","DynamicTool","RunnableToolLike","ServerTool","Record","ClientTool"],"sources":["../../src/agents/tools.d.ts"],"sourcesContent":["import type { StructuredToolInterface, DynamicTool } from \"@langchain/core/tools\";\nimport type { RunnableToolLike } from \"@langchain/core/runnables\";\nexport type ServerTool = Record<string, unknown>;\nexport type ClientTool = StructuredToolInterface | DynamicTool | RunnableToolLike;\n"],"mappings":";;;;KAEYG,UAAAA,GAAaC;KACbC,UAAAA,GAAaL,0BAA0BC,cAAcC"}

package/dist/agents/tools.d.ts

@@ -0,0 +1,9 @@
+import { RunnableToolLike } from "@langchain/core/runnables";
+import { DynamicTool, StructuredToolInterface } from "@langchain/core/tools";
+
+//#region src/agents/tools.d.ts
+type ServerTool = Record<string, unknown>;
+type ClientTool = StructuredToolInterface | DynamicTool | RunnableToolLike;
+//#endregion
+export { ClientTool, ServerTool };
+//# sourceMappingURL=tools.d.ts.map

package/dist/agents/tools.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"tools.d.ts","names":["StructuredToolInterface","DynamicTool","RunnableToolLike","ServerTool","Record","ClientTool"],"sources":["../../src/agents/tools.d.ts"],"sourcesContent":["import type { StructuredToolInterface, DynamicTool } from \"@langchain/core/tools\";\nimport type { RunnableToolLike } from \"@langchain/core/runnables\";\nexport type ServerTool = Record<string, unknown>;\nexport type ClientTool = StructuredToolInterface | DynamicTool | RunnableToolLike;\n"],"mappings":";;;;KAEYG,UAAAA,GAAaC;KACbC,UAAAA,GAAaL,0BAA0BC,cAAcC"}

package/dist/agents/types.d.cts

@@ -1,21 +1,56 @@
-import { AnyAnnotationRoot, PreHookAnnotation, ResponseFormatUndefined, ToAnnotationRoot } from "./annotation.cjs";
-import { JsonSchemaFormat, ProviderStrategy, ResponseFormat, ToolStrategy, TypedToolStrategy } from "./responses.cjs";
-import { ToolNode } from "./nodes/ToolNode.cjs";
+import { JsonSchemaFormat, ProviderStrategy, ResponseFormat, ResponseFormatUndefined, ToolStrategy, TypedToolStrategy } from "./responses.cjs";
+import { JumpToTarget } from "./constants.cjs";
+import { ClientTool, ServerTool } from "./tools.cjs";
+import { AgentMiddleware, AnyAnnotationRoot, InferSchemaInput } from "./middleware/types.cjs";
 import { LanguageModelLike } from "@langchain/core/language_models/base";
+import { BaseMessage } from "@langchain/core/messages";
 import { InteropZodObject, InteropZodType } from "@langchain/core/utils/types";
-import { BaseMessage, BaseMessageLike, SystemMessage } from "@langchain/core/messages";
-import { LangGraphRunnableConfig, START, StateGraph } from "@langchain/langgraph";
-import { All, BaseCheckpointSaver, BaseStore } from "@langchain/langgraph-checkpoint";
-import { Runnable, RunnableLike, RunnableToolLike } from "@langchain/core/runnables";
-import { DynamicTool, StructuredToolInterface } from "@langchain/core/tools";
+import { END, START, StateGraph } from "@langchain/langgraph";
+import { BaseCheckpointSaver, BaseStore } from "@langchain/langgraph-checkpoint";
+import { Messages } from "@langchain/langgraph/";
 
 //#region src/agents/types.d.ts
-declare const META_EXTRAS_DESCRIPTION_PREFIX = "lg:";
-type N = typeof START | "agent" | "tools";
+type N = typeof START | "model_request" | "tools";
+/**
+ * Represents information about an interrupt.
+ */
+interface Interrupt<TValue = unknown> {
+  /**
+   * The ID of the interrupt.
+   */
+  id: string;
+  /**
+   * The requests for human input.
+   */
+  value: TValue;
+}
+interface BuiltInState {
+  messages: BaseMessage[];
+  __interrupt__?: Interrupt[];
+  /**
+   * Optional property to control routing after afterModel middleware execution.
+   * When set by middleware, the agent will jump to the specified node instead of
+   * following normal routing logic. The property is automatically cleared after use.
+   *
+   * - "model_request": Jump back to the model for another LLM call
+   * - "tools": Jump to tool execution (requires tools to be available)
+   */
+  jumpTo?: JumpToTarget;
+}
+/**
+ * Base input type for `.invoke` and `.stream` methods.
+ */
+type UserInput<TStateSchema extends AnyAnnotationRoot | InteropZodObject | undefined = undefined> = InferSchemaInput<TStateSchema> & {
+  messages: Messages;
+};
 /**
  * Information about a tool call that has been executed.
  */
-interface ExecutedToolCall {
+interface ToolCall$1 {
+  /**
+   * The ID of the tool call.
+   */
+  id: string;
   /**
    * The name of the tool that was called.
    */
@@ -23,61 +58,62 @@ interface ExecutedToolCall {
   /**
    * The arguments that were passed to the tool.
    */
-  args: Record<string, unknown>;
+  args: Record<string, any>;
   /**
-   * The ID of the tool call.
+   * The result of the tool call.
    */
-  tool_id: string;
+  result?: unknown;
   /**
-   * The result of the tool call (if available).
+   * An optional error message if the tool call failed.
    */
-  result?: unknown;
+  error?: string;
 }
 /**
- * Type helper to extract the inferred type from a single Zod schema or array of schemas
+ * Information about a tool result from a tool execution.
  */
-type ExtractZodType<T> = T extends InteropZodType<infer U> ? U : T extends readonly InteropZodType<any>[] ? ExtractZodArrayTypes<T> : never;
+interface ToolResult {
+  /**
+   * The ID of the tool call.
+   */
+  id: string;
+  /**
+   * The result of the tool call.
+   */
+  result: any;
+  /**
+   * An optional error message if the tool call failed.
+   */
+  error?: string;
+}
 /**
- * Type helper to extract union type from an array of Zod schemas
+ * jump targets (internal)
  */
-type ExtractZodArrayTypes<T extends readonly InteropZodType<any>[]> = T extends readonly [InteropZodType<infer A>, ...infer Rest] ? Rest extends readonly InteropZodType<any>[] ? A | ExtractZodArrayTypes<Rest> : A : never;
+type JumpTo = "model_request" | "tools" | typeof END;
 /**
- * Type helper to extract the structured response type from responseFormat
+ * Information about a tool call that has been executed.
  */
-type InferResponseFormatType<T> = T extends InteropZodType<infer U> ? U extends Record<string, any> ? U : Record<string, any> : T extends readonly InteropZodType<any>[] ? ExtractZodArrayTypes<T> : T extends ToolStrategy[] ? Record<string, any> // ToolStrategy arrays will be handled at runtime
-: T extends ResponseFormat ? Record<string, any> // Single ResponseFormat will be handled at runtime
-: Record<string, any>;
-/** @internal */
-type ReducedZodChannel<T extends InteropZodType, TReducerSchema extends InteropZodType> = T & {
-  lg_reducer_schema: TReducerSchema;
-};
-type ServerTool = Record<string, unknown>;
-type ClientTool = StructuredToolInterface | DynamicTool | RunnableToolLike;
-type Prompt<StateSchema extends AnyAnnotationRoot | InteropZodObject = AnyAnnotationRoot, ContextSchema extends AnyAnnotationRoot | InteropZodObject = AnyAnnotationRoot> = SystemMessage | string | ((state: ToAnnotationRoot<StateSchema>["State"] & PreHookAnnotation["State"], config: LangGraphRunnableConfig<ToAnnotationRoot<ContextSchema>["State"]>) => BaseMessageLike[] | Promise<BaseMessageLike[]>) | Runnable;
-type AgentState<AnnotationRoot extends AnyAnnotationRoot | InteropZodObject = AnyAnnotationRoot> = ToAnnotationRoot<AnnotationRoot>["State"] & PreHookAnnotation["State"];
-interface AgentRuntime<ContextType = Record<string, unknown>> {
+interface ExecutedToolCall {
   /**
-   * The context of the agent.
+   * The name of the tool that was called.
    */
-  context?: ContextType;
+  name: string;
   /**
-   * The store passed to the agent.
+   * The arguments that were passed to the tool.
    */
-  store?: BaseStore;
+  args: Record<string, unknown>;
   /**
-   * The writer of the agent to write to the output stream.
+   * The ID of the tool call.
    */
-  writer?: (chunk: unknown) => void;
+  tool_id: string;
   /**
-   * An optional abort signal that indicates that the overall operation should be aborted.
+   * The result of the tool call (if available).
    */
-  signal?: AbortSignal;
+  result?: unknown;
 }
-type CreateAgentParams<StateSchema extends AnyAnnotationRoot | InteropZodObject = AnyAnnotationRoot, StructuredResponseType extends Record<string, unknown> = Record<string, unknown>, ContextSchema extends AnyAnnotationRoot | InteropZodObject = AnyAnnotationRoot, ResponseFormatType = InteropZodType<StructuredResponseType> | InteropZodType<unknown>[] | JsonSchemaFormat | JsonSchemaFormat[] | ResponseFormat | TypedToolStrategy<StructuredResponseType> | ToolStrategy<StructuredResponseType> | ProviderStrategy<StructuredResponseType> | ResponseFormatUndefined> = {
-  /** The chat model that can utilize OpenAI-style tool calling. */
-  llm?: LanguageModelLike | DynamicLLMFunction<StateSchema, ContextSchema>;
+type CreateAgentParams<StructuredResponseType extends Record<string, any> = Record<string, any>, StateSchema extends AnyAnnotationRoot | InteropZodObject | undefined = undefined, ContextSchema extends AnyAnnotationRoot | InteropZodObject = AnyAnnotationRoot, ResponseFormatType = InteropZodType<StructuredResponseType> | InteropZodType<unknown>[] | JsonSchemaFormat | JsonSchemaFormat[] | ResponseFormat | TypedToolStrategy<StructuredResponseType> | ToolStrategy<StructuredResponseType> | ProviderStrategy<StructuredResponseType> | ResponseFormatUndefined> = {
   /**
-   * Initializes a ChatModel based on the provided model name and provider.
+   * Defines a model to use for the agent. You can either pass in an instance of a LangChain chat model
+   * or a string. If a string is provided the agent initializes a ChatModel based on the provided model name and provider.
    * It supports various model providers and allows for runtime configuration of model parameters.
    *
    * @uses {@link initChatModel}
@@ -88,76 +124,91 @@ type CreateAgentParams<StateSchema extends AnyAnnotationRoot | InteropZodObject
    *   // ...
    * });
    * ```
+   *
+   * @example
+   * ```ts
+   * import { ChatOpenAI } from "@langchain/openai";
+   * const agent = createAgent({
+   *   model: new ChatOpenAI({ model: "gpt-4o" }),
+   *   // ...
+   * });
+   * ```
    */
-  model?: string;
-  /** A list of tools or a ToolNode. */
-  tools: ToolNode | (ServerTool | ClientTool)[];
+  model: string | LanguageModelLike;
   /**
-   * An optional prompt for the LLM. This takes full graph state BEFORE the LLM is called and prepares the input to LLM.
-   *
-   * Can take a few different forms:
+   * A list of tools or a ToolNode.
    *
-   * - str: This is converted to a SystemMessage and added to the beginning of the list of messages in state["messages"].
-   * - SystemMessage: this is added to the beginning of the list of messages in state["messages"].
-   * - Function: This function should take in full graph state and the output is then passed to the language model.
-   * - Runnable: This runnable should take in full graph state and the output is then passed to the language model.
+   * @example
+   * ```ts
+   * import { tool } from "langchain";
    *
-   * Note:
-   * Prior to `v0.2.46`, the prompt was set using `stateModifier` / `messagesModifier` parameters.
-   * This is now deprecated and will be removed in a future release.
+   * const weatherTool = tool(() => "Sunny!", {
+   *   name: "get_weather",
+   *   description: "Get the weather for a location",
+   *   schema: z.object({
+   *     location: z.string().describe("The location to get weather for"),
+   *   }),
+   * });
    *
-   * Cannot be used together with `prepareCall`.
+   * const agent = createAgent({
+   *   tools: [weatherTool],
+   *   // ...
+   * });
+   * ```
+   */
+  tools?: (ServerTool | ClientTool)[];
+  /**
+   * An optional system message for the model.
    */
-  prompt?: Prompt<StateSchema, ContextSchema>;
+  systemPrompt?: string;
   /**
-   * Additional state schema for the agent. It allows to define additional state keys that will be
-   * persisted between agent invocations.
+   * An optional schema for the agent state. It allows you to define custom state properties that persist
+   * across agent invocations and can be accessed in hooks, middleware, and throughout the agent's execution.
+   * The state is persisted when using a checkpointer and can be updated by middleware or during execution.
+   *
+   * As opposed to the context (defined in `contextSchema`), the state is persisted between agent invocations
+   * when using a checkpointer, making it suitable for maintaining conversation history, user preferences,
+   * or any other data that should persist across multiple interactions.
    *
    * @example
    * ```ts
-   * // State schema defines data that persists across agent invocations
-   * const stateSchema = z.object({
-   *   userPreferences: z.object({
-   *     theme: z.enum(["light", "dark"]),
-   *     language: z.string(),
-   *   }),
-   *   taskHistory: z.array(z.string()),
-   *   currentWorkflow: z.string().optional(),
-   * });
-   *
-   * // Context schema defines runtime parameters passed per invocation
-   * const contextSchema = z.object({ ... });
+   * import { z } from "zod";
+   * import { createAgent } from "@langchain/langgraph";
    *
    * const agent = createAgent({
-   *   llm: model,
-   *   tools: [updatePreferences, addTask],
-   *   stateSchema,    // Persisted: preferences, e.g. task history, workflow state
-   *   contextSchema,  // Per-invocation: user ID, session, API keys, etc.
+   *   model: "openai:gpt-4o",
+   *   tools: [getWeather],
+   *   stateSchema: z.object({
+   *     userPreferences: z.object({
+   *       temperatureUnit: z.enum(["celsius", "fahrenheit"]).default("celsius"),
+   *       location: z.string().optional(),
+   *     }).optional(),
+   *     conversationCount: z.number().default(0),
+   *   }),
    *   prompt: (state, config) => {
-   *     // ...
+   *     const unit = state.userPreferences?.temperatureUnit || "celsius";
+   *     return [
+   *       new SystemMessage(`You are a helpful assistant. Use ${unit} for temperature.`),
+   *     ];
    *   },
    * });
    *
-   * // First invocation - state starts empty, context provided
-   * await agent.invoke({
-   *   messages: [new HumanMessage("Set my theme to dark")],
-   * }, {
-   *   context: { userId: "user123", sessionId: "sess456", apiKeys: {...} }
-   * });
-   *
-   * // Second invocation - state persists, new context
-   * await agent.invoke({
-   *   messages: [new HumanMessage("Add a task to review code")],
-   * }, {
-   *   context: { userId: "user123", sessionId: "sess789", apiKeys: {...} }
+   * const result = await agent.invoke({
+   *   messages: [
+   *     new HumanMessage("What's the weather like?"),
+   *   ],
+   *   userPreferences: {
+   *     temperatureUnit: "fahrenheit",
+   *     location: "New York",
+   *   },
+   *   conversationCount: 1,
    * });
-   * // State now contains: userPreferences.theme="dark", taskHistory=["review code"]
    * ```
    */
   stateSchema?: StateSchema;
   /**
    * An optional schema for the context. It allows to pass in a typed context object into the agent
-   * invocation and allows to access it in hooks such as `prompt`, `preModelHook`, `postModelHook`, etc.
+   * invocation and allows to access it in hooks such as `prompt` and middleware.
    * As opposed to the agent state, defined in `stateSchema`, the context is not persisted between
    * agent invocations.
    *
@@ -189,14 +240,15 @@ type CreateAgentParams<StateSchema extends AnyAnnotationRoot | InteropZodObject
    * ```
    */
   contextSchema?: ContextSchema;
-  /** An optional checkpoint saver to persist the agent's state. */
-  checkpointSaver?: BaseCheckpointSaver | boolean;
-  /** An optional checkpoint saver to persist the agent's state. Alias of "checkpointSaver". */
+  /**
+   * An optional checkpoint saver to persist the agent's state.
+   * @see {@link https://docs.langchain.com/oss/javascript/langgraph/persistence | Checkpointing}
+   */
   checkpointer?: BaseCheckpointSaver | boolean;
-  /** An optional list of node names to interrupt before running. */
-  interruptBefore?: N[] | All;
-  /** An optional list of node names to interrupt after running. */
-  interruptAfter?: N[] | All;
+  /**
+   * An optional store to persist the agent's state.
+   * @see {@link https://docs.langchain.com/oss/javascript/langgraph/memory#memory-storage | Long-term memory}
+   */
   store?: BaseStore;
   /**
    * An optional schema for the final agent output.
@@ -252,6 +304,13 @@ type CreateAgentParams<StateSchema extends AnyAnnotationRoot | InteropZodObject
    * This is not the only strategy to get structured responses, see more options in [this guide](https://langchain-ai.github.io/langgraph/how-tos/react-agent-structured-output/).
    */
   responseFormat?: ResponseFormatType;
+  /**
+   * Middleware instances to run during agent execution.
+   * Each middleware can define its own state schema and hook into the agent lifecycle.
+   *
+   * @see {@link https://docs.langchain.com/oss/javascript/langchain/middleware | Middleware}
+   */
+  middleware?: readonly AgentMiddleware<any, any, any>[];
   /**
    * An optional name for the agent.
    */
@@ -268,16 +327,6 @@ type CreateAgentParams<StateSchema extends AnyAnnotationRoot | InteropZodObject
    *       Example: `"How can I help you"` -> `"<name>agent_name</name><content>How can I help you?</content>"`
    */
   includeAgentName?: "inline" | undefined;
-  /**
-   * An optional node to add before the `agent` node (i.e., the node that calls the LLM).
-   * Useful for managing long message histories (e.g., message trimming, summarization, etc.).
-   */
-  preModelHook?: RunnableLike<ToAnnotationRoot<StateSchema>["State"] & PreHookAnnotation["State"], ToAnnotationRoot<StateSchema>["Update"] & PreHookAnnotation["Update"], LangGraphRunnableConfig<ToAnnotationRoot<ContextSchema>["State"]>>;
-  /**
-   * An optional node to add after the `agent` node (i.e., the node that calls the LLM).
-   * Useful for implementing human-in-the-loop, guardrails, validation, or other post-processing.
-   */
-  postModelHook?: RunnableLike<ToAnnotationRoot<StateSchema>["State"] & PreHookAnnotation["State"], ToAnnotationRoot<StateSchema>["Update"] & PreHookAnnotation["Update"], LangGraphRunnableConfig<ToAnnotationRoot<ContextSchema>["State"]>>;
   /**
    * An optional abort signal that indicates that the overall operation should be aborted.
    */
@@ -295,16 +344,11 @@ type CreateAgentParams<StateSchema extends AnyAnnotationRoot | InteropZodObject
    */
   version?: "v1" | "v2";
 };
-type InternalAgentState<StructuredResponseType extends Record<string, unknown> | undefined = Record<string, unknown>> = {
-  messages: BaseMessage[];
-} & (StructuredResponseType extends ResponseFormatUndefined ? Record<string, never> : {
-  structuredResponse: StructuredResponseType;
-});
-type WithStateGraphNodes<K extends string, Graph> = Graph extends StateGraph<infer SD, infer S, infer U, infer N, infer I, infer O, infer C> ? StateGraph<SD, S, U, N | K, I, O, C> : never;
 /**
- * @deprecated likely to be removed in the next version of the agent
+ * Type helper to extract union type from an array of Zod schemas
  */
-type DynamicLLMFunction<StateSchema extends AnyAnnotationRoot | InteropZodObject = AnyAnnotationRoot, ContextSchema extends AnyAnnotationRoot | InteropZodObject = AnyAnnotationRoot> = (state: ToAnnotationRoot<StateSchema>["State"] & PreHookAnnotation["State"], runtime: AgentRuntime<ToAnnotationRoot<ContextSchema>["State"]>) => Promise<LanguageModelLike> | LanguageModelLike;
+type ExtractZodArrayTypes<T extends readonly InteropZodType<any>[]> = T extends readonly [InteropZodType<infer A>, ...infer Rest] ? Rest extends readonly InteropZodType<any>[] ? A | ExtractZodArrayTypes<Rest> : A : never;
+type WithStateGraphNodes<K extends string, Graph> = Graph extends StateGraph<infer SD, infer S, infer U, infer N, infer I, infer O, infer C> ? StateGraph<SD, S, U, N | K, I, O, C> : never;
 //#endregion
-export { AgentRuntime, AgentState, ClientTool, CreateAgentParams, ExecutedToolCall, ExtractZodArrayTypes, ExtractZodType, InferResponseFormatType, InternalAgentState, META_EXTRAS_DESCRIPTION_PREFIX, N, Prompt, ReducedZodChannel, ServerTool, WithStateGraphNodes };
+export { BuiltInState, CreateAgentParams, ExecutedToolCall, ExtractZodArrayTypes, Interrupt, JumpTo, N, ToolCall$1 as ToolCall, ToolResult, UserInput, WithStateGraphNodes };
 //# sourceMappingURL=types.d.cts.map