@langchain/classic 1.0.0-alpha.1

package/dist/evaluation/types.d.cts

@@ -0,0 +1,40 @@
+import { AgentTrajectoryEvaluator, LLMPairwiseStringEvaluator, LLMStringEvaluator, PairwiseStringEvaluator, StringEvaluator } from "./base.cjs";
+
+//#region src/evaluation/types.d.ts
+interface EvaluatorType {
+  /**
+   * The criteria evaluator, which evaluates a model based on a
+   * custom set of criteria without any reference labels.
+   * */
+  criteria: LLMStringEvaluator;
+  /**
+   * The labeled criteria evaluator, which evaluates a model based on a
+   * custom set of criteria, with a reference label.
+   * */
+  labeled_criteria: LLMStringEvaluator;
+  /**
+   *  The pairwise string evaluator, which predicts the preferred prediction from
+   *  between two models.
+   */
+  pairwise_string: LLMPairwiseStringEvaluator;
+  /**
+   * The labeled pairwise string evaluator, which predicts the preferred prediction
+   * from between two models based on a ground truth reference label.
+   * */
+  labeled_pairwise_string: LLMPairwiseStringEvaluator;
+  /**
+   * The agent trajectory evaluator, which grades the agent's intermediate steps.
+   */
+  trajectory: AgentTrajectoryEvaluator;
+  /**
+   * Compare a prediction to a reference label using embedding distance.
+   * */
+  embedding_distance: StringEvaluator;
+  /**
+   * Compare two predictions using embedding distance.
+   * */
+  pairwise_embedding_distance: PairwiseStringEvaluator;
+}
+//#endregion
+export { EvaluatorType };
+//# sourceMappingURL=types.d.cts.map

package/dist/evaluation/types.d.cts.map

@@ -0,0 +1 @@
+{"version":3,"file":"types.d.cts","names":["AgentTrajectoryEvaluator","LLMPairwiseStringEvaluator","LLMStringEvaluator","PairwiseStringEvaluator","StringEvaluator","EvaluatorType"],"sources":["../../src/evaluation/types.d.ts"],"sourcesContent":["import { AgentTrajectoryEvaluator, LLMPairwiseStringEvaluator, LLMStringEvaluator, PairwiseStringEvaluator, StringEvaluator } from \"./base.js\";\nexport interface EvaluatorType {\n    /**\n     * The criteria evaluator, which evaluates a model based on a\n     * custom set of criteria without any reference labels.\n     * */\n    criteria: LLMStringEvaluator;\n    /**\n     * The labeled criteria evaluator, which evaluates a model based on a\n     * custom set of criteria, with a reference label.\n     * */\n    labeled_criteria: LLMStringEvaluator;\n    /**\n     *  The pairwise string evaluator, which predicts the preferred prediction from\n     *  between two models.\n     */\n    pairwise_string: LLMPairwiseStringEvaluator;\n    /**\n     * The labeled pairwise string evaluator, which predicts the preferred prediction\n     * from between two models based on a ground truth reference label.\n     * */\n    labeled_pairwise_string: LLMPairwiseStringEvaluator;\n    /**\n     * The agent trajectory evaluator, which grades the agent's intermediate steps.\n     */\n    trajectory: AgentTrajectoryEvaluator;\n    /**\n     * Compare a prediction to a reference label using embedding distance.\n     * */\n    embedding_distance: StringEvaluator;\n    /**\n     * Compare two predictions using embedding distance.\n     * */\n    pairwise_embedding_distance: PairwiseStringEvaluator;\n}\n"],"mappings":";;;UACiBK,aAAAA;;AAAjB;;;EAKgC,QAKVH,EALRA,kBAKQA;EAAkB;;;;EAkBD,gBAINC,EAtBXD,kBAsBWC;EAAuB;;;;mBAjBnCF;;;;;2BAKQA;;;;cAIbD;;;;sBAIQI;;;;+BAISD"}

package/dist/evaluation/types.d.ts

@@ -0,0 +1,40 @@
+import { AgentTrajectoryEvaluator, LLMPairwiseStringEvaluator, LLMStringEvaluator, PairwiseStringEvaluator, StringEvaluator } from "./base.js";
+
+//#region src/evaluation/types.d.ts
+interface EvaluatorType {
+  /**
+   * The criteria evaluator, which evaluates a model based on a
+   * custom set of criteria without any reference labels.
+   * */
+  criteria: LLMStringEvaluator;
+  /**
+   * The labeled criteria evaluator, which evaluates a model based on a
+   * custom set of criteria, with a reference label.
+   * */
+  labeled_criteria: LLMStringEvaluator;
+  /**
+   *  The pairwise string evaluator, which predicts the preferred prediction from
+   *  between two models.
+   */
+  pairwise_string: LLMPairwiseStringEvaluator;
+  /**
+   * The labeled pairwise string evaluator, which predicts the preferred prediction
+   * from between two models based on a ground truth reference label.
+   * */
+  labeled_pairwise_string: LLMPairwiseStringEvaluator;
+  /**
+   * The agent trajectory evaluator, which grades the agent's intermediate steps.
+   */
+  trajectory: AgentTrajectoryEvaluator;
+  /**
+   * Compare a prediction to a reference label using embedding distance.
+   * */
+  embedding_distance: StringEvaluator;
+  /**
+   * Compare two predictions using embedding distance.
+   * */
+  pairwise_embedding_distance: PairwiseStringEvaluator;
+}
+//#endregion
+export { EvaluatorType };
+//# sourceMappingURL=types.d.ts.map

package/dist/evaluation/types.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"types.d.ts","names":["AgentTrajectoryEvaluator","LLMPairwiseStringEvaluator","LLMStringEvaluator","PairwiseStringEvaluator","StringEvaluator","EvaluatorType"],"sources":["../../src/evaluation/types.d.ts"],"sourcesContent":["import { AgentTrajectoryEvaluator, LLMPairwiseStringEvaluator, LLMStringEvaluator, PairwiseStringEvaluator, StringEvaluator } from \"./base.js\";\nexport interface EvaluatorType {\n    /**\n     * The criteria evaluator, which evaluates a model based on a\n     * custom set of criteria without any reference labels.\n     * */\n    criteria: LLMStringEvaluator;\n    /**\n     * The labeled criteria evaluator, which evaluates a model based on a\n     * custom set of criteria, with a reference label.\n     * */\n    labeled_criteria: LLMStringEvaluator;\n    /**\n     *  The pairwise string evaluator, which predicts the preferred prediction from\n     *  between two models.\n     */\n    pairwise_string: LLMPairwiseStringEvaluator;\n    /**\n     * The labeled pairwise string evaluator, which predicts the preferred prediction\n     * from between two models based on a ground truth reference label.\n     * */\n    labeled_pairwise_string: LLMPairwiseStringEvaluator;\n    /**\n     * The agent trajectory evaluator, which grades the agent's intermediate steps.\n     */\n    trajectory: AgentTrajectoryEvaluator;\n    /**\n     * Compare a prediction to a reference label using embedding distance.\n     * */\n    embedding_distance: StringEvaluator;\n    /**\n     * Compare two predictions using embedding distance.\n     * */\n    pairwise_embedding_distance: PairwiseStringEvaluator;\n}\n"],"mappings":";;;UACiBK,aAAAA;;AAAjB;;;EAKgC,QAKVH,EALRA,kBAKQA;EAAkB;;;;EAkBD,gBAINC,EAtBXD,kBAsBWC;EAAuB;;;;mBAjBnCF;;;;;2BAKQA;;;;cAIbD;;;;sBAIQI;;;;+BAISD"}

package/dist/experimental/autogpt/agent.cjs

@@ -0,0 +1,152 @@
+const require_rolldown_runtime = require('../../_virtual/rolldown_runtime.cjs');
+const require_text_splitter = require('../../text_splitter.cjs');
+const require_llm_chain = require('../../chains/llm_chain.cjs');
+const require_schema = require('./schema.cjs');
+const require_prompt = require('./prompt.cjs');
+const require_output_parser = require('./output_parser.cjs');
+const __langchain_core_language_models_base = require_rolldown_runtime.__toESM(require("@langchain/core/language_models/base"));
+const __langchain_core_messages = require_rolldown_runtime.__toESM(require("@langchain/core/messages"));
+
+//#region src/experimental/autogpt/agent.ts
+/**
+* Class representing the AutoGPT concept with LangChain primitives. It is
+* designed to be used with a set of tools such as a search tool,
+* write-file tool, and a read-file tool.
+* @example
+* ```typescript
+* const autogpt = AutoGPT.fromLLMAndTools(
+*   new ChatOpenAI({ model: "gpt-4o-mini", temperature: 0 }),
+*   [
+*     new ReadFileTool({ store: new InMemoryFileStore() }),
+*     new WriteFileTool({ store: new InMemoryFileStore() }),
+*     new SerpAPI("YOUR_SERPAPI_API_KEY", {
+*       location: "San Francisco,California,United States",
+*       hl: "en",
+*       gl: "us",
+*     }),
+*   ],
+*   {
+*     memory: new MemoryVectorStore(new OpenAIEmbeddings()).asRetriever(),
+*     aiName: "Tom",
+*     aiRole: "Assistant",
+*   },
+* );
+* const result = await autogpt.run(["write a weather report for SF today"]);
+* ```
+*/
+var AutoGPT = class AutoGPT {
+	aiName;
+	memory;
+	fullMessageHistory;
+	nextActionCount;
+	chain;
+	outputParser;
+	tools;
+	feedbackTool;
+	maxIterations;
+	textSplitter;
+	constructor({ aiName, memory, chain, outputParser, tools, feedbackTool, maxIterations }) {
+		this.aiName = aiName;
+		this.memory = memory;
+		this.fullMessageHistory = [];
+		this.nextActionCount = 0;
+		this.chain = chain;
+		this.outputParser = outputParser;
+		this.tools = tools;
+		this.feedbackTool = feedbackTool;
+		this.maxIterations = maxIterations;
+		const chunkSize = (0, __langchain_core_language_models_base.getEmbeddingContextSize)("modelName" in memory.vectorStore.embeddings ? memory.vectorStore.embeddings.modelName : void 0);
+		this.textSplitter = new require_text_splitter.text_splitter_exports.TokenTextSplitter({
+			chunkSize,
+			chunkOverlap: Math.round(chunkSize / 10)
+		});
+	}
+	/**
+	* Creates a new AutoGPT instance from a given LLM and a set of tools.
+	* @param llm A BaseChatModel object.
+	* @param tools An array of ObjectTool objects.
+	* @param options.aiName The name of the AI.
+	* @param options.aiRole The role of the AI.
+	* @param options.memory A VectorStoreRetriever object that represents the memory of the AI.
+	* @param options.maxIterations The maximum number of iterations the AI can perform.
+	* @param options.outputParser An AutoGPTOutputParser object that parses the output of the AI.
+	* @returns A new instance of the AutoGPT class.
+	*/
+	static fromLLMAndTools(llm, tools, { aiName, aiRole, memory, maxIterations = 100, outputParser = new require_output_parser.AutoGPTOutputParser() }) {
+		const prompt = new require_prompt.AutoGPTPrompt({
+			aiName,
+			aiRole,
+			tools,
+			tokenCounter: llm.getNumTokens.bind(llm),
+			sendTokenLimit: (0, __langchain_core_language_models_base.getModelContextSize)("modelName" in llm ? llm.modelName : "gpt2")
+		});
+		const chain = new require_llm_chain.LLMChain({
+			llm,
+			prompt
+		});
+		return new AutoGPT({
+			aiName,
+			memory,
+			chain,
+			outputParser,
+			tools,
+			maxIterations
+		});
+	}
+	/**
+	* Runs the AI with a given set of goals.
+	* @param goals An array of strings representing the goals.
+	* @returns A string representing the result of the run or undefined if the maximum number of iterations is reached without a result.
+	*/
+	async run(goals) {
+		const user_input = "Determine which next command to use, and respond using the format specified above:";
+		let loopCount = 0;
+		while (loopCount < this.maxIterations) {
+			loopCount += 1;
+			const { text: assistantReply } = await this.chain.call({
+				goals,
+				user_input,
+				memory: this.memory,
+				messages: this.fullMessageHistory
+			});
+			console.log(assistantReply);
+			this.fullMessageHistory.push(new __langchain_core_messages.HumanMessage(user_input));
+			this.fullMessageHistory.push(new __langchain_core_messages.AIMessage(assistantReply));
+			const action = await this.outputParser.parse(assistantReply);
+			const tools = this.tools.reduce((acc, tool) => ({
+				...acc,
+				[tool.name]: tool
+			}), {});
+			if (action.name === require_schema.FINISH_NAME) return action.args.response;
+			let result;
+			if (action.name in tools) {
+				const tool = tools[action.name];
+				let observation;
+				try {
+					observation = await tool.call(action.args);
+				} catch (e) {
+					observation = `Error in args: ${e}`;
+				}
+				result = `Command ${tool.name} returned: ${observation}`;
+			} else if (action.name === "ERROR") result = `Error: ${action.args}. `;
+			else result = `Unknown command '${action.name}'. Please refer to the 'COMMANDS' list for available commands and only respond in the specified JSON format.`;
+			let memoryToAdd = `Assistant Reply: ${assistantReply}\nResult: ${result} `;
+			if (this.feedbackTool) {
+				const feedback = `\n${await this.feedbackTool.call("Input: ")}`;
+				if (feedback === "q" || feedback === "stop") {
+					console.log("EXITING");
+					return "EXITING";
+				}
+				memoryToAdd += feedback;
+			}
+			const documents = await this.textSplitter.createDocuments([memoryToAdd]);
+			await this.memory.addDocuments(documents);
+			this.fullMessageHistory.push(new __langchain_core_messages.SystemMessage(result));
+		}
+		return void 0;
+	}
+};
+
+//#endregion
+exports.AutoGPT = AutoGPT;
+//# sourceMappingURL=agent.cjs.map

package/dist/experimental/autogpt/agent.cjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"agent.cjs","names":["llm: BaseChatModel","tools: ObjectTool[]","AutoGPTOutputParser","AutoGPTPrompt","LLMChain","goals: string[]","HumanMessage","AIMessage","FINISH_NAME","result: string","SystemMessage"],"sources":["../../../src/experimental/autogpt/agent.ts"],"sourcesContent":["import type { VectorStoreRetrieverInterface } from \"@langchain/core/vectorstores\";\nimport { Tool } from \"@langchain/core/tools\";\nimport {\n  AIMessage,\n  BaseMessage,\n  HumanMessage,\n  SystemMessage,\n} from \"@langchain/core/messages\";\nimport { BaseChatModel } from \"@langchain/core/language_models/chat_models\";\nimport {\n  getEmbeddingContextSize,\n  getModelContextSize,\n} from \"@langchain/core/language_models/base\";\nimport { LLMChain } from \"../../chains/llm_chain.js\";\n\nimport { AutoGPTOutputParser } from \"./output_parser.js\";\nimport { AutoGPTPrompt } from \"./prompt.js\";\n// import { HumanInputRun } from \"./tools/human/tool\"; // TODO\nimport { ObjectTool, FINISH_NAME } from \"./schema.js\";\nimport { TokenTextSplitter } from \"../../text_splitter.js\";\n\n/**\n * Interface for the input parameters of the AutoGPT class.\n */\nexport interface AutoGPTInput {\n  aiName: string;\n  aiRole: string;\n  memory: VectorStoreRetrieverInterface;\n  humanInTheLoop?: boolean;\n  outputParser?: AutoGPTOutputParser;\n  maxIterations?: number;\n}\n\n/**\n * Class representing the AutoGPT concept with LangChain primitives. It is\n * designed to be used with a set of tools such as a search tool,\n * write-file tool, and a read-file tool.\n * @example\n * ```typescript\n * const autogpt = AutoGPT.fromLLMAndTools(\n *   new ChatOpenAI({ model: \"gpt-4o-mini\", temperature: 0 }),\n *   [\n *     new ReadFileTool({ store: new InMemoryFileStore() }),\n *     new WriteFileTool({ store: new InMemoryFileStore() }),\n *     new SerpAPI(\"YOUR_SERPAPI_API_KEY\", {\n *       location: \"San Francisco,California,United States\",\n *       hl: \"en\",\n *       gl: \"us\",\n *     }),\n *   ],\n *   {\n *     memory: new MemoryVectorStore(new OpenAIEmbeddings()).asRetriever(),\n *     aiName: \"Tom\",\n *     aiRole: \"Assistant\",\n *   },\n * );\n * const result = await autogpt.run([\"write a weather report for SF today\"]);\n * ```\n */\nexport class AutoGPT {\n  aiName: string;\n\n  memory: VectorStoreRetrieverInterface;\n\n  fullMessageHistory: BaseMessage[];\n\n  nextActionCount: number;\n\n  chain: LLMChain;\n\n  outputParser: AutoGPTOutputParser;\n\n  tools: ObjectTool[];\n\n  feedbackTool?: Tool;\n\n  maxIterations: number;\n\n  // Currently not generic enough to support any text splitter.\n  textSplitter: TokenTextSplitter;\n\n  constructor({\n    aiName,\n    memory,\n    chain,\n    outputParser,\n    tools,\n    feedbackTool,\n    maxIterations,\n  }: Omit<Required<AutoGPTInput>, \"aiRole\" | \"humanInTheLoop\"> & {\n    chain: LLMChain;\n    tools: ObjectTool[];\n    feedbackTool?: Tool;\n  }) {\n    this.aiName = aiName;\n    this.memory = memory;\n    this.fullMessageHistory = [];\n    this.nextActionCount = 0;\n    this.chain = chain;\n    this.outputParser = outputParser;\n    this.tools = tools;\n    this.feedbackTool = feedbackTool;\n    this.maxIterations = maxIterations;\n    const chunkSize = getEmbeddingContextSize(\n      \"modelName\" in memory.vectorStore.embeddings\n        ? (memory.vectorStore.embeddings.modelName as string)\n        : undefined\n    );\n    this.textSplitter = new TokenTextSplitter({\n      chunkSize,\n      chunkOverlap: Math.round(chunkSize / 10),\n    });\n  }\n\n  /**\n   * Creates a new AutoGPT instance from a given LLM and a set of tools.\n   * @param llm A BaseChatModel object.\n   * @param tools An array of ObjectTool objects.\n   * @param options.aiName The name of the AI.\n   * @param options.aiRole The role of the AI.\n   * @param options.memory A VectorStoreRetriever object that represents the memory of the AI.\n   * @param options.maxIterations The maximum number of iterations the AI can perform.\n   * @param options.outputParser An AutoGPTOutputParser object that parses the output of the AI.\n   * @returns A new instance of the AutoGPT class.\n   */\n  static fromLLMAndTools(\n    llm: BaseChatModel,\n    tools: ObjectTool[],\n    {\n      aiName,\n      aiRole,\n      memory,\n      maxIterations = 100,\n      // humanInTheLoop = false,\n      outputParser = new AutoGPTOutputParser(),\n    }: AutoGPTInput\n  ): AutoGPT {\n    const prompt = new AutoGPTPrompt({\n      aiName,\n      aiRole,\n      tools,\n      tokenCounter: llm.getNumTokens.bind(llm),\n      sendTokenLimit: getModelContextSize(\n        \"modelName\" in llm ? (llm.modelName as string) : \"gpt2\"\n      ),\n    });\n    // const feedbackTool = humanInTheLoop ? new HumanInputRun() : null;\n    const chain = new LLMChain({ llm, prompt });\n    return new AutoGPT({\n      aiName,\n      memory,\n      chain,\n      outputParser,\n      tools,\n      // feedbackTool,\n      maxIterations,\n    });\n  }\n\n  /**\n   * Runs the AI with a given set of goals.\n   * @param goals An array of strings representing the goals.\n   * @returns A string representing the result of the run or undefined if the maximum number of iterations is reached without a result.\n   */\n  async run(goals: string[]): Promise<string | undefined> {\n    const user_input =\n      \"Determine which next command to use, and respond using the format specified above:\";\n    let loopCount = 0;\n    while (loopCount < this.maxIterations) {\n      loopCount += 1;\n\n      const { text: assistantReply } = await this.chain.call({\n        goals,\n        user_input,\n        memory: this.memory,\n        messages: this.fullMessageHistory,\n      });\n\n      // Print the assistant reply\n      console.log(assistantReply);\n      this.fullMessageHistory.push(new HumanMessage(user_input));\n      this.fullMessageHistory.push(new AIMessage(assistantReply));\n\n      const action = await this.outputParser.parse(assistantReply);\n      const tools = this.tools.reduce(\n        (acc, tool) => ({ ...acc, [tool.name]: tool }),\n        {} as { [key: string]: ObjectTool }\n      );\n      if (action.name === FINISH_NAME) {\n        return action.args.response;\n      }\n      let result: string;\n      if (action.name in tools) {\n        const tool = tools[action.name];\n        let observation;\n        try {\n          observation = await tool.call(action.args);\n        } catch (e) {\n          observation = `Error in args: ${e}`;\n        }\n        result = `Command ${tool.name} returned: ${observation}`;\n      } else if (action.name === \"ERROR\") {\n        result = `Error: ${action.args}. `;\n      } else {\n        result = `Unknown command '${action.name}'. Please refer to the 'COMMANDS' list for available commands and only respond in the specified JSON format.`;\n      }\n\n      let memoryToAdd = `Assistant Reply: ${assistantReply}\\nResult: ${result} `;\n      if (this.feedbackTool) {\n        const feedback = `\\n${await this.feedbackTool.call(\"Input: \")}`;\n        if (feedback === \"q\" || feedback === \"stop\") {\n          console.log(\"EXITING\");\n          return \"EXITING\";\n        }\n        memoryToAdd += feedback;\n      }\n\n      const documents = await this.textSplitter.createDocuments([memoryToAdd]);\n      await this.memory.addDocuments(documents);\n      this.fullMessageHistory.push(new SystemMessage(result));\n    }\n\n    return undefined;\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AA2DA,IAAa,UAAb,MAAa,QAAQ;CACnB;CAEA;CAEA;CAEA;CAEA;CAEA;CAEA;CAEA;CAEA;CAGA;CAEA,YAAY,EACV,QACA,QACA,OACA,cACA,OACA,cACA,eAKD,EAAE;EACD,KAAK,SAAS;EACd,KAAK,SAAS;EACd,KAAK,qBAAqB,CAAE;EAC5B,KAAK,kBAAkB;EACvB,KAAK,QAAQ;EACb,KAAK,eAAe;EACpB,KAAK,QAAQ;EACb,KAAK,eAAe;EACpB,KAAK,gBAAgB;EACrB,MAAM,+EACJ,eAAe,OAAO,YAAY,aAC7B,OAAO,YAAY,WAAW,YAC/B,OACL;EACD,KAAK,eAAe,gDAAI,kBAAkB;GACxC;GACA,cAAc,KAAK,MAAM,YAAY,GAAG;EACzC;CACF;;;;;;;;;;;;CAaD,OAAO,gBACLA,KACAC,OACA,EACE,QACA,QACA,QACA,gBAAgB,KAEhB,eAAe,IAAIC,6CACN,EACN;EACT,MAAM,SAAS,IAAIC,6BAAc;GAC/B;GACA;GACA;GACA,cAAc,IAAI,aAAa,KAAK,IAAI;GACxC,+EACE,eAAe,MAAO,IAAI,YAAuB,OAClD;EACF;EAED,MAAM,QAAQ,IAAIC,2BAAS;GAAE;GAAK;EAAQ;AAC1C,SAAO,IAAI,QAAQ;GACjB;GACA;GACA;GACA;GACA;GAEA;EACD;CACF;;;;;;CAOD,MAAM,IAAIC,OAA8C;EACtD,MAAM,aACJ;EACF,IAAI,YAAY;AAChB,SAAO,YAAY,KAAK,eAAe;GACrC,aAAa;GAEb,MAAM,EAAE,MAAM,gBAAgB,GAAG,MAAM,KAAK,MAAM,KAAK;IACrD;IACA;IACA,QAAQ,KAAK;IACb,UAAU,KAAK;GAChB,EAAC;GAGF,QAAQ,IAAI,eAAe;GAC3B,KAAK,mBAAmB,KAAK,IAAIC,uCAAa,YAAY;GAC1D,KAAK,mBAAmB,KAAK,IAAIC,oCAAU,gBAAgB;GAE3D,MAAM,SAAS,MAAM,KAAK,aAAa,MAAM,eAAe;GAC5D,MAAM,QAAQ,KAAK,MAAM,OACvB,CAAC,KAAK,UAAU;IAAE,GAAG;KAAM,KAAK,OAAO;GAAM,IAC7C,CAAE,EACH;AACD,OAAI,OAAO,SAASC,2BAClB,QAAO,OAAO,KAAK;GAErB,IAAIC;AACJ,OAAI,OAAO,QAAQ,OAAO;IACxB,MAAM,OAAO,MAAM,OAAO;IAC1B,IAAI;AACJ,QAAI;KACF,cAAc,MAAM,KAAK,KAAK,OAAO,KAAK;IAC3C,SAAQ,GAAG;KACV,cAAc,CAAC,eAAe,EAAE,GAAG;IACpC;IACD,SAAS,CAAC,QAAQ,EAAE,KAAK,KAAK,WAAW,EAAE,aAAa;GACzD,WAAU,OAAO,SAAS,SACzB,SAAS,CAAC,OAAO,EAAE,OAAO,KAAK,EAAE,CAAC;QAElC,SAAS,CAAC,iBAAiB,EAAE,OAAO,KAAK,4GAA4G,CAAC;GAGxJ,IAAI,cAAc,CAAC,iBAAiB,EAAE,eAAe,UAAU,EAAE,OAAO,CAAC,CAAC;AAC1E,OAAI,KAAK,cAAc;IACrB,MAAM,WAAW,CAAC,EAAE,EAAE,MAAM,KAAK,aAAa,KAAK,UAAU,EAAE;AAC/D,QAAI,aAAa,OAAO,aAAa,QAAQ;KAC3C,QAAQ,IAAI,UAAU;AACtB,YAAO;IACR;IACD,eAAe;GAChB;GAED,MAAM,YAAY,MAAM,KAAK,aAAa,gBAAgB,CAAC,WAAY,EAAC;GACxE,MAAM,KAAK,OAAO,aAAa,UAAU;GACzC,KAAK,mBAAmB,KAAK,IAAIC,wCAAc,QAAQ;EACxD;AAED,SAAO;CACR;AACF"}

package/dist/experimental/autogpt/agent.d.cts

@@ -0,0 +1,102 @@
+import { LLMChain } from "../../chains/llm_chain.cjs";
+import { ObjectTool } from "./schema.cjs";
+import { AutoGPTOutputParser } from "./output_parser.cjs";
+import { text_splitter_d_exports } from "../../text_splitter.cjs";
+import { BaseMessage } from "@langchain/core/messages";
+import { Tool } from "@langchain/core/tools";
+import { VectorStoreRetrieverInterface } from "@langchain/core/vectorstores";
+import { BaseChatModel } from "@langchain/core/language_models/chat_models";
+
+//#region src/experimental/autogpt/agent.d.ts
+
+/**
+ * Interface for the input parameters of the AutoGPT class.
+ */
+interface AutoGPTInput {
+  aiName: string;
+  aiRole: string;
+  memory: VectorStoreRetrieverInterface;
+  humanInTheLoop?: boolean;
+  outputParser?: AutoGPTOutputParser;
+  maxIterations?: number;
+}
+/**
+ * Class representing the AutoGPT concept with LangChain primitives. It is
+ * designed to be used with a set of tools such as a search tool,
+ * write-file tool, and a read-file tool.
+ * @example
+ * ```typescript
+ * const autogpt = AutoGPT.fromLLMAndTools(
+ *   new ChatOpenAI({ model: "gpt-4o-mini", temperature: 0 }),
+ *   [
+ *     new ReadFileTool({ store: new InMemoryFileStore() }),
+ *     new WriteFileTool({ store: new InMemoryFileStore() }),
+ *     new SerpAPI("YOUR_SERPAPI_API_KEY", {
+ *       location: "San Francisco,California,United States",
+ *       hl: "en",
+ *       gl: "us",
+ *     }),
+ *   ],
+ *   {
+ *     memory: new MemoryVectorStore(new OpenAIEmbeddings()).asRetriever(),
+ *     aiName: "Tom",
+ *     aiRole: "Assistant",
+ *   },
+ * );
+ * const result = await autogpt.run(["write a weather report for SF today"]);
+ * ```
+ */
+declare class AutoGPT {
+  aiName: string;
+  memory: VectorStoreRetrieverInterface;
+  fullMessageHistory: BaseMessage[];
+  nextActionCount: number;
+  chain: LLMChain;
+  outputParser: AutoGPTOutputParser;
+  tools: ObjectTool[];
+  feedbackTool?: Tool;
+  maxIterations: number;
+  // Currently not generic enough to support any text splitter.
+  textSplitter: text_splitter_d_exports.TokenTextSplitter;
+  constructor({
+    aiName,
+    memory,
+    chain,
+    outputParser,
+    tools,
+    feedbackTool,
+    maxIterations
+  }: Omit<Required<AutoGPTInput>, "aiRole" | "humanInTheLoop"> & {
+    chain: LLMChain;
+    tools: ObjectTool[];
+    feedbackTool?: Tool;
+  });
+  /**
+   * Creates a new AutoGPT instance from a given LLM and a set of tools.
+   * @param llm A BaseChatModel object.
+   * @param tools An array of ObjectTool objects.
+   * @param options.aiName The name of the AI.
+   * @param options.aiRole The role of the AI.
+   * @param options.memory A VectorStoreRetriever object that represents the memory of the AI.
+   * @param options.maxIterations The maximum number of iterations the AI can perform.
+   * @param options.outputParser An AutoGPTOutputParser object that parses the output of the AI.
+   * @returns A new instance of the AutoGPT class.
+   */
+  static fromLLMAndTools(llm: BaseChatModel, tools: ObjectTool[], {
+    aiName,
+    aiRole,
+    memory,
+    maxIterations,
+    // humanInTheLoop = false,
+    outputParser
+  }: AutoGPTInput): AutoGPT;
+  /**
+   * Runs the AI with a given set of goals.
+   * @param goals An array of strings representing the goals.
+   * @returns A string representing the result of the run or undefined if the maximum number of iterations is reached without a result.
+   */
+  run(goals: string[]): Promise<string | undefined>;
+}
+//#endregion
+export { AutoGPT, AutoGPTInput };
+//# sourceMappingURL=agent.d.cts.map

package/dist/experimental/autogpt/agent.d.cts.map

@@ -0,0 +1 @@
+{"version":3,"file":"agent.d.cts","names":["VectorStoreRetrieverInterface","Tool","BaseMessage","BaseChatModel","LLMChain","AutoGPTOutputParser","ObjectTool","TokenTextSplitter","AutoGPTInput","AutoGPT","aiName","memory","chain","outputParser","tools","feedbackTool","maxIterations","Required","Omit","aiRole","Promise"],"sources":["../../../src/experimental/autogpt/agent.d.ts"],"sourcesContent":["import type { VectorStoreRetrieverInterface } from \"@langchain/core/vectorstores\";\nimport { Tool } from \"@langchain/core/tools\";\nimport { BaseMessage } from \"@langchain/core/messages\";\nimport { BaseChatModel } from \"@langchain/core/language_models/chat_models\";\nimport { LLMChain } from \"../../chains/llm_chain.js\";\nimport { AutoGPTOutputParser } from \"./output_parser.js\";\n// import { HumanInputRun } from \"./tools/human/tool\"; // TODO\nimport { ObjectTool } from \"./schema.js\";\nimport { TokenTextSplitter } from \"../../text_splitter.js\";\n/**\n * Interface for the input parameters of the AutoGPT class.\n */\nexport interface AutoGPTInput {\n    aiName: string;\n    aiRole: string;\n    memory: VectorStoreRetrieverInterface;\n    humanInTheLoop?: boolean;\n    outputParser?: AutoGPTOutputParser;\n    maxIterations?: number;\n}\n/**\n * Class representing the AutoGPT concept with LangChain primitives. It is\n * designed to be used with a set of tools such as a search tool,\n * write-file tool, and a read-file tool.\n * @example\n * ```typescript\n * const autogpt = AutoGPT.fromLLMAndTools(\n *   new ChatOpenAI({ model: \"gpt-4o-mini\", temperature: 0 }),\n *   [\n *     new ReadFileTool({ store: new InMemoryFileStore() }),\n *     new WriteFileTool({ store: new InMemoryFileStore() }),\n *     new SerpAPI(\"YOUR_SERPAPI_API_KEY\", {\n *       location: \"San Francisco,California,United States\",\n *       hl: \"en\",\n *       gl: \"us\",\n *     }),\n *   ],\n *   {\n *     memory: new MemoryVectorStore(new OpenAIEmbeddings()).asRetriever(),\n *     aiName: \"Tom\",\n *     aiRole: \"Assistant\",\n *   },\n * );\n * const result = await autogpt.run([\"write a weather report for SF today\"]);\n * ```\n */\nexport declare class AutoGPT {\n    aiName: string;\n    memory: VectorStoreRetrieverInterface;\n    fullMessageHistory: BaseMessage[];\n    nextActionCount: number;\n    chain: LLMChain;\n    outputParser: AutoGPTOutputParser;\n    tools: ObjectTool[];\n    feedbackTool?: Tool;\n    maxIterations: number;\n    // Currently not generic enough to support any text splitter.\n    textSplitter: TokenTextSplitter;\n    constructor({ aiName, memory, chain, outputParser, tools, feedbackTool, maxIterations }: Omit<Required<AutoGPTInput>, \"aiRole\" | \"humanInTheLoop\"> & {\n        chain: LLMChain;\n        tools: ObjectTool[];\n        feedbackTool?: Tool;\n    });\n    /**\n     * Creates a new AutoGPT instance from a given LLM and a set of tools.\n     * @param llm A BaseChatModel object.\n     * @param tools An array of ObjectTool objects.\n     * @param options.aiName The name of the AI.\n     * @param options.aiRole The role of the AI.\n     * @param options.memory A VectorStoreRetriever object that represents the memory of the AI.\n     * @param options.maxIterations The maximum number of iterations the AI can perform.\n     * @param options.outputParser An AutoGPTOutputParser object that parses the output of the AI.\n     * @returns A new instance of the AutoGPT class.\n     */\n    static fromLLMAndTools(llm: BaseChatModel, tools: ObjectTool[], { aiName, aiRole, memory, maxIterations, \n    // humanInTheLoop = false,\n    outputParser }: AutoGPTInput): AutoGPT;\n    /**\n     * Runs the AI with a given set of goals.\n     * @param goals An array of strings representing the goals.\n     * @returns A string representing the result of the run or undefined if the maximum number of iterations is reached without a result.\n     */\n    run(goals: string[]): Promise<string | undefined>;\n}\n"],"mappings":";;;;;;;;;;;;AAYA;;AAGYA,UAHKQ,YAAAA,CAGLR;EAA6B,MAEtBK,EAAAA,MAAAA;EAAmB,MAAA,EAAA,MAAA;EA6BjBI,MAAAA,EA/BTT,6BA+BgB;EAAA,cAAA,CAAA,EAAA,OAAA;EAAA,YAEhBA,CAAAA,EA/BOK,mBA+BPL;EAA6B,aACjBE,CAAAA,EAAAA,MAAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAiCEkB,cApCLX,OAAAA,CAoCKW;EAAO,MAAA,EAAA,MAAA;UAlCrBpB;sBACYE;;SAEbE;gBACOC;SACPC;iBACQL;;;gBAGDM,uBAAAA,CAAAA;;;;;;;;;KAC2EW,KAAKD,SAAST;WAC5FJ;WACAE;mBACQL;;;;;;;;;;;;;8BAaSE,sBAAsBG;;;;;;;KAElCE,eAAeC;;;;;;wBAMTW"}

package/dist/experimental/autogpt/agent.d.ts

@@ -0,0 +1,102 @@
+import { LLMChain } from "../../chains/llm_chain.js";
+import { ObjectTool } from "./schema.js";
+import { AutoGPTOutputParser } from "./output_parser.js";
+import { text_splitter_d_exports } from "../../text_splitter.js";
+import { Tool } from "@langchain/core/tools";
+import { BaseMessage } from "@langchain/core/messages";
+import { BaseChatModel } from "@langchain/core/language_models/chat_models";
+import { VectorStoreRetrieverInterface } from "@langchain/core/vectorstores";
+
+//#region src/experimental/autogpt/agent.d.ts
+
+/**
+ * Interface for the input parameters of the AutoGPT class.
+ */
+interface AutoGPTInput {
+  aiName: string;
+  aiRole: string;
+  memory: VectorStoreRetrieverInterface;
+  humanInTheLoop?: boolean;
+  outputParser?: AutoGPTOutputParser;
+  maxIterations?: number;
+}
+/**
+ * Class representing the AutoGPT concept with LangChain primitives. It is
+ * designed to be used with a set of tools such as a search tool,
+ * write-file tool, and a read-file tool.
+ * @example
+ * ```typescript
+ * const autogpt = AutoGPT.fromLLMAndTools(
+ *   new ChatOpenAI({ model: "gpt-4o-mini", temperature: 0 }),
+ *   [
+ *     new ReadFileTool({ store: new InMemoryFileStore() }),
+ *     new WriteFileTool({ store: new InMemoryFileStore() }),
+ *     new SerpAPI("YOUR_SERPAPI_API_KEY", {
+ *       location: "San Francisco,California,United States",
+ *       hl: "en",
+ *       gl: "us",
+ *     }),
+ *   ],
+ *   {
+ *     memory: new MemoryVectorStore(new OpenAIEmbeddings()).asRetriever(),
+ *     aiName: "Tom",
+ *     aiRole: "Assistant",
+ *   },
+ * );
+ * const result = await autogpt.run(["write a weather report for SF today"]);
+ * ```
+ */
+declare class AutoGPT {
+  aiName: string;
+  memory: VectorStoreRetrieverInterface;
+  fullMessageHistory: BaseMessage[];
+  nextActionCount: number;
+  chain: LLMChain;
+  outputParser: AutoGPTOutputParser;
+  tools: ObjectTool[];
+  feedbackTool?: Tool;
+  maxIterations: number;
+  // Currently not generic enough to support any text splitter.
+  textSplitter: text_splitter_d_exports.TokenTextSplitter;
+  constructor({
+    aiName,
+    memory,
+    chain,
+    outputParser,
+    tools,
+    feedbackTool,
+    maxIterations
+  }: Omit<Required<AutoGPTInput>, "aiRole" | "humanInTheLoop"> & {
+    chain: LLMChain;
+    tools: ObjectTool[];
+    feedbackTool?: Tool;
+  });
+  /**
+   * Creates a new AutoGPT instance from a given LLM and a set of tools.
+   * @param llm A BaseChatModel object.
+   * @param tools An array of ObjectTool objects.
+   * @param options.aiName The name of the AI.
+   * @param options.aiRole The role of the AI.
+   * @param options.memory A VectorStoreRetriever object that represents the memory of the AI.
+   * @param options.maxIterations The maximum number of iterations the AI can perform.
+   * @param options.outputParser An AutoGPTOutputParser object that parses the output of the AI.
+   * @returns A new instance of the AutoGPT class.
+   */
+  static fromLLMAndTools(llm: BaseChatModel, tools: ObjectTool[], {
+    aiName,
+    aiRole,
+    memory,
+    maxIterations,
+    // humanInTheLoop = false,
+    outputParser
+  }: AutoGPTInput): AutoGPT;
+  /**
+   * Runs the AI with a given set of goals.
+   * @param goals An array of strings representing the goals.
+   * @returns A string representing the result of the run or undefined if the maximum number of iterations is reached without a result.
+   */
+  run(goals: string[]): Promise<string | undefined>;
+}
+//#endregion
+export { AutoGPT, AutoGPTInput };
+//# sourceMappingURL=agent.d.ts.map

package/dist/experimental/autogpt/agent.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"agent.d.ts","names":["VectorStoreRetrieverInterface","Tool","BaseMessage","BaseChatModel","LLMChain","AutoGPTOutputParser","ObjectTool","TokenTextSplitter","AutoGPTInput","AutoGPT","aiName","memory","chain","outputParser","tools","feedbackTool","maxIterations","Required","Omit","aiRole","Promise"],"sources":["../../../src/experimental/autogpt/agent.d.ts"],"sourcesContent":["import type { VectorStoreRetrieverInterface } from \"@langchain/core/vectorstores\";\nimport { Tool } from \"@langchain/core/tools\";\nimport { BaseMessage } from \"@langchain/core/messages\";\nimport { BaseChatModel } from \"@langchain/core/language_models/chat_models\";\nimport { LLMChain } from \"../../chains/llm_chain.js\";\nimport { AutoGPTOutputParser } from \"./output_parser.js\";\n// import { HumanInputRun } from \"./tools/human/tool\"; // TODO\nimport { ObjectTool } from \"./schema.js\";\nimport { TokenTextSplitter } from \"../../text_splitter.js\";\n/**\n * Interface for the input parameters of the AutoGPT class.\n */\nexport interface AutoGPTInput {\n    aiName: string;\n    aiRole: string;\n    memory: VectorStoreRetrieverInterface;\n    humanInTheLoop?: boolean;\n    outputParser?: AutoGPTOutputParser;\n    maxIterations?: number;\n}\n/**\n * Class representing the AutoGPT concept with LangChain primitives. It is\n * designed to be used with a set of tools such as a search tool,\n * write-file tool, and a read-file tool.\n * @example\n * ```typescript\n * const autogpt = AutoGPT.fromLLMAndTools(\n *   new ChatOpenAI({ model: \"gpt-4o-mini\", temperature: 0 }),\n *   [\n *     new ReadFileTool({ store: new InMemoryFileStore() }),\n *     new WriteFileTool({ store: new InMemoryFileStore() }),\n *     new SerpAPI(\"YOUR_SERPAPI_API_KEY\", {\n *       location: \"San Francisco,California,United States\",\n *       hl: \"en\",\n *       gl: \"us\",\n *     }),\n *   ],\n *   {\n *     memory: new MemoryVectorStore(new OpenAIEmbeddings()).asRetriever(),\n *     aiName: \"Tom\",\n *     aiRole: \"Assistant\",\n *   },\n * );\n * const result = await autogpt.run([\"write a weather report for SF today\"]);\n * ```\n */\nexport declare class AutoGPT {\n    aiName: string;\n    memory: VectorStoreRetrieverInterface;\n    fullMessageHistory: BaseMessage[];\n    nextActionCount: number;\n    chain: LLMChain;\n    outputParser: AutoGPTOutputParser;\n    tools: ObjectTool[];\n    feedbackTool?: Tool;\n    maxIterations: number;\n    // Currently not generic enough to support any text splitter.\n    textSplitter: TokenTextSplitter;\n    constructor({ aiName, memory, chain, outputParser, tools, feedbackTool, maxIterations }: Omit<Required<AutoGPTInput>, \"aiRole\" | \"humanInTheLoop\"> & {\n        chain: LLMChain;\n        tools: ObjectTool[];\n        feedbackTool?: Tool;\n    });\n    /**\n     * Creates a new AutoGPT instance from a given LLM and a set of tools.\n     * @param llm A BaseChatModel object.\n     * @param tools An array of ObjectTool objects.\n     * @param options.aiName The name of the AI.\n     * @param options.aiRole The role of the AI.\n     * @param options.memory A VectorStoreRetriever object that represents the memory of the AI.\n     * @param options.maxIterations The maximum number of iterations the AI can perform.\n     * @param options.outputParser An AutoGPTOutputParser object that parses the output of the AI.\n     * @returns A new instance of the AutoGPT class.\n     */\n    static fromLLMAndTools(llm: BaseChatModel, tools: ObjectTool[], { aiName, aiRole, memory, maxIterations, \n    // humanInTheLoop = false,\n    outputParser }: AutoGPTInput): AutoGPT;\n    /**\n     * Runs the AI with a given set of goals.\n     * @param goals An array of strings representing the goals.\n     * @returns A string representing the result of the run or undefined if the maximum number of iterations is reached without a result.\n     */\n    run(goals: string[]): Promise<string | undefined>;\n}\n"],"mappings":";;;;;;;;;;;;AAYA;;AAGYA,UAHKQ,YAAAA,CAGLR;EAA6B,MAEtBK,EAAAA,MAAAA;EAAmB,MAAA,EAAA,MAAA;EA6BjBI,MAAAA,EA/BTT,6BA+BgB;EAAA,cAAA,CAAA,EAAA,OAAA;EAAA,YAEhBA,CAAAA,EA/BOK,mBA+BPL;EAA6B,aACjBE,CAAAA,EAAAA,MAAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAiCEkB,cApCLX,OAAAA,CAoCKW;EAAO,MAAA,EAAA,MAAA;UAlCrBpB;sBACYE;;SAEbE;gBACOC;SACPC;iBACQL;;;gBAGDM,uBAAAA,CAAAA;;;;;;;;;KAC2EW,KAAKD,SAAST;WAC5FJ;WACAE;mBACQL;;;;;;;;;;;;;8BAaSE,sBAAsBG;;;;;;;KAElCE,eAAeC;;;;;;wBAMTW"}

package/dist/experimental/autogpt/agent.js

@@ -0,0 +1,151 @@
+import { text_splitter_exports } from "../../text_splitter.js";
+import { LLMChain } from "../../chains/llm_chain.js";
+import { FINISH_NAME } from "./schema.js";
+import { AutoGPTPrompt } from "./prompt.js";
+import { AutoGPTOutputParser } from "./output_parser.js";
+import { getEmbeddingContextSize, getModelContextSize } from "@langchain/core/language_models/base";
+import { AIMessage, HumanMessage, SystemMessage } from "@langchain/core/messages";
+
+//#region src/experimental/autogpt/agent.ts
+/**
+* Class representing the AutoGPT concept with LangChain primitives. It is
+* designed to be used with a set of tools such as a search tool,
+* write-file tool, and a read-file tool.
+* @example
+* ```typescript
+* const autogpt = AutoGPT.fromLLMAndTools(
+*   new ChatOpenAI({ model: "gpt-4o-mini", temperature: 0 }),
+*   [
+*     new ReadFileTool({ store: new InMemoryFileStore() }),
+*     new WriteFileTool({ store: new InMemoryFileStore() }),
+*     new SerpAPI("YOUR_SERPAPI_API_KEY", {
+*       location: "San Francisco,California,United States",
+*       hl: "en",
+*       gl: "us",
+*     }),
+*   ],
+*   {
+*     memory: new MemoryVectorStore(new OpenAIEmbeddings()).asRetriever(),
+*     aiName: "Tom",
+*     aiRole: "Assistant",
+*   },
+* );
+* const result = await autogpt.run(["write a weather report for SF today"]);
+* ```
+*/
+var AutoGPT = class AutoGPT {
+	aiName;
+	memory;
+	fullMessageHistory;
+	nextActionCount;
+	chain;
+	outputParser;
+	tools;
+	feedbackTool;
+	maxIterations;
+	textSplitter;
+	constructor({ aiName, memory, chain, outputParser, tools, feedbackTool, maxIterations }) {
+		this.aiName = aiName;
+		this.memory = memory;
+		this.fullMessageHistory = [];
+		this.nextActionCount = 0;
+		this.chain = chain;
+		this.outputParser = outputParser;
+		this.tools = tools;
+		this.feedbackTool = feedbackTool;
+		this.maxIterations = maxIterations;
+		const chunkSize = getEmbeddingContextSize("modelName" in memory.vectorStore.embeddings ? memory.vectorStore.embeddings.modelName : void 0);
+		this.textSplitter = new text_splitter_exports.TokenTextSplitter({
+			chunkSize,
+			chunkOverlap: Math.round(chunkSize / 10)
+		});
+	}
+	/**
+	* Creates a new AutoGPT instance from a given LLM and a set of tools.
+	* @param llm A BaseChatModel object.
+	* @param tools An array of ObjectTool objects.
+	* @param options.aiName The name of the AI.
+	* @param options.aiRole The role of the AI.
+	* @param options.memory A VectorStoreRetriever object that represents the memory of the AI.
+	* @param options.maxIterations The maximum number of iterations the AI can perform.
+	* @param options.outputParser An AutoGPTOutputParser object that parses the output of the AI.
+	* @returns A new instance of the AutoGPT class.
+	*/
+	static fromLLMAndTools(llm, tools, { aiName, aiRole, memory, maxIterations = 100, outputParser = new AutoGPTOutputParser() }) {
+		const prompt = new AutoGPTPrompt({
+			aiName,
+			aiRole,
+			tools,
+			tokenCounter: llm.getNumTokens.bind(llm),
+			sendTokenLimit: getModelContextSize("modelName" in llm ? llm.modelName : "gpt2")
+		});
+		const chain = new LLMChain({
+			llm,
+			prompt
+		});
+		return new AutoGPT({
+			aiName,
+			memory,
+			chain,
+			outputParser,
+			tools,
+			maxIterations
+		});
+	}
+	/**
+	* Runs the AI with a given set of goals.
+	* @param goals An array of strings representing the goals.
+	* @returns A string representing the result of the run or undefined if the maximum number of iterations is reached without a result.
+	*/
+	async run(goals) {
+		const user_input = "Determine which next command to use, and respond using the format specified above:";
+		let loopCount = 0;
+		while (loopCount < this.maxIterations) {
+			loopCount += 1;
+			const { text: assistantReply } = await this.chain.call({
+				goals,
+				user_input,
+				memory: this.memory,
+				messages: this.fullMessageHistory
+			});
+			console.log(assistantReply);
+			this.fullMessageHistory.push(new HumanMessage(user_input));
+			this.fullMessageHistory.push(new AIMessage(assistantReply));
+			const action = await this.outputParser.parse(assistantReply);
+			const tools = this.tools.reduce((acc, tool) => ({
+				...acc,
+				[tool.name]: tool
+			}), {});
+			if (action.name === FINISH_NAME) return action.args.response;
+			let result;
+			if (action.name in tools) {
+				const tool = tools[action.name];
+				let observation;
+				try {
+					observation = await tool.call(action.args);
+				} catch (e) {
+					observation = `Error in args: ${e}`;
+				}
+				result = `Command ${tool.name} returned: ${observation}`;
+			} else if (action.name === "ERROR") result = `Error: ${action.args}. `;
+			else result = `Unknown command '${action.name}'. Please refer to the 'COMMANDS' list for available commands and only respond in the specified JSON format.`;
+			let memoryToAdd = `Assistant Reply: ${assistantReply}\nResult: ${result} `;
+			if (this.feedbackTool) {
+				const feedback = `\n${await this.feedbackTool.call("Input: ")}`;
+				if (feedback === "q" || feedback === "stop") {
+					console.log("EXITING");
+					return "EXITING";
+				}
+				memoryToAdd += feedback;
+			}
+			const documents = await this.textSplitter.createDocuments([memoryToAdd]);
+			await this.memory.addDocuments(documents);
+			this.fullMessageHistory.push(new SystemMessage(result));
+		}
+		return void 0;
+	}
+};
+
+//#endregion
+export { AutoGPT };
+//# sourceMappingURL=agent.js.map