Automata/packages/@n8n/nodes-langchain/nodes/agents/Agent/agents/OpenAiFunctionsAgent/execute.ts

import {
	type IExecuteFunctions,
	type INodeExecutionData,
	NodeConnectionType,
	NodeOperationError,
} from 'n8n-workflow';

import type { AgentExecutorInput } from 'langchain/agents';
import { AgentExecutor, OpenAIAgent } from 'langchain/agents';
import type { Tool } from 'langchain/tools';
import type { BaseOutputParser } from 'langchain/schema/output_parser';
import { PromptTemplate } from 'langchain/prompts';
import { CombiningOutputParser } from 'langchain/output_parsers';
import { BufferMemory, type BaseChatMemory } from 'langchain/memory';
import { ChatOpenAI } from 'langchain/chat_models/openai';
import { getOptionalOutputParsers, getPromptInputByType } from '../../../../../utils/helpers';

export async function openAiFunctionsAgentExecute(
	this: IExecuteFunctions,
): Promise<INodeExecutionData[][]> {
	this.logger.verbose('Executing OpenAi Functions Agent');
	const model = (await this.getInputConnectionData(
		NodeConnectionType.AiLanguageModel,
		0,
	)) as ChatOpenAI;

	if (!(model instanceof ChatOpenAI)) {
		throw new NodeOperationError(
			this.getNode(),
			'OpenAI Functions Agent requires OpenAI Chat Model',
		);
	}
	const memory = (await this.getInputConnectionData(NodeConnectionType.AiMemory, 0)) as
		| BaseChatMemory
		| undefined;
	const tools = (await this.getInputConnectionData(NodeConnectionType.AiTool, 0)) as Tool[];
	const outputParsers = await getOptionalOutputParsers(this);
	const options = this.getNodeParameter('options', 0, {}) as {
		systemMessage?: string;
		maxIterations?: number;
		returnIntermediateSteps?: boolean;
	};

	const agentConfig: AgentExecutorInput = {
		tags: ['openai-functions'],
		agent: OpenAIAgent.fromLLMAndTools(model, tools, {
			prefix: options.systemMessage,
		}),
		tools,
		maxIterations: options.maxIterations ?? 10,
		returnIntermediateSteps: options?.returnIntermediateSteps === true,
		memory:
			memory ??
			new BufferMemory({
				returnMessages: true,
				memoryKey: 'chat_history',
				inputKey: 'input',
				outputKey: 'output',
			}),
	};

	const agentExecutor = AgentExecutor.fromAgentAndTools(agentConfig);

	const returnData: INodeExecutionData[] = [];

	let outputParser: BaseOutputParser | undefined;
	let prompt: PromptTemplate | undefined;
	if (outputParsers.length) {
		outputParser =
			outputParsers.length === 1 ? outputParsers[0] : new CombiningOutputParser(...outputParsers);

		const formatInstructions = outputParser.getFormatInstructions();

		prompt = new PromptTemplate({
			template: '{input}\n{formatInstructions}',
			inputVariables: ['input'],
			partialVariables: { formatInstructions },
		});
	}

	const items = this.getInputData();
	for (let itemIndex = 0; itemIndex < items.length; itemIndex++) {
		let input;
		if (this.getNode().typeVersion <= 1.2) {
			input = this.getNodeParameter('text', itemIndex) as string;
		} else {
			input = getPromptInputByType({
				ctx: this,
				i: itemIndex,
				inputKey: 'text',
				promptTypeKey: 'promptType',
			});
		}

		if (input === undefined) {
			throw new NodeOperationError(this.getNode(), 'The ‘text‘ parameter is empty.');
		}

		if (prompt) {
			input = (await prompt.invoke({ input })).value;
		}

		let response = await agentExecutor.call({ input, outputParsers });

		if (outputParser) {
			response = { output: await outputParser.parse(response.output as string) };
		}

		returnData.push({ json: response });
	}

	return await this.prepareOutputData(returnData);
}