GETApiChain.ts

import { BaseLanguageModel } from '@langchain/core/language_models/base'
import { PromptTemplate } from '@langchain/core/prompts'
import { APIChain } from 'langchain/chains'
import { getBaseClasses } from '../../../src/utils'
import { ICommonObject, INode, INodeData, INodeParams, IServerSideEventStreamer } from '../../../src/Interface'
import { ConsoleCallbackHandler, CustomChainHandler, additionalCallbacks } from '../../../src/handler'

export const API_URL_RAW_PROMPT_TEMPLATE = `You are given the below API Documentation:
{api_docs}
Using this documentation, generate the full API url to call for answering the user question.
You should build the API url in order to get a response that is as short as possible, while still getting the necessary information to answer the question. Pay attention to deliberately exclude any unnecessary pieces of data in the API call.

Question:{question}
API url:`

export const API_RESPONSE_RAW_PROMPT_TEMPLATE =
    'Given this {api_response} response for {api_url}. use the given response to answer this {question}'

class GETApiChain_Chains implements INode {
    label: string
    name: string
    version: number
    type: string
    icon: string
    category: string
    baseClasses: string[]
    description: string
    inputs: INodeParams[]

    constructor() {
        this.label = 'GET API Chain'
        this.name = 'getApiChain'
        this.version = 1.0
        this.type = 'GETApiChain'
        this.icon = 'get.svg'
        this.category = 'Chains'
        this.description = 'Chain to run queries against GET API'
        this.baseClasses = [this.type, ...getBaseClasses(APIChain)]
        this.inputs = [
            {
                label: 'Language Model',
                name: 'model',
                type: 'BaseLanguageModel'
            },
            {
                label: 'API Documentation',
                name: 'apiDocs',
                type: 'string',
                description:
                    'Description of how API works. Please refer to more <a target="_blank" href="https://github.com/langchain-ai/langchain/blob/master/libs/langchain/langchain/chains/api/open_meteo_docs.py">examples</a>',
                rows: 4
            },
            {
                label: 'Headers',
                name: 'headers',
                type: 'json',
                additionalParams: true,
                optional: true
            },
            {
                label: 'URL Prompt',
                name: 'urlPrompt',
                type: 'string',
                description: 'Prompt used to tell LLMs how to construct the URL. Must contains {api_docs} and {question}',
                default: API_URL_RAW_PROMPT_TEMPLATE,
                rows: 4,
                additionalParams: true
            },
            {
                label: 'Answer Prompt',
                name: 'ansPrompt',
                type: 'string',
                description:
                    'Prompt used to tell LLMs how to return the API response. Must contains {api_response}, {api_url}, and {question}',
                default: API_RESPONSE_RAW_PROMPT_TEMPLATE,
                rows: 4,
                additionalParams: true
            }
        ]
    }

    async init(nodeData: INodeData): Promise<any> {
        const model = nodeData.inputs?.model as BaseLanguageModel
        const apiDocs = nodeData.inputs?.apiDocs as string
        const headers = nodeData.inputs?.headers as string
        const urlPrompt = nodeData.inputs?.urlPrompt as string
        const ansPrompt = nodeData.inputs?.ansPrompt as string

        const chain = await getAPIChain(apiDocs, model, headers, urlPrompt, ansPrompt)
        return chain
    }

    async run(nodeData: INodeData, input: string, options: ICommonObject): Promise<string> {
        const model = nodeData.inputs?.model as BaseLanguageModel
        const apiDocs = nodeData.inputs?.apiDocs as string
        const headers = nodeData.inputs?.headers as string
        const urlPrompt = nodeData.inputs?.urlPrompt as string
        const ansPrompt = nodeData.inputs?.ansPrompt as string

        const chain = await getAPIChain(apiDocs, model, headers, urlPrompt, ansPrompt)
        const loggerHandler = new ConsoleCallbackHandler(options.logger)
        const callbacks = await additionalCallbacks(nodeData, options)
        const shouldStreamResponse = options.shouldStreamResponse
        const sseStreamer: IServerSideEventStreamer = options.sseStreamer as IServerSideEventStreamer
        const chatId = options.chatId

        if (shouldStreamResponse) {
            const handler = new CustomChainHandler(sseStreamer, chatId)
            const res = await chain.run(input, [loggerHandler, handler, ...callbacks])
            return res
        } else {
            const res = await chain.run(input, [loggerHandler, ...callbacks])
            return res
        }
    }
}

const getAPIChain = async (documents: string, llm: BaseLanguageModel, headers: string, urlPrompt: string, ansPrompt: string) => {
    const apiUrlPrompt = new PromptTemplate({
        inputVariables: ['api_docs', 'question'],
        template: urlPrompt ? urlPrompt : API_URL_RAW_PROMPT_TEMPLATE
    })

    const apiResponsePrompt = new PromptTemplate({
        inputVariables: ['api_docs', 'question', 'api_url', 'api_response'],
        template: ansPrompt ? ansPrompt : API_RESPONSE_RAW_PROMPT_TEMPLATE
    })

    const chain = APIChain.fromLLMAndAPIDocs(llm, documents, {
        apiUrlPrompt,
        apiResponsePrompt,
        verbose: process.env.DEBUG === 'true',
        headers: typeof headers === 'object' ? headers : headers ? JSON.parse(headers) : {}
    })
    return chain
}

module.exports = { nodeClass: GETApiChain_Chains }