import { ICommonObject, INode, INodeData, INodeOptionsValue, INodeOutputsValue, INodeParams } from '../../../src/Interface' import { S3Loader } from '@langchain/community/document_loaders/web/s3' import { UnstructuredLoader, UnstructuredLoaderOptions, UnstructuredLoaderStrategy, SkipInferTableTypes, HiResModelName } from '@langchain/community/document_loaders/fs/unstructured' import { getCredentialData, getCredentialParam, handleDocumentLoaderDocuments, handleDocumentLoaderMetadata, handleDocumentLoaderOutput } from '../../../src/utils' import { S3Client, GetObjectCommand, S3ClientConfig } from '@aws-sdk/client-s3' import { getRegions, MODEL_TYPE } from '../../../src/modelLoader' import { Readable } from 'node:stream' import * as fsDefault from 'node:fs' import * as path from 'node:path' import * as os from 'node:os' class S3_DocumentLoaders implements INode { label: string name: string version: number description: string type: string icon: string category: string baseClasses: string[] credential: INodeParams inputs?: INodeParams[] outputs: INodeOutputsValue[] constructor() { this.label = 'S3' this.name = 'S3' this.version = 4.0 this.type = 'Document' this.icon = 's3.svg' this.category = 'Document Loaders' this.description = 'Load Data from S3 Buckets' this.baseClasses = [this.type] this.credential = { label: 'AWS Credential', name: 'credential', type: 'credential', credentialNames: ['awsApi'], optional: true } this.inputs = [ { label: 'Bucket', name: 'bucketName', type: 'string' }, { label: 'Object Key', name: 'keyName', type: 'string', description: 'The object key (or key name) that uniquely identifies object in an Amazon S3 bucket', placeholder: 'AI-Paper.pdf' }, { label: 'Region', name: 'region', type: 'asyncOptions', loadMethod: 'listRegions', default: 'us-east-1' }, { label: 'Unstructured API URL', name: 'unstructuredAPIUrl', description: 'Your Unstructured.io URL. Read more on how to get started', type: 'string', placeholder: process.env.UNSTRUCTURED_API_URL || 'http://localhost:8000/general/v0/general', optional: !!process.env.UNSTRUCTURED_API_URL }, { label: 'Unstructured API KEY', name: 'unstructuredAPIKey', type: 'password', optional: true }, { label: 'Strategy', name: 'strategy', description: 'The strategy to use for partitioning PDF/image. Options are fast, hi_res, auto. Default: auto.', type: 'options', options: [ { label: 'Hi-Res', name: 'hi_res' }, { label: 'Fast', name: 'fast' }, { label: 'OCR Only', name: 'ocr_only' }, { label: 'Auto', name: 'auto' } ], optional: true, additionalParams: true, default: 'auto' }, { label: 'Encoding', name: 'encoding', description: 'The encoding method used to decode the text input. Default: utf-8.', type: 'string', optional: true, additionalParams: true, default: 'utf-8' }, { label: 'Skip Infer Table Types', name: 'skipInferTableTypes', description: 'The document types that you want to skip table extraction with. Default: pdf, jpg, png.', type: 'multiOptions', options: [ { label: 'doc', name: 'doc' }, { label: 'docx', name: 'docx' }, { label: 'eml', name: 'eml' }, { label: 'epub', name: 'epub' }, { label: 'heic', name: 'heic' }, { label: 'htm', name: 'htm' }, { label: 'html', name: 'html' }, { label: 'jpeg', name: 'jpeg' }, { label: 'jpg', name: 'jpg' }, { label: 'md', name: 'md' }, { label: 'msg', name: 'msg' }, { label: 'odt', name: 'odt' }, { label: 'pdf', name: 'pdf' }, { label: 'png', name: 'png' }, { label: 'ppt', name: 'ppt' }, { label: 'pptx', name: 'pptx' }, { label: 'rtf', name: 'rtf' }, { label: 'text', name: 'text' }, { label: 'txt', name: 'txt' }, { label: 'xls', name: 'xls' }, { label: 'xlsx', name: 'xlsx' } ], optional: true, additionalParams: true, default: '["pdf", "jpg", "png"]' }, { label: 'Hi-Res Model Name', name: 'hiResModelName', description: 'The name of the inference model used when strategy is hi_res. Default: detectron2_onnx.', type: 'options', options: [ { label: 'chipper', name: 'chipper', description: 'Exlusive to Unstructured hosted API. The Chipper model is Unstructured in-house image-to-text model based on transformer-based Visual Document Understanding (VDU) models.' }, { label: 'detectron2_onnx', name: 'detectron2_onnx', description: 'A Computer Vision model by Facebook AI that provides object detection and segmentation algorithms with ONNX Runtime. It is the fastest model with the hi_res strategy.' }, { label: 'yolox', name: 'yolox', description: 'A single-stage real-time object detector that modifies YOLOv3 with a DarkNet53 backbone.' }, { label: 'yolox_quantized', name: 'yolox_quantized', description: 'Runs faster than YoloX and its speed is closer to Detectron2.' } ], optional: true, additionalParams: true, default: 'detectron2_onnx' }, { label: 'Chunking Strategy', name: 'chunkingStrategy', description: 'Use one of the supported strategies to chunk the returned elements. When omitted, no chunking is performed and any other chunking parameters provided are ignored. Default: by_title', type: 'options', options: [ { label: 'None', name: 'None' }, { label: 'By Title', name: 'by_title' } ], optional: true, additionalParams: true, default: 'by_title' }, { label: 'OCR Languages', name: 'ocrLanguages', description: 'The languages to use for OCR. Note: Being depricated as languages is the new type. Pending langchain update.', type: 'multiOptions', options: [ { label: 'English', name: 'eng' }, { label: 'Spanish (Español)', name: 'spa' }, { label: 'Mandarin Chinese (普通话)', name: 'cmn' }, { label: 'Hindi (हिन्दी)', name: 'hin' }, { label: 'Arabic (اَلْعَرَبِيَّةُ)', name: 'ara' }, { label: 'Portuguese (Português)', name: 'por' }, { label: 'Bengali (বাংলা)', name: 'ben' }, { label: 'Russian (Русский)', name: 'rus' }, { label: 'Japanese (日本語)', name: 'jpn' }, { label: 'Punjabi (ਪੰਜਾਬੀ)', name: 'pan' }, { label: 'German (Deutsch)', name: 'deu' }, { label: 'Korean (한국어)', name: 'kor' }, { label: 'French (Français)', name: 'fra' }, { label: 'Italian (Italiano)', name: 'ita' }, { label: 'Vietnamese (Tiếng Việt)', name: 'vie' } ], optional: true, additionalParams: true }, { label: 'Source ID Key', name: 'sourceIdKey', type: 'string', description: 'Key used to get the true source of document, to be compared against the record. Document metadata must contain the Source ID Key.', default: 'source', placeholder: 'source', optional: true, additionalParams: true }, { label: 'Coordinates', name: 'coordinates', type: 'boolean', description: 'If true, return coordinates for each element. Default: false.', optional: true, additionalParams: true, default: false }, { label: 'XML Keep Tags', name: 'xmlKeepTags', description: 'If True, will retain the XML tags in the output. Otherwise it will simply extract the text from within the tags. Only applies to partition_xml.', type: 'boolean', optional: true, additionalParams: true }, { label: 'Include Page Breaks', name: 'includePageBreaks', description: 'When true, the output will include page break elements when the filetype supports it.', type: 'boolean', optional: true, additionalParams: true }, { label: 'XML Keep Tags', name: 'xmlKeepTags', description: 'Whether to keep XML tags in the output.', type: 'boolean', optional: true, additionalParams: true }, { label: 'Multi-Page Sections', name: 'multiPageSections', description: 'Whether to treat multi-page documents as separate sections.', type: 'boolean', optional: true, additionalParams: true }, { label: 'Combine Under N Chars', name: 'combineUnderNChars', description: "If chunking strategy is set, combine elements until a section reaches a length of n chars. Default: value of max_characters. Can't exceed value of max_characters.", type: 'number', optional: true, additionalParams: true }, { label: 'New After N Chars', name: 'newAfterNChars', description: "If chunking strategy is set, cut off new sections after reaching a length of n chars (soft max). value of max_characters. Can't exceed value of max_characters.", type: 'number', optional: true, additionalParams: true }, { label: 'Max Characters', name: 'maxCharacters', description: 'If chunking strategy is set, cut off new sections after reaching a length of n chars (hard max). Default: 500', type: 'number', optional: true, additionalParams: true, default: '500' }, { label: 'Additional Metadata', name: 'metadata', type: 'json', description: 'Additional metadata to be added to the extracted documents', optional: true, additionalParams: true }, { label: 'Omit Metadata Keys', name: 'omitMetadataKeys', type: 'string', rows: 4, description: 'Each document loader comes with a default set of metadata keys that are extracted from the document. You can use this field to omit some of the default metadata keys. The value should be a list of keys, seperated by comma. Use * to omit all metadata keys execept the ones you specify in the Additional Metadata field', placeholder: 'key1, key2, key3.nestedKey1', optional: true, additionalParams: true } ] this.outputs = [ { label: 'Document', name: 'document', description: 'Array of document objects containing metadata and pageContent', baseClasses: [...this.baseClasses, 'json'] }, { label: 'Text', name: 'text', description: 'Concatenated string from pageContent of documents', baseClasses: ['string', 'json'] } ] } loadMethods = { async listRegions(): Promise { return await getRegions(MODEL_TYPE.CHAT, 'awsChatBedrock') } } async init(nodeData: INodeData, _: string, options: ICommonObject): Promise { const bucketName = nodeData.inputs?.bucketName as string const keyName = nodeData.inputs?.keyName as string const region = nodeData.inputs?.region as string const unstructuredAPIUrl = nodeData.inputs?.unstructuredAPIUrl as string const unstructuredAPIKey = nodeData.inputs?.unstructuredAPIKey as string const strategy = nodeData.inputs?.strategy as UnstructuredLoaderStrategy const encoding = nodeData.inputs?.encoding as string const coordinates = nodeData.inputs?.coordinates as boolean const skipInferTableTypes = nodeData.inputs?.skipInferTableTypes ? JSON.parse(nodeData.inputs?.skipInferTableTypes as string) : ([] as SkipInferTableTypes[]) const hiResModelName = nodeData.inputs?.hiResModelName as HiResModelName const includePageBreaks = nodeData.inputs?.includePageBreaks as boolean const chunkingStrategy = nodeData.inputs?.chunkingStrategy as 'None' | 'by_title' const metadata = nodeData.inputs?.metadata const sourceIdKey = (nodeData.inputs?.sourceIdKey as string) || 'source' const ocrLanguages = nodeData.inputs?.ocrLanguages ? JSON.parse(nodeData.inputs?.ocrLanguages as string) : ([] as string[]) const xmlKeepTags = nodeData.inputs?.xmlKeepTags as boolean const multiPageSections = nodeData.inputs?.multiPageSections as boolean const combineUnderNChars = nodeData.inputs?.combineUnderNChars as number const newAfterNChars = nodeData.inputs?.newAfterNChars as number const maxCharacters = nodeData.inputs?.maxCharacters as number const _omitMetadataKeys = nodeData.inputs?.omitMetadataKeys as string const output = nodeData.outputs?.output as string let credentials: S3ClientConfig['credentials'] | undefined if (nodeData.credential) { const credentialData = await getCredentialData(nodeData.credential, options) const accessKeyId = getCredentialParam('awsKey', credentialData, nodeData) const secretAccessKey = getCredentialParam('awsSecret', credentialData, nodeData) if (accessKeyId && secretAccessKey) { credentials = { accessKeyId, secretAccessKey } } } const s3Config: S3ClientConfig = { region, credentials } const loader = new S3Loader({ bucket: bucketName, key: keyName, s3Config, unstructuredAPIURL: unstructuredAPIUrl, unstructuredAPIKey: unstructuredAPIKey }) loader.load = async () => { const tempDir = fsDefault.mkdtempSync(path.join(os.tmpdir(), 's3fileloader-')) const filePath = path.join(tempDir, keyName) try { const s3Client = new S3Client(s3Config) const getObjectCommand = new GetObjectCommand({ Bucket: bucketName, Key: keyName }) const response = await s3Client.send(getObjectCommand) const objectData = await new Promise((resolve, reject) => { const chunks: Buffer[] = [] if (response.Body instanceof Readable) { response.Body.on('data', (chunk: Buffer) => chunks.push(chunk)) response.Body.on('end', () => resolve(Buffer.concat(chunks))) response.Body.on('error', reject) } else { reject(new Error('Response body is not a readable stream.')) } }) fsDefault.mkdirSync(path.dirname(filePath), { recursive: true }) fsDefault.writeFileSync(filePath, objectData) } catch (e: any) { throw new Error(`Failed to download file ${keyName} from S3 bucket ${bucketName}: ${e.message}`) } try { const obj: UnstructuredLoaderOptions = { apiUrl: unstructuredAPIUrl, strategy, encoding, coordinates, skipInferTableTypes, hiResModelName, includePageBreaks, chunkingStrategy, ocrLanguages, xmlKeepTags, multiPageSections, combineUnderNChars, newAfterNChars, maxCharacters } if (unstructuredAPIKey) obj.apiKey = unstructuredAPIKey const unstructuredLoader = new UnstructuredLoader(filePath, obj) let docs = await handleDocumentLoaderDocuments(unstructuredLoader) docs = handleDocumentLoaderMetadata(docs, _omitMetadataKeys, metadata, sourceIdKey) return handleDocumentLoaderOutput(docs, output) } catch { throw new Error(`Failed to load file ${filePath} using unstructured loader.`) } finally { fsDefault.rmSync(path.dirname(filePath), { recursive: true }) } } return loader.load() } } module.exports = { nodeClass: S3_DocumentLoaders }