farcasclaudiu/Flowise
1
0
Fork 0
mirror of https://github.com/farcasclaudiu/Flowise.git synced 2026-06-28 15:00:57 +03:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'main' of github.com:0xi4o/Flowise into feature/scrapped-links

Browse Source
This commit is contained in:
Ilango
2024-01-26 03:57:21 +05:30
parent 98acb35376 eb4d54552c
commit 94d8e003e7
60 changed files with 2471 additions and 1028 deletions
Download Patch File Download Diff File Expand all files Collapse all files
packages/components/nodes/chains/ConversationChain/ConversationChain.ts
+58 -28
View File
@@ -1,13 +1,12 @@
import { FlowiseMemory, ICommonObject, IMessage, INode, INodeData, INodeParams } from '../../../src/Interface'
import { ConversationChain } from 'langchain/chains'
import { getBaseClasses } from '../../../src/utils'
import { getBaseClasses, handleEscapeCharacters } from '../../../src/utils'
import { ChatPromptTemplate, HumanMessagePromptTemplate, MessagesPlaceholder, SystemMessagePromptTemplate } from 'langchain/prompts'
import { BaseChatModel } from 'langchain/chat_models/base'
import { ConsoleCallbackHandler, CustomChainHandler, additionalCallbacks } from '../../../src/handler'
import { flatten } from 'lodash'
import { Document } from 'langchain/document'
import { RunnableSequence } from 'langchain/schema/runnable'
import { StringOutputParser } from 'langchain/schema/output_parser'
import { ConsoleCallbackHandler as LCConsoleCallbackHandler } from '@langchain/core/tracers/console'
let systemMessage = `The following is a friendly conversation between a human and an AI. The AI is talkative and provides lots of specific details from its context. If the AI does not know the answer to a question, it truthfully says it does not know.`
const inputKey = 'input'
@@ -27,7 +26,7 @@ class ConversationChain_Chains implements INode {
constructor(fields?: { sessionId?: string }) {
this.label = 'Conversation Chain'
this.name = 'conversationChain'
this.version = 1.0
this.version = 2.0
this.type = 'ConversationChain'
this.icon = 'conv.svg'
this.category = 'Chains'
@@ -44,6 +43,14 @@ class ConversationChain_Chains implements INode {
name: 'memory',
type: 'BaseMemory'
},
{
label: 'Chat Prompt Template',
name: 'chatPromptTemplate',
type: 'ChatPromptTemplate',
description: 'Override existing prompt with Chat Prompt Template. Human Message must includes {input} variable',
optional: true
},
/* Deprecated
{
label: 'Document',
name: 'document',
@@ -52,15 +59,17 @@ class ConversationChain_Chains implements INode {
'Include whole document into the context window, if you get maximum context length error, please use model with higher context window like Claude 100k, or gpt4 32k',
optional: true,
list: true
},
},*/
{
label: 'System Message',
name: 'systemMessagePrompt',
type: 'string',
rows: 4,
description: 'If Chat Prompt Template is provided, this will be ignored',
additionalParams: true,
optional: true,
placeholder: 'You are a helpful assistant that write codes'
default: systemMessage,
placeholder: systemMessage
}
]
this.sessionId = fields?.sessionId
@@ -76,15 +85,21 @@ class ConversationChain_Chains implements INode {
const chain = prepareChain(nodeData, this.sessionId, options.chatHistory)
const loggerHandler = new ConsoleCallbackHandler(options.logger)
const callbacks = await additionalCallbacks(nodeData, options)
const additionalCallback = await additionalCallbacks(nodeData, options)
let res = ''
let callbacks = [loggerHandler, ...additionalCallback]
if (process.env.DEBUG === 'true') {
callbacks.push(new LCConsoleCallbackHandler())
}
if (options.socketIO && options.socketIOClientId) {
const handler = new CustomChainHandler(options.socketIO, options.socketIOClientId)
res = await chain.invoke({ input }, { callbacks: [loggerHandler, handler, ...callbacks] })
callbacks.push(handler)
res = await chain.invoke({ input }, { callbacks })
} else {
res = await chain.invoke({ input }, { callbacks: [loggerHandler, ...callbacks] })
res = await chain.invoke({ input }, { callbacks })
}
await memory.addChatMessages(
@@ -108,28 +123,27 @@ class ConversationChain_Chains implements INode {
const prepareChatPrompt = (nodeData: INodeData) => {
const memory = nodeData.inputs?.memory as FlowiseMemory
const prompt = nodeData.inputs?.systemMessagePrompt as string
const docs = nodeData.inputs?.document as Document[]
const chatPromptTemplate = nodeData.inputs?.chatPromptTemplate as ChatPromptTemplate
const flattenDocs = docs && docs.length ? flatten(docs) : []
const finalDocs = []
for (let i = 0; i < flattenDocs.length; i += 1) {
if (flattenDocs[i] && flattenDocs[i].pageContent) {
finalDocs.push(new Document(flattenDocs[i]))
if (chatPromptTemplate && chatPromptTemplate.promptMessages.length) {
const sysPrompt = chatPromptTemplate.promptMessages[0]
const humanPrompt = chatPromptTemplate.promptMessages[chatPromptTemplate.promptMessages.length - 1]
const chatPrompt = ChatPromptTemplate.fromMessages([
sysPrompt,
new MessagesPlaceholder(memory.memoryKey ?? 'chat_history'),
humanPrompt
])
if ((chatPromptTemplate as any).promptValues) {
// @ts-ignore
chatPrompt.promptValues = (chatPromptTemplate as any).promptValues
}
return chatPrompt
}
let finalText = ''
for (let i = 0; i < finalDocs.length; i += 1) {
finalText += finalDocs[i].pageContent
}
const replaceChar: string[] = ['{', '}']
for (const char of replaceChar) finalText = finalText.replaceAll(char, '')
if (finalText) systemMessage = `${systemMessage}\nThe AI has the following context:\n${finalText}`
const chatPrompt = ChatPromptTemplate.fromMessages([
SystemMessagePromptTemplate.fromTemplate(prompt ? `${prompt}\n${systemMessage}` : systemMessage),
SystemMessagePromptTemplate.fromTemplate(prompt ? prompt : systemMessage),
new MessagesPlaceholder(memory.memoryKey ?? 'chat_history'),
HumanMessagePromptTemplate.fromTemplate(`{${inputKey}}`)
])
@@ -142,15 +156,31 @@ const prepareChain = (nodeData: INodeData, sessionId?: string, chatHistory: IMes
const memory = nodeData.inputs?.memory as FlowiseMemory
const memoryKey = memory.memoryKey ?? 'chat_history'
const chatPrompt = prepareChatPrompt(nodeData)
let promptVariables = {}
const promptValuesRaw = (chatPrompt as any).promptValues
if (promptValuesRaw) {
const promptValues = handleEscapeCharacters(promptValuesRaw, true)
for (const val in promptValues) {
promptVariables = {
...promptVariables,
[val]: () => {
return promptValues[val]
}
}
}
}
const conversationChain = RunnableSequence.from([
{
[inputKey]: (input: { input: string }) => input.input,
[memoryKey]: async () => {
const history = await memory.getChatMessages(sessionId, true, chatHistory)
return history
}
},
...promptVariables
},
prepareChatPrompt(nodeData),
chatPrompt,
model,
new StringOutputParser()
])
packages/components/nodes/chains/ConversationalRetrievalQAChain/ConversationalRetrievalQAChain.ts
+9 -2
View File
@@ -13,6 +13,7 @@ import { applyPatch } from 'fast-json-patch'
import { convertBaseMessagetoIMessage, getBaseClasses } from '../../../src/utils'
import { ConsoleCallbackHandler, additionalCallbacks } from '../../../src/handler'
import { FlowiseMemory, ICommonObject, IMessage, INode, INodeData, INodeParams, MemoryMethods } from '../../../src/Interface'
import { ConsoleCallbackHandler as LCConsoleCallbackHandler } from '@langchain/core/tracers/console'
type RetrievalChainInput = {
chat_history: string
@@ -176,11 +177,17 @@ class ConversationalRetrievalQAChain_Chains implements INode {
const history = ((await memory.getChatMessages(this.sessionId, false, options.chatHistory)) as IMessage[]) ?? []
const loggerHandler = new ConsoleCallbackHandler(options.logger)
const callbacks = await additionalCallbacks(nodeData, options)
const additionalCallback = await additionalCallbacks(nodeData, options)
let callbacks = [loggerHandler, ...additionalCallback]
if (process.env.DEBUG === 'true') {
callbacks.push(new LCConsoleCallbackHandler())
}
const stream = answerChain.streamLog(
{ question: input, chat_history: history },
{ callbacks: [loggerHandler, ...callbacks] },
{ callbacks },
{
includeNames: [sourceRunnableName]
}
packages/components/nodes/chains/LLMChain/LLMChain.ts
+9 -2
View File
@@ -82,7 +82,7 @@ class LLMChain_Chains implements INode {
const model = nodeData.inputs?.model as BaseLanguageModel
const prompt = nodeData.inputs?.prompt
const output = nodeData.outputs?.output as string
const promptValues = prompt.promptValues as ICommonObject
let promptValues: ICommonObject | undefined = nodeData.inputs?.prompt.promptValues as ICommonObject
const llmOutputParser = nodeData.inputs?.outputParser as BaseOutputParser
this.outputParser = llmOutputParser
if (llmOutputParser) {
@@ -107,17 +107,24 @@ class LLMChain_Chains implements INode {
verbose: process.env.DEBUG === 'true'
})
const inputVariables = chain.prompt.inputVariables as string[] // ["product"]
promptValues = injectOutputParser(this.outputParser, chain, promptValues)
const res = await runPrediction(inputVariables, chain, input, promptValues, options, nodeData)
// eslint-disable-next-line no-console
console.log('\x1b[92m\x1b[1m\n*****OUTPUT PREDICTION*****\n\x1b[0m\x1b[0m')
// eslint-disable-next-line no-console
console.log(res)
let finalRes = res
if (this.outputParser && typeof res === 'object' && Object.prototype.hasOwnProperty.call(res, 'json')) {
finalRes = (res as ICommonObject).json
}
/**
* Apply string transformation to convert special chars:
* FROM: hello i am ben\n\n\thow are you?
* TO: hello i am benFLOWISE_NEWLINEFLOWISE_NEWLINEFLOWISE_TABhow are you?
*/
return handleEscapeCharacters(res, false)
return handleEscapeCharacters(finalRes, false)
}
}
packages/components/nodes/documentloaders/Figma/Figma.ts
+18 -1
View File
@@ -1,6 +1,7 @@
import { getCredentialData, getCredentialParam } from '../../../src'
import { ICommonObject, INode, INodeData, INodeParams } from '../../../src/Interface'
import { FigmaFileLoader, FigmaLoaderParams } from 'langchain/document_loaders/web/figma'
import { TextSplitter } from 'langchain/text_splitter'
class Figma_DocumentLoaders implements INode {
label: string
@@ -71,6 +72,8 @@ class Figma_DocumentLoaders implements INode {
async init(nodeData: INodeData, _: string, options: ICommonObject): Promise<any> {
const nodeIds = (nodeData.inputs?.nodeIds as string)?.trim().split(',') || []
const fileKey = nodeData.inputs?.fileKey as string
const textSplitter = nodeData.inputs?.textSplitter as TextSplitter
const metadata = nodeData.inputs?.metadata
const credentialData = await getCredentialData(nodeData.credential ?? '', options)
const accessToken = getCredentialParam('accessToken', credentialData, nodeData)
@@ -82,7 +85,21 @@ class Figma_DocumentLoaders implements INode {
}
const loader = new FigmaFileLoader(figmaOptions)
const docs = await loader.load()
const docs = textSplitter ? await loader.loadAndSplit() : await loader.load()
if (metadata) {
const parsedMetadata = typeof metadata === 'object' ? metadata : JSON.parse(metadata)
return docs.map((doc) => {
return {
...doc,
metadata: {
...doc.metadata,
...parsedMetadata
}
}
})
}
return docs
}
packages/components/nodes/llms/Azure OpenAI/AzureOpenAI.ts
+9 -1
View File
@@ -18,7 +18,7 @@ class AzureOpenAI_LLMs implements INode {
constructor() {
this.label = 'Azure OpenAI'
this.name = 'azureOpenAI'
this.version = 2.0
this.version = 2.1
this.type = 'AzureOpenAI'
this.icon = 'Azure.svg'
this.category = 'LLMs'
@@ -89,6 +89,14 @@ class AzureOpenAI_LLMs implements INode {
{
label: 'gpt-35-turbo',
name: 'gpt-35-turbo'
},
{
label: 'gpt-4',
name: 'gpt-4'
},
{
label: 'gpt-4-32k',
name: 'gpt-4-32k'
}
],
default: 'text-davinci-003',
packages/components/nodes/retrievers/CohereRerankRetriever/Cohere.svg
+1
View File
@@ -0,0 +1 @@
<svg width="32" height="32" fill="none" xmlns="http://www.w3.org/2000/svg"><path fill-rule="evenodd" clip-rule="evenodd" d="M11.776 18.304c.64 0 1.92-.032 3.712-.768 2.08-.864 6.176-2.4 9.152-4 2.08-1.12 2.976-2.592 2.976-4.576 0-2.72-2.208-4.96-4.96-4.96h-11.52A7.143 7.143 0 0 0 4 11.136c0 3.936 3.008 7.168 7.776 7.168Z" fill="#39594D"/><path fill-rule="evenodd" clip-rule="evenodd" d="M13.728 23.2c0-1.92 1.152-3.68 2.944-4.416l3.616-1.504C23.968 15.776 28 18.464 28 22.432A5.572 5.572 0 0 1 22.432 28h-3.936c-2.624 0-4.768-2.144-4.768-4.8Z" fill="#D18EE2"/><path d="M8.128 19.232A4.138 4.138 0 0 0 4 23.36v.544C4 26.144 5.856 28 8.128 28a4.138 4.138 0 0 0 4.128-4.128v-.544c-.032-2.24-1.856-4.096-4.128-4.096Z" fill="#FF7759"/></svg>
Side by Side

After

Width:  |  Height:  |  Size: 738 B

packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerank.ts
+55
View File
@@ -0,0 +1,55 @@
import { Callbacks } from 'langchain/callbacks'
import { Document } from 'langchain/document'
import { BaseDocumentCompressor } from 'langchain/retrievers/document_compressors'
import axios from 'axios'
export class CohereRerank extends BaseDocumentCompressor {
private cohereAPIKey: any
private COHERE_API_URL = 'https://api.cohere.ai/v1/rerank'
private readonly model: string
private readonly k: number
private readonly maxChunksPerDoc: number
constructor(cohereAPIKey: string, model: string, k: number, maxChunksPerDoc: number) {
super()
this.cohereAPIKey = cohereAPIKey
this.model = model
this.k = k
this.maxChunksPerDoc = maxChunksPerDoc
}
async compressDocuments(
documents: Document<Record<string, any>>[],
query: string,
_?: Callbacks | undefined
): Promise<Document<Record<string, any>>[]> {
// avoid empty api call
if (documents.length === 0) {
return []
}
const config = {
headers: {
Authorization: `Bearer ${this.cohereAPIKey}`,
'Content-Type': 'application/json',
Accept: 'application/json'
}
}
const data = {
model: this.model,
topN: this.k,
max_chunks_per_doc: this.maxChunksPerDoc,
query: query,
return_documents: false,
documents: documents.map((doc) => doc.pageContent)
}
try {
let returnedDocs = await axios.post(this.COHERE_API_URL, data, config)
const finalResults: Document<Record<string, any>>[] = []
returnedDocs.data.results.forEach((result: any) => {
const doc = documents[result.index]
doc.metadata.relevance_score = result.relevance_score
finalResults.push(doc)
})
return finalResults
} catch (error) {
return documents
}
}
}
packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerankRetriever.ts
+142
View File
@@ -0,0 +1,142 @@
import { ICommonObject, INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface'
import { BaseRetriever } from 'langchain/schema/retriever'
import { ContextualCompressionRetriever } from 'langchain/retrievers/contextual_compression'
import { getCredentialData, getCredentialParam, handleEscapeCharacters } from '../../../src'
import { CohereRerank } from './CohereRerank'
import { VectorStoreRetriever } from 'langchain/vectorstores/base'
class CohereRerankRetriever_Retrievers implements INode {
label: string
name: string
version: number
description: string
type: string
icon: string
category: string
baseClasses: string[]
inputs: INodeParams[]
credential: INodeParams
badge: string
outputs: INodeOutputsValue[]
constructor() {
this.label = 'Cohere Rerank Retriever'
this.name = 'cohereRerankRetriever'
this.version = 1.0
this.type = 'Cohere Rerank Retriever'
this.icon = 'Cohere.svg'
this.category = 'Retrievers'
this.badge = 'NEW'
this.description = 'Cohere Rerank indexes the documents from most to least semantically relevant to the query.'
this.baseClasses = [this.type, 'BaseRetriever']
this.credential = {
label: 'Connect Credential',
name: 'credential',
type: 'credential',
credentialNames: ['cohereApi']
}
this.inputs = [
{
label: 'Vector Store Retriever',
name: 'baseRetriever',
type: 'VectorStoreRetriever'
},
{
label: 'Model Name',
name: 'model',
type: 'options',
options: [
{
label: 'rerank-english-v2.0',
name: 'rerank-english-v2.0'
},
{
label: 'rerank-multilingual-v2.0',
name: 'rerank-multilingual-v2.0'
}
],
default: 'rerank-english-v2.0',
optional: true
},
{
label: 'Query',
name: 'query',
type: 'string',
description: 'Query to retrieve documents from retriever. If not specified, user question will be used',
optional: true,
acceptVariable: true
},
{
label: 'Top K',
name: 'topK',
description: 'Number of top results to fetch. Default to the TopK of the Base Retriever',
placeholder: '4',
type: 'number',
additionalParams: true,
optional: true
},
{
label: 'Max Chunks Per Doc',
name: 'maxChunksPerDoc',
description: 'The maximum number of chunks to produce internally from a document. Default to 10',
placeholder: '10',
type: 'number',
additionalParams: true,
optional: true
}
]
this.outputs = [
{
label: 'Cohere Rerank Retriever',
name: 'retriever',
baseClasses: this.baseClasses
},
{
label: 'Document',
name: 'document',
baseClasses: ['Document']
},
{
label: 'Text',
name: 'text',
baseClasses: ['string', 'json']
}
]
}
async init(nodeData: INodeData, input: string, options: ICommonObject): Promise<any> {
const baseRetriever = nodeData.inputs?.baseRetriever as BaseRetriever
const model = nodeData.inputs?.model as string
const query = nodeData.inputs?.query as string
const credentialData = await getCredentialData(nodeData.credential ?? '', options)
const cohereApiKey = getCredentialParam('cohereApiKey', credentialData, nodeData)
const topK = nodeData.inputs?.topK as string
const k = topK ? parseFloat(topK) : (baseRetriever as VectorStoreRetriever).k ?? 4
const maxChunksPerDoc = nodeData.inputs?.maxChunksPerDoc as string
const max_chunks_per_doc = maxChunksPerDoc ? parseFloat(maxChunksPerDoc) : 10
const output = nodeData.outputs?.output as string
const cohereCompressor = new CohereRerank(cohereApiKey, model, k, max_chunks_per_doc)
const retriever = new ContextualCompressionRetriever({
baseCompressor: cohereCompressor,
baseRetriever: baseRetriever
})
if (output === 'retriever') return retriever
else if (output === 'document') return await retriever.getRelevantDocuments(query ? query : input)
else if (output === 'text') {
let finaltext = ''
const docs = await retriever.getRelevantDocuments(query ? query : input)
for (const doc of docs) finaltext += `${doc.pageContent}\n`
return handleEscapeCharacters(finaltext, false)
}
return retriever
}
}
module.exports = { nodeClass: CohereRerankRetriever_Retrievers }
packages/components/nodes/retrievers/EmbeddingsFilterRetriever/EmbeddingsFilterRetriever.ts
+133
View File
@@ -0,0 +1,133 @@
import { INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface'
import { BaseRetriever } from 'langchain/schema/retriever'
import { Embeddings } from 'langchain/embeddings/base'
import { ContextualCompressionRetriever } from 'langchain/retrievers/contextual_compression'
import { EmbeddingsFilter } from 'langchain/retrievers/document_compressors/embeddings_filter'
import { handleEscapeCharacters } from '../../../src/utils'
class EmbeddingsFilterRetriever_Retrievers implements INode {
label: string
name: string
version: number
description: string
type: string
icon: string
category: string
baseClasses: string[]
inputs: INodeParams[]
outputs: INodeOutputsValue[]
badge: string
constructor() {
this.label = 'Embeddings Filter Retriever'
this.name = 'embeddingsFilterRetriever'
this.version = 1.0
this.type = 'EmbeddingsFilterRetriever'
this.icon = 'compressionRetriever.svg'
this.category = 'Retrievers'
this.badge = 'NEW'
this.description = 'A document compressor that uses embeddings to drop documents unrelated to the query'
this.baseClasses = [this.type, 'BaseRetriever']
this.inputs = [
{
label: 'Vector Store Retriever',
name: 'baseRetriever',
type: 'VectorStoreRetriever'
},
{
label: 'Embeddings',
name: 'embeddings',
type: 'Embeddings'
},
{
label: 'Query',
name: 'query',
type: 'string',
description: 'Query to retrieve documents from retriever. If not specified, user question will be used',
optional: true,
acceptVariable: true
},
{
label: 'Similarity Threshold',
name: 'similarityThreshold',
description:
'Threshold for determining when two documents are similar enough to be considered redundant. Must be specified if `k` is not set',
type: 'number',
default: 0.8,
step: 0.1,
optional: true
},
{
label: 'K',
name: 'k',
description:
'The number of relevant documents to return. Can be explicitly set to undefined, in which case similarity_threshold must be specified. Defaults to 20',
type: 'number',
default: 20,
step: 1,
optional: true,
additionalParams: true
}
]
this.outputs = [
{
label: 'Embeddings Filter Retriever',
name: 'retriever',
baseClasses: this.baseClasses
},
{
label: 'Document',
name: 'document',
baseClasses: ['Document']
},
{
label: 'Text',
name: 'text',
baseClasses: ['string', 'json']
}
]
}
async init(nodeData: INodeData, input: string): Promise<any> {
const baseRetriever = nodeData.inputs?.baseRetriever as BaseRetriever
const embeddings = nodeData.inputs?.embeddings as Embeddings
const query = nodeData.inputs?.query as string
const similarityThreshold = nodeData.inputs?.similarityThreshold as string
const k = nodeData.inputs?.k as string
const output = nodeData.outputs?.output as string
if (k === undefined && similarityThreshold === undefined) {
throw new Error(`Must specify one of "k" or "similarity_threshold".`)
}
const similarityThresholdNumber = similarityThreshold ? parseFloat(similarityThreshold) : 0.8
const kNumber = k ? parseFloat(k) : undefined
const baseCompressor = new EmbeddingsFilter({
embeddings: embeddings,
similarityThreshold: similarityThresholdNumber,
k: kNumber
})
const retriever = new ContextualCompressionRetriever({
baseCompressor,
baseRetriever: baseRetriever
})
if (output === 'retriever') return retriever
else if (output === 'document') return await retriever.getRelevantDocuments(query ? query : input)
else if (output === 'text') {
let finaltext = ''
const docs = await retriever.getRelevantDocuments(query ? query : input)
for (const doc of docs) finaltext += `${doc.pageContent}\n`
return handleEscapeCharacters(finaltext, false)
}
return retriever
}
}
module.exports = { nodeClass: EmbeddingsFilterRetriever_Retrievers }
packages/components/nodes/retrievers/EmbeddingsFilterRetriever/compressionRetriever.svg
+7
View File
@@ -0,0 +1,7 @@
<svg xmlns="http://www.w3.org/2000/svg" class="icon icon-tabler icon-tabler-chart-bar" width="24" height="24" viewBox="0 0 24 24" stroke-width="2" stroke="currentColor" fill="none" stroke-linecap="round" stroke-linejoin="round">
<path stroke="none" d="M0 0h24v24H0z" fill="none"/>
<path d="M3 12m0 1a1 1 0 0 1 1 -1h4a1 1 0 0 1 1 1v6a1 1 0 0 1 -1 1h-4a1 1 0 0 1 -1 -1z" />
<path d="M9 8m0 1a1 1 0 0 1 1 -1h4a1 1 0 0 1 1 1v10a1 1 0 0 1 -1 1h-4a1 1 0 0 1 -1 -1z" />
<path d="M15 4m0 1a1 1 0 0 1 1 -1h4a1 1 0 0 1 1 1v14a1 1 0 0 1 -1 1h-4a1 1 0 0 1 -1 -1z" />
<path d="M4 20l14 0" />
</svg>
Side by Side

After

Width:  |  Height:  |  Size: 600 B

packages/components/nodes/retrievers/HydeRetriever/HydeRetriever.ts
+46 -4
View File
@@ -1,8 +1,9 @@
import { VectorStore } from 'langchain/vectorstores/base'
import { INode, INodeData, INodeParams } from '../../../src/Interface'
import { INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface'
import { HydeRetriever, HydeRetrieverOptions, PromptKey } from 'langchain/retrievers/hyde'
import { BaseLanguageModel } from 'langchain/base_language'
import { PromptTemplate } from 'langchain/prompts'
import { handleEscapeCharacters } from '../../../src/utils'
class HydeRetriever_Retrievers implements INode {
label: string
@@ -14,11 +15,12 @@ class HydeRetriever_Retrievers implements INode {
category: string
baseClasses: string[]
inputs: INodeParams[]
outputs: INodeOutputsValue[]
constructor() {
this.label = 'Hyde Retriever'
this.label = 'HyDE Retriever'
this.name = 'HydeRetriever'
this.version = 2.0
this.version = 3.0
this.type = 'HydeRetriever'
this.icon = 'hyderetriever.svg'
this.category = 'Retrievers'
@@ -35,6 +37,14 @@ class HydeRetriever_Retrievers implements INode {
name: 'vectorStore',
type: 'VectorStore'
},
{
label: 'Query',
name: 'query',
type: 'string',
description: 'Query to retrieve documents from retriever. If not specified, user question will be used',
optional: true,
acceptVariable: true
},
{
label: 'Select Defined Prompt',
name: 'promptKey',
@@ -121,15 +131,34 @@ Passage:`
optional: true
}
]
this.outputs = [
{
label: 'HyDE Retriever',
name: 'retriever',
baseClasses: this.baseClasses
},
{
label: 'Document',
name: 'document',
baseClasses: ['Document']
},
{
label: 'Text',
name: 'text',
baseClasses: ['string', 'json']
}
]
}
async init(nodeData: INodeData): Promise<any> {
async init(nodeData: INodeData, input: string): Promise<any> {
const llm = nodeData.inputs?.model as BaseLanguageModel
const vectorStore = nodeData.inputs?.vectorStore as VectorStore
const promptKey = nodeData.inputs?.promptKey as PromptKey
const customPrompt = nodeData.inputs?.customPrompt as string
const query = nodeData.inputs?.query as string
const topK = nodeData.inputs?.topK as string
const k = topK ? parseFloat(topK) : 4
const output = nodeData.outputs?.output as string
const obj: HydeRetrieverOptions<any> = {
llm,
@@ -141,6 +170,19 @@ Passage:`
else if (promptKey) obj.promptTemplate = promptKey
const retriever = new HydeRetriever(obj)
if (output === 'retriever') return retriever
else if (output === 'document') return await retriever.getRelevantDocuments(query ? query : input)
else if (output === 'text') {
let finaltext = ''
const docs = await retriever.getRelevantDocuments(query ? query : input)
for (const doc of docs) finaltext += `${doc.pageContent}\n`
return handleEscapeCharacters(finaltext, false)
}
return retriever
}
}
packages/components/nodes/retrievers/LLMFilterRetriever/LLMFilterCompressionRetriever.ts
+100
View File
@@ -0,0 +1,100 @@
import { INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface'
import { BaseRetriever } from 'langchain/schema/retriever'
import { ContextualCompressionRetriever } from 'langchain/retrievers/contextual_compression'
import { BaseLanguageModel } from 'langchain/base_language'
import { LLMChainExtractor } from 'langchain/retrievers/document_compressors/chain_extract'
import { handleEscapeCharacters } from '../../../src/utils'
class LLMFilterCompressionRetriever_Retrievers implements INode {
label: string
name: string
version: number
description: string
type: string
icon: string
category: string
baseClasses: string[]
inputs: INodeParams[]
outputs: INodeOutputsValue[]
badge: string
constructor() {
this.label = 'LLM Filter Retriever'
this.name = 'llmFilterRetriever'
this.version = 1.0
this.type = 'LLMFilterRetriever'
this.icon = 'llmFilterRetriever.svg'
this.category = 'Retrievers'
this.badge = 'NEW'
this.description =
'Iterate over the initially returned documents and extract, from each, only the content that is relevant to the query'
this.baseClasses = [this.type, 'BaseRetriever']
this.inputs = [
{
label: 'Vector Store Retriever',
name: 'baseRetriever',
type: 'VectorStoreRetriever'
},
{
label: 'Language Model',
name: 'model',
type: 'BaseLanguageModel'
},
{
label: 'Query',
name: 'query',
type: 'string',
description: 'Query to retrieve documents from retriever. If not specified, user question will be used',
optional: true,
acceptVariable: true
}
]
this.outputs = [
{
label: 'LLM Filter Retriever',
name: 'retriever',
baseClasses: this.baseClasses
},
{
label: 'Document',
name: 'document',
baseClasses: ['Document']
},
{
label: 'Text',
name: 'text',
baseClasses: ['string', 'json']
}
]
}
async init(nodeData: INodeData, input: string): Promise<any> {
const baseRetriever = nodeData.inputs?.baseRetriever as BaseRetriever
const model = nodeData.inputs?.model as BaseLanguageModel
const query = nodeData.inputs?.query as string
const output = nodeData.outputs?.output as string
if (!model) throw new Error('There must be a LLM model connected to LLM Filter Retriever')
const retriever = new ContextualCompressionRetriever({
baseCompressor: LLMChainExtractor.fromLLM(model),
baseRetriever: baseRetriever
})
if (output === 'retriever') return retriever
else if (output === 'document') return await retriever.getRelevantDocuments(query ? query : input)
else if (output === 'text') {
let finaltext = ''
const docs = await retriever.getRelevantDocuments(query ? query : input)
for (const doc of docs) finaltext += `${doc.pageContent}\n`
return handleEscapeCharacters(finaltext, false)
}
return retriever
}
}
module.exports = { nodeClass: LLMFilterCompressionRetriever_Retrievers }
packages/components/nodes/retrievers/LLMFilterRetriever/llmFilterRetriever.svg
+1
View File
@@ -0,0 +1 @@
<svg xmlns="http://www.w3.org/2000/svg" class="icon icon-tabler icon-tabler-filter-check" width="24" height="24" viewBox="0 0 24 24" stroke-width="2" stroke="currentColor" fill="none" stroke-linecap="round" stroke-linejoin="round"><path stroke="none" d="M0 0h24v24H0z" fill="none"/><path d="M11.18 20.274l-2.18 .726v-8.5l-4.48 -4.928a2 2 0 0 1 -.52 -1.345v-2.227h16v2.172a2 2 0 0 1 -.586 1.414l-4.414 4.414v3" /><path d="M15 19l2 2l4 -4" /></svg>
Side by Side

After

Width:  |  Height:  |  Size: 446 B

packages/components/nodes/retrievers/RRFRetriever/RRFRetriever.ts
+136
View File
@@ -0,0 +1,136 @@
import { INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface'
import { BaseLanguageModel } from 'langchain/base_language'
import { ContextualCompressionRetriever } from 'langchain/retrievers/contextual_compression'
import { BaseRetriever } from 'langchain/schema/retriever'
import { ReciprocalRankFusion } from './ReciprocalRankFusion'
import { VectorStoreRetriever } from 'langchain/vectorstores/base'
import { handleEscapeCharacters } from '../../../src/utils'
class RRFRetriever_Retrievers implements INode {
label: string
name: string
version: number
description: string
type: string
icon: string
category: string
baseClasses: string[]
inputs: INodeParams[]
badge: string
outputs: INodeOutputsValue[]
constructor() {
this.label = 'Reciprocal Rank Fusion Retriever'
this.name = 'RRFRetriever'
this.version = 1.0
this.type = 'RRFRetriever'
this.badge = 'NEW'
this.icon = 'rrfRetriever.svg'
this.category = 'Retrievers'
this.description = 'Reciprocal Rank Fusion to re-rank search results by multiple query generation.'
this.baseClasses = [this.type, 'BaseRetriever']
this.inputs = [
{
label: 'Vector Store Retriever',
name: 'baseRetriever',
type: 'VectorStoreRetriever'
},
{
label: 'Language Model',
name: 'model',
type: 'BaseLanguageModel'
},
{
label: 'Query',
name: 'query',
type: 'string',
description: 'Query to retrieve documents from retriever. If not specified, user question will be used',
optional: true,
acceptVariable: true
},
{
label: 'Query Count',
name: 'queryCount',
description: 'Number of synthetic queries to generate. Default to 4',
placeholder: '4',
type: 'number',
default: 4,
additionalParams: true,
optional: true
},
{
label: 'Top K',
name: 'topK',
description: 'Number of top results to fetch. Default to the TopK of the Base Retriever',
placeholder: '0',
type: 'number',
additionalParams: true,
optional: true
},
{
label: 'Constant',
name: 'c',
description:
'A constant added to the rank, controlling the balance between the importance of high-ranked items and the consideration given to lower-ranked items.\n' +
'Default is 60',
placeholder: '60',
type: 'number',
default: 60,
additionalParams: true,
optional: true
}
]
this.outputs = [
{
label: 'Reciprocal Rank Fusion Retriever',
name: 'retriever',
baseClasses: this.baseClasses
},
{
label: 'Document',
name: 'document',
baseClasses: ['Document']
},
{
label: 'Text',
name: 'text',
baseClasses: ['string', 'json']
}
]
}
async init(nodeData: INodeData, input: string): Promise<any> {
const llm = nodeData.inputs?.model as BaseLanguageModel
const baseRetriever = nodeData.inputs?.baseRetriever as BaseRetriever
const query = nodeData.inputs?.query as string
const queryCount = nodeData.inputs?.queryCount as string
const q = queryCount ? parseFloat(queryCount) : 4
const topK = nodeData.inputs?.topK as string
const k = topK ? parseFloat(topK) : (baseRetriever as VectorStoreRetriever).k ?? 4
const constantC = nodeData.inputs?.c as string
const c = topK ? parseFloat(constantC) : 60
const output = nodeData.outputs?.output as string
const ragFusion = new ReciprocalRankFusion(llm, baseRetriever as VectorStoreRetriever, q, k, c)
const retriever = new ContextualCompressionRetriever({
baseCompressor: ragFusion,
baseRetriever: baseRetriever
})
if (output === 'retriever') return retriever
else if (output === 'document') return await retriever.getRelevantDocuments(query ? query : input)
else if (output === 'text') {
let finaltext = ''
const docs = await retriever.getRelevantDocuments(query ? query : input)
for (const doc of docs) finaltext += `${doc.pageContent}\n`
return handleEscapeCharacters(finaltext, false)
}
return retriever
}
}
module.exports = { nodeClass: RRFRetriever_Retrievers }
packages/components/nodes/retrievers/RRFRetriever/ReciprocalRankFusion.ts
+96
View File
@@ -0,0 +1,96 @@
import { BaseDocumentCompressor } from 'langchain/retrievers/document_compressors'
import { Document } from 'langchain/document'
import { Callbacks } from 'langchain/callbacks'
import { BaseLanguageModel } from 'langchain/base_language'
import { ChatPromptTemplate, HumanMessagePromptTemplate, SystemMessagePromptTemplate } from 'langchain/prompts'
import { LLMChain } from 'langchain/chains'
import { VectorStoreRetriever } from 'langchain/vectorstores/base'
export class ReciprocalRankFusion extends BaseDocumentCompressor {
private readonly llm: BaseLanguageModel
private readonly queryCount: number
private readonly topK: number
private readonly c: number
private baseRetriever: VectorStoreRetriever
constructor(llm: BaseLanguageModel, baseRetriever: VectorStoreRetriever, queryCount: number, topK: number, c: number) {
super()
this.queryCount = queryCount
this.llm = llm
this.baseRetriever = baseRetriever
this.topK = topK
this.c = c
}
async compressDocuments(
documents: Document<Record<string, any>>[],
query: string,
_?: Callbacks | undefined
): Promise<Document<Record<string, any>>[]> {
// avoid empty api call
if (documents.length === 0) {
return []
}
const chatPrompt = ChatPromptTemplate.fromMessages([
SystemMessagePromptTemplate.fromTemplate(
'You are a helpful assistant that generates multiple search queries based on a single input query.'
),
HumanMessagePromptTemplate.fromTemplate(
'Generate multiple search queries related to: {input}. Provide these alternative questions separated by newlines, do not add any numbers.'
),
HumanMessagePromptTemplate.fromTemplate('OUTPUT (' + this.queryCount + ' queries):')
])
const llmChain = new LLMChain({
llm: this.llm,
prompt: chatPrompt
})
const multipleQueries = await llmChain.call({ input: query })
const queries = []
queries.push(query)
multipleQueries.text.split('\n').map((q: string) => {
queries.push(q)
})
const docList: Document<Record<string, any>>[][] = []
for (let i = 0; i < queries.length; i++) {
const resultOne = await this.baseRetriever.vectorStore.similaritySearch(queries[i], 5)
const docs: any[] = []
resultOne.forEach((doc) => {
docs.push(doc)
})
docList.push(docs)
}
return this.reciprocalRankFunction(docList, this.c)
}
reciprocalRankFunction(docList: Document<Record<string, any>>[][], k: number): Document<Record<string, any>>[] {
docList.forEach((docs: Document<Record<string, any>>[]) => {
docs.forEach((doc: any, index: number) => {
let rank = index + 1
if (doc.metadata.relevancy_score) {
doc.metadata.relevancy_score += 1 / (rank + k)
} else {
doc.metadata.relevancy_score = 1 / (rank + k)
}
})
})
const scoreArray: any[] = []
docList.forEach((docs: Document<Record<string, any>>[]) => {
docs.forEach((doc: any) => {
scoreArray.push(doc.metadata.relevancy_score)
})
})
scoreArray.sort((a, b) => b - a)
const rerankedDocuments: Document<Record<string, any>>[] = []
const seenScores: any[] = []
scoreArray.forEach((score) => {
docList.forEach((docs) => {
docs.forEach((doc: any) => {
if (doc.metadata.relevancy_score === score && seenScores.indexOf(score) === -1) {
rerankedDocuments.push(doc)
seenScores.push(doc.metadata.relevancy_score)
}
})
})
})
return rerankedDocuments.splice(0, this.topK)
}
}
packages/components/nodes/retrievers/RRFRetriever/rrfRetriever.svg
+1
View File
@@ -0,0 +1 @@
<svg xmlns="http://www.w3.org/2000/svg" class="icon icon-tabler icon-tabler-math-x-divide-y-2" width="24" height="24" viewBox="0 0 24 24" stroke-width="2" stroke="currentColor" fill="none" stroke-linecap="round" stroke-linejoin="round"><path stroke="none" d="M0 0h24v24H0z" fill="none"/><path d="M3 21l18 -18" /><path d="M15 14l3 4.5" /><path d="M21 14l-4.5 7" /><path d="M3 4l6 6" /><path d="M3 10l6 -6" /></svg>
Side by Side

After

Width:  |  Height:  |  Size: 413 B

packages/components/nodes/retrievers/SimilarityThresholdRetriever/SimilarityThresholdRetriever.ts
+16 -5
View File
@@ -18,7 +18,7 @@ class SimilarityThresholdRetriever_Retrievers implements INode {
constructor() {
this.label = 'Similarity Score Threshold Retriever'
this.name = 'similarityThresholdRetriever'
this.version = 1.0
this.version = 2.0
this.type = 'SimilarityThresholdRetriever'
this.icon = 'similaritythreshold.svg'
this.category = 'Retrievers'
@@ -30,6 +30,14 @@ class SimilarityThresholdRetriever_Retrievers implements INode {
name: 'vectorStore',
type: 'VectorStore'
},
{
label: 'Query',
name: 'query',
type: 'string',
description: 'Query to retrieve documents from retriever. If not specified, user question will be used',
optional: true,
acceptVariable: true
},
{
label: 'Minimum Similarity Score (%)',
name: 'minSimilarityScore',
@@ -44,7 +52,8 @@ class SimilarityThresholdRetriever_Retrievers implements INode {
description: `The maximum number of results to fetch`,
type: 'number',
default: 20,
step: 1
step: 1,
additionalParams: true
},
{
label: 'K Increment',
@@ -52,7 +61,8 @@ class SimilarityThresholdRetriever_Retrievers implements INode {
description: `How much to increase K by each time. It'll fetch N results, then N + kIncrement, then N + kIncrement * 2, etc.`,
type: 'number',
default: 2,
step: 1
step: 1,
additionalParams: true
}
]
this.outputs = [
@@ -77,6 +87,7 @@ class SimilarityThresholdRetriever_Retrievers implements INode {
async init(nodeData: INodeData, input: string): Promise<any> {
const vectorStore = nodeData.inputs?.vectorStore as VectorStore
const minSimilarityScore = nodeData.inputs?.minSimilarityScore as number
const query = nodeData.inputs?.query as string
const maxK = nodeData.inputs?.maxK as string
const kIncrement = nodeData.inputs?.kIncrement as string
@@ -89,11 +100,11 @@ class SimilarityThresholdRetriever_Retrievers implements INode {
})
if (output === 'retriever') return retriever
else if (output === 'document') return await retriever.getRelevantDocuments(input)
else if (output === 'document') return await retriever.getRelevantDocuments(query ? query : input)
else if (output === 'text') {
let finaltext = ''
const docs = await retriever.getRelevantDocuments(input)
const docs = await retriever.getRelevantDocuments(query ? query : input)
for (const doc of docs) finaltext += `${doc.pageContent}\n`
packages/components/nodes/tools/CustomTool/CustomTool.ts
+2 -18
View File
@@ -1,5 +1,5 @@
import { ICommonObject, IDatabaseEntity, INode, INodeData, INodeOptionsValue, INodeParams } from '../../../src/Interface'
import { convertSchemaToZod, getBaseClasses } from '../../../src/utils'
import { convertSchemaToZod, getBaseClasses, getVars } from '../../../src/utils'
import { DynamicStructuredTool } from './core'
import { z } from 'zod'
import { DataSource } from 'typeorm'
@@ -81,23 +81,7 @@ class CustomTool_Tools implements INode {
}
if (customToolFunc) obj.code = customToolFunc
const variables = await appDataSource.getRepository(databaseEntities['Variable']).find()
// override variables defined in overrideConfig
// nodeData.inputs.variables is an Object, check each property and override the variable
if (nodeData?.inputs?.vars) {
for (const propertyName of Object.getOwnPropertyNames(nodeData.inputs.vars)) {
const foundVar = variables.find((v) => v.name === propertyName)
if (foundVar) {
// even if the variable was defined as runtime, we override it with static value
foundVar.type = 'static'
foundVar.value = nodeData.inputs.vars[propertyName]
} else {
// add it the variables, if not found locally in the db
variables.push({ name: propertyName, type: 'static', value: nodeData.inputs.vars[propertyName] })
}
}
}
const variables = await getVars(appDataSource, databaseEntities, nodeData)
const flow = { chatflowId: options.chatflowid }
packages/components/nodes/tools/CustomTool/core.ts
+2 -37
View File
@@ -1,6 +1,6 @@
import { z } from 'zod'
import { NodeVM } from 'vm2'
import { availableDependencies } from '../../../src/utils'
import { availableDependencies, defaultAllowBuiltInDep, prepareSandboxVars } from '../../../src/utils'
import { RunnableConfig } from '@langchain/core/runnables'
import { StructuredTool, ToolParams } from '@langchain/core/tools'
import { CallbackManagerForToolRun, Callbacks, CallbackManager, parseCallbackConfigArg } from '@langchain/core/callbacks/manager'
@@ -112,48 +112,13 @@ export class DynamicStructuredTool<
}
}
// inject variables
let vars = {}
if (this.variables) {
for (const item of this.variables) {
let value = item.value
// read from .env file
if (item.type === 'runtime') {
value = process.env[item.name]
}
Object.defineProperty(vars, item.name, {
enumerable: true,
configurable: true,
writable: true,
value: value
})
}
}
sandbox['$vars'] = vars
sandbox['$vars'] = prepareSandboxVars(this.variables)
// inject flow properties
if (this.flowObj) {
sandbox['$flow'] = { ...this.flowObj, ...flowConfig }
}
const defaultAllowBuiltInDep = [
'assert',
'buffer',
'crypto',
'events',
'http',
'https',
'net',
'path',
'querystring',
'timers',
'tls',
'url',
'zlib'
]
const builtinDeps = process.env.TOOL_FUNCTION_BUILTIN_DEP
? defaultAllowBuiltInDep.concat(process.env.TOOL_FUNCTION_BUILTIN_DEP.split(','))
: defaultAllowBuiltInDep
packages/components/nodes/utilities/CustomFunction/CustomFunction.ts
+27 -19
View File
@@ -1,6 +1,7 @@
import { ICommonObject, INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface'
import { ICommonObject, IDatabaseEntity, INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface'
import { NodeVM } from 'vm2'
import { availableDependencies, handleEscapeCharacters } from '../../../src/utils'
import { DataSource } from 'typeorm'
import { availableDependencies, defaultAllowBuiltInDep, getVars, handleEscapeCharacters, prepareSandboxVars } from '../../../src/utils'
class CustomFunction_Utilities implements INode {
label: string
@@ -55,9 +56,19 @@ class CustomFunction_Utilities implements INode {
]
}
async init(nodeData: INodeData, input: string): Promise<any> {
async init(nodeData: INodeData, input: string, options: ICommonObject): Promise<any> {
const javascriptFunction = nodeData.inputs?.javascriptFunction as string
const functionInputVariablesRaw = nodeData.inputs?.functionInputVariables
const appDataSource = options.appDataSource as DataSource
const databaseEntities = options.databaseEntities as IDatabaseEntity
const variables = await getVars(appDataSource, databaseEntities, nodeData)
const flow = {
chatflowId: options.chatflowid,
sessionId: options.sessionId,
chatId: options.chatId,
input
}
let inputVars: ICommonObject = {}
if (functionInputVariablesRaw) {
@@ -69,7 +80,20 @@ class CustomFunction_Utilities implements INode {
}
}
// Some values might be a stringified JSON, parse it
for (const key in inputVars) {
if (typeof inputVars[key] === 'string' && inputVars[key].startsWith('{') && inputVars[key].endsWith('}')) {
try {
inputVars[key] = JSON.parse(inputVars[key])
} catch (e) {
continue
}
}
}
let sandbox: any = { $input: input }
sandbox['$vars'] = prepareSandboxVars(variables)
sandbox['$flow'] = flow
if (Object.keys(inputVars).length) {
for (const item in inputVars) {
@@ -81,22 +105,6 @@ class CustomFunction_Utilities implements INode {
}
}
const defaultAllowBuiltInDep = [
'assert',
'buffer',
'crypto',
'events',
'http',
'https',
'net',
'path',
'querystring',
'timers',
'tls',
'url',
'zlib'
]
const builtinDeps = process.env.TOOL_FUNCTION_BUILTIN_DEP
? defaultAllowBuiltInDep.concat(process.env.TOOL_FUNCTION_BUILTIN_DEP.split(','))
: defaultAllowBuiltInDep
packages/components/nodes/utilities/IfElseFunction/IfElseFunction.ts
+33 -21
View File
@@ -1,6 +1,7 @@
import { ICommonObject, INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface'
import { ICommonObject, IDatabaseEntity, INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface'
import { NodeVM } from 'vm2'
import { availableDependencies } from '../../../src/utils'
import { DataSource } from 'typeorm'
import { availableDependencies, defaultAllowBuiltInDep, getVars, handleEscapeCharacters, prepareSandboxVars } from '../../../src/utils'
class IfElseFunction_Utilities implements INode {
label: string
@@ -73,10 +74,20 @@ class IfElseFunction_Utilities implements INode {
]
}
async init(nodeData: INodeData, input: string): Promise<any> {
async init(nodeData: INodeData, input: string, options: ICommonObject): Promise<any> {
const ifFunction = nodeData.inputs?.ifFunction as string
const elseFunction = nodeData.inputs?.elseFunction as string
const functionInputVariablesRaw = nodeData.inputs?.functionInputVariables
const appDataSource = options.appDataSource as DataSource
const databaseEntities = options.databaseEntities as IDatabaseEntity
const variables = await getVars(appDataSource, databaseEntities, nodeData)
const flow = {
chatflowId: options.chatflowid,
sessionId: options.sessionId,
chatId: options.chatId,
input
}
let inputVars: ICommonObject = {}
if (functionInputVariablesRaw) {
@@ -84,34 +95,35 @@ class IfElseFunction_Utilities implements INode {
inputVars =
typeof functionInputVariablesRaw === 'object' ? functionInputVariablesRaw : JSON.parse(functionInputVariablesRaw)
} catch (exception) {
throw new Error("Invalid JSON in the PromptTemplate's promptValues: " + exception)
throw new Error("Invalid JSON in the IfElse's Input Variables: " + exception)
}
}
// Some values might be a stringified JSON, parse it
for (const key in inputVars) {
if (typeof inputVars[key] === 'string' && inputVars[key].startsWith('{') && inputVars[key].endsWith('}')) {
try {
inputVars[key] = JSON.parse(inputVars[key])
} catch (e) {
continue
}
}
}
let sandbox: any = { $input: input }
sandbox['$vars'] = prepareSandboxVars(variables)
sandbox['$flow'] = flow
if (Object.keys(inputVars).length) {
for (const item in inputVars) {
sandbox[`$${item}`] = inputVars[item]
let value = inputVars[item]
if (typeof value === 'string') {
value = handleEscapeCharacters(value, true)
}
sandbox[`$${item}`] = value
}
}
const defaultAllowBuiltInDep = [
'assert',
'buffer',
'crypto',
'events',
'http',
'https',
'net',
'path',
'querystring',
'timers',
'tls',
'url',
'zlib'
]
const builtinDeps = process.env.TOOL_FUNCTION_BUILTIN_DEP
? defaultAllowBuiltInDep.concat(process.env.TOOL_FUNCTION_BUILTIN_DEP.split(','))
: defaultAllowBuiltInDep
packages/components/nodes/vectorstores/Astra/Astra.ts
+4 -12
View File
@@ -4,6 +4,7 @@ import { Document } from 'langchain/document'
import { ICommonObject, INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface'
import { getBaseClasses, getCredentialData } from '../../../src/utils'
import { AstraDBVectorStore, AstraLibArgs } from '@langchain/community/vectorstores/astradb'
import { addMMRInputParams, resolveVectorStoreOrRetriever } from '../VectorStoreUtils'
class Astra_VectorStores implements INode {
label: string
@@ -26,7 +27,7 @@ class Astra_VectorStores implements INode {
this.type = 'Astra'
this.icon = 'astra.svg'
this.category = 'Vector Stores'
this.description = `Upsert embedded data and perform similarity search upon query using DataStax Astra DB, a serverless vector database that’s perfect for managing mission-critical AI workloads`
this.description = `Upsert embedded data and perform similarity or mmr search upon query using DataStax Astra DB, a serverless vector database that’s perfect for managing mission-critical AI workloads`
this.baseClasses = [this.type, 'VectorStoreRetriever', 'BaseRetriever']
this.badge = 'NEW'
this.credential = {
@@ -74,6 +75,7 @@ class Astra_VectorStores implements INode {
optional: true
}
]
addMMRInputParams(this.inputs)
this.outputs = [
{
label: 'Astra Retriever',
@@ -139,9 +141,6 @@ class Astra_VectorStores implements INode {
const embeddings = nodeData.inputs?.embeddings as Embeddings
const vectorDimension = nodeData.inputs?.vectorDimension as number
const similarityMetric = nodeData.inputs?.similarityMetric as 'cosine' | 'euclidean' | 'dot_product' | undefined
const output = nodeData.outputs?.output as string
const topK = nodeData.inputs?.topK as string
const k = topK ? parseFloat(topK) : 4
const credentialData = await getCredentialData(nodeData.credential ?? '', options)
@@ -176,14 +175,7 @@ class Astra_VectorStores implements INode {
const vectorStore = await AstraDBVectorStore.fromExistingIndex(embeddings, astraConfig)
if (output === 'retriever') {
const retriever = vectorStore.asRetriever(k)
return retriever
} else if (output === 'vectorStore') {
;(vectorStore as any).k = k
return vectorStore
}
return vectorStore
return resolveVectorStoreOrRetriever(nodeData, vectorStore)
}
}
packages/components/nodes/vectorstores/MongoDBAtlas/MongoDBAtlas.ts
+4 -11
View File
@@ -5,6 +5,7 @@ import { Embeddings } from 'langchain/embeddings/base'
import { Document } from 'langchain/document'
import { ICommonObject, INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface'
import { getBaseClasses, getCredentialData, getCredentialParam } from '../../../src/utils'
import { addMMRInputParams, resolveVectorStoreOrRetriever } from '../VectorStoreUtils'
class MongoDBAtlas_VectorStores implements INode {
label: string
@@ -24,7 +25,7 @@ class MongoDBAtlas_VectorStores implements INode {
this.label = 'MongoDB Atlas'
this.name = 'mongoDBAtlas'
this.version = 1.0
this.description = `Upsert embedded data and perform similarity search upon query using MongoDB Atlas, a managed cloud mongodb database`
this.description = `Upsert embedded data and perform similarity or mmr search upon query using MongoDB Atlas, a managed cloud mongodb database`
this.type = 'MongoDB Atlas'
this.icon = 'mongodb.svg'
this.category = 'Vector Stores'
@@ -95,6 +96,7 @@ class MongoDBAtlas_VectorStores implements INode {
optional: true
}
]
addMMRInputParams(this.inputs)
this.outputs = [
{
label: 'MongoDB Retriever',
@@ -162,9 +164,6 @@ class MongoDBAtlas_VectorStores implements INode {
let textKey = nodeData.inputs?.textKey as string
let embeddingKey = nodeData.inputs?.embeddingKey as string
const embeddings = nodeData.inputs?.embeddings as Embeddings
const topK = nodeData.inputs?.topK as string
const k = topK ? parseFloat(topK) : 4
const output = nodeData.outputs?.output as string
let mongoDBConnectUrl = getCredentialParam('mongoDBConnectUrl', credentialData, nodeData)
@@ -181,13 +180,7 @@ class MongoDBAtlas_VectorStores implements INode {
embeddingKey
})
if (output === 'retriever') {
return vectorStore.asRetriever(k)
} else if (output === 'vectorStore') {
;(vectorStore as any).k = k
return vectorStore
}
return vectorStore
return resolveVectorStoreOrRetriever(nodeData, vectorStore)
}
}
packages/components/nodes/vectorstores/Pinecone/Pinecone.ts
+7 -19
View File
@@ -5,6 +5,7 @@ import { Embeddings } from 'langchain/embeddings/base'
import { Document } from 'langchain/document'
import { ICommonObject, INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface'
import { getBaseClasses, getCredentialData, getCredentialParam } from '../../../src/utils'
import { addMMRInputParams, resolveVectorStoreOrRetriever } from '../VectorStoreUtils'
class Pinecone_VectorStores implements INode {
label: string
@@ -23,11 +24,11 @@ class Pinecone_VectorStores implements INode {
constructor() {
this.label = 'Pinecone'
this.name = 'pinecone'
this.version = 1.0
this.version = 2.0
this.type = 'Pinecone'
this.icon = 'pinecone.svg'
this.category = 'Vector Stores'
this.description = `Upsert embedded data and perform similarity search upon query using Pinecone, a leading fully managed hosted vector database`
this.description = `Upsert embedded data and perform similarity or mmr search using Pinecone, a leading fully managed hosted vector database`
this.baseClasses = [this.type, 'VectorStoreRetriever', 'BaseRetriever']
this.badge = 'NEW'
this.credential = {
@@ -79,6 +80,7 @@ class Pinecone_VectorStores implements INode {
optional: true
}
]
addMMRInputParams(this.inputs)
this.outputs = [
{
label: 'Pinecone Retriever',
@@ -103,11 +105,9 @@ class Pinecone_VectorStores implements INode {
const credentialData = await getCredentialData(nodeData.credential ?? '', options)
const pineconeApiKey = getCredentialParam('pineconeApiKey', credentialData, nodeData)
const pineconeEnv = getCredentialParam('pineconeEnv', credentialData, nodeData)
const client = new Pinecone({
apiKey: pineconeApiKey,
environment: pineconeEnv
apiKey: pineconeApiKey
})
const pineconeIndex = client.Index(index)
@@ -140,17 +140,12 @@ class Pinecone_VectorStores implements INode {
const pineconeMetadataFilter = nodeData.inputs?.pineconeMetadataFilter
const docs = nodeData.inputs?.document as Document[]
const embeddings = nodeData.inputs?.embeddings as Embeddings
const output = nodeData.outputs?.output as string
const topK = nodeData.inputs?.topK as string
const k = topK ? parseFloat(topK) : 4
const credentialData = await getCredentialData(nodeData.credential ?? '', options)
const pineconeApiKey = getCredentialParam('pineconeApiKey', credentialData, nodeData)
const pineconeEnv = getCredentialParam('pineconeEnv', credentialData, nodeData)
const client = new Pinecone({
apiKey: pineconeApiKey,
environment: pineconeEnv
apiKey: pineconeApiKey
})
const pineconeIndex = client.Index(index)
@@ -175,14 +170,7 @@ class Pinecone_VectorStores implements INode {
const vectorStore = await PineconeStore.fromExistingIndex(embeddings, obj)
if (output === 'retriever') {
const retriever = vectorStore.asRetriever(k)
return retriever
} else if (output === 'vectorStore') {
;(vectorStore as any).k = k
return vectorStore
}
return vectorStore
return resolveVectorStoreOrRetriever(nodeData, vectorStore)
}
}
packages/components/nodes/vectorstores/Pinecone/Pinecone_Existing.ts
+1 -3
View File
@@ -95,11 +95,9 @@ class Pinecone_Existing_VectorStores implements INode {
const credentialData = await getCredentialData(nodeData.credential ?? '', options)
const pineconeApiKey = getCredentialParam('pineconeApiKey', credentialData, nodeData)
const pineconeEnv = getCredentialParam('pineconeEnv', credentialData, nodeData)
const client = new Pinecone({
apiKey: pineconeApiKey,
environment: pineconeEnv
apiKey: pineconeApiKey
})
const pineconeIndex = client.Index(index)
packages/components/nodes/vectorstores/Pinecone/Pinecone_Upsert.ts
+1 -3
View File
@@ -96,11 +96,9 @@ class PineconeUpsert_VectorStores implements INode {
const credentialData = await getCredentialData(nodeData.credential ?? '', options)
const pineconeApiKey = getCredentialParam('pineconeApiKey', credentialData, nodeData)
const pineconeEnv = getCredentialParam('pineconeEnv', credentialData, nodeData)
const client = new Pinecone({
apiKey: pineconeApiKey,
environment: pineconeEnv
apiKey: pineconeApiKey
})
const pineconeIndex = client.Index(index)
packages/components/nodes/vectorstores/Supabase/Supabase.ts
+5 -13
View File
@@ -5,6 +5,7 @@ import { Embeddings } from 'langchain/embeddings/base'
import { ICommonObject, INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface'
import { getBaseClasses, getCredentialData, getCredentialParam } from '../../../src/utils'
import { SupabaseLibArgs, SupabaseVectorStore } from 'langchain/vectorstores/supabase'
import { addMMRInputParams, resolveVectorStoreOrRetriever } from '../VectorStoreUtils'
class Supabase_VectorStores implements INode {
label: string
@@ -23,11 +24,11 @@ class Supabase_VectorStores implements INode {
constructor() {
this.label = 'Supabase'
this.name = 'supabase'
this.version = 1.0
this.version = 2.0
this.type = 'Supabase'
this.icon = 'supabase.svg'
this.category = 'Vector Stores'
this.description = 'Upsert embedded data and perform similarity search upon query using Supabase via pgvector extension'
this.description = 'Upsert embedded data and perform similarity or mmr search upon query using Supabase via pgvector extension'
this.baseClasses = [this.type, 'VectorStoreRetriever', 'BaseRetriever']
this.badge = 'NEW'
this.credential = {
@@ -81,6 +82,7 @@ class Supabase_VectorStores implements INode {
optional: true
}
]
addMMRInputParams(this.inputs)
this.outputs = [
{
label: 'Supabase Retriever',
@@ -135,9 +137,6 @@ class Supabase_VectorStores implements INode {
const queryName = nodeData.inputs?.queryName as string
const embeddings = nodeData.inputs?.embeddings as Embeddings
const supabaseMetadataFilter = nodeData.inputs?.supabaseMetadataFilter
const output = nodeData.outputs?.output as string
const topK = nodeData.inputs?.topK as string
const k = topK ? parseFloat(topK) : 4
const credentialData = await getCredentialData(nodeData.credential ?? '', options)
const supabaseApiKey = getCredentialParam('supabaseApiKey', credentialData, nodeData)
@@ -157,14 +156,7 @@ class Supabase_VectorStores implements INode {
const vectorStore = await SupabaseVectorStore.fromExistingIndex(embeddings, obj)
if (output === 'retriever') {
const retriever = vectorStore.asRetriever(k)
return retriever
} else if (output === 'vectorStore') {
;(vectorStore as any).k = k
return vectorStore
}
return vectorStore
return resolveVectorStoreOrRetriever(nodeData, vectorStore)
}
}
packages/components/nodes/vectorstores/VectorStoreUtils.ts
+75
View File
@@ -0,0 +1,75 @@
import { INodeData } from '../../src'
export const resolveVectorStoreOrRetriever = (nodeData: INodeData, vectorStore: any) => {
const output = nodeData.outputs?.output as string
const searchType = nodeData.outputs?.searchType as string
const topK = nodeData.inputs?.topK as string
const k = topK ? parseFloat(topK) : 4
if (output === 'retriever') {
if ('mmr' === searchType) {
const fetchK = nodeData.inputs?.fetchK as string
const lambda = nodeData.inputs?.lambda as string
const f = fetchK ? parseInt(fetchK) : 20
const l = lambda ? parseFloat(lambda) : 0.5
return vectorStore.asRetriever({
searchType: 'mmr',
k: k,
searchKwargs: {
fetchK: f,
lambda: l
}
})
} else {
// "searchType" is "similarity"
return vectorStore.asRetriever(k)
}
} else if (output === 'vectorStore') {
;(vectorStore as any).k = k
return vectorStore
}
}
export const addMMRInputParams = (inputs: any[]) => {
const mmrInputParams = [
{
label: 'Search Type',
name: 'searchType',
type: 'options',
default: 'similarity',
options: [
{
label: 'Similarity',
name: 'similarity'
},
{
label: 'Max Marginal Relevance',
name: 'mmr'
}
],
additionalParams: true,
optional: true
},
{
label: 'Fetch K (for MMR Search)',
name: 'fetchK',
description: 'Number of initial documents to fetch for MMR reranking. Default to 20. Used only when the search type is MMR',
placeholder: '20',
type: 'number',
additionalParams: true,
optional: true
},
{
label: 'Lambda (for MMR Search)',
name: 'lambda',
description:
'Number between 0 and 1 that determines the degree of diversity among the results, where 0 corresponds to maximum diversity and 1 to minimum diversity. Used only when the search type is MMR',
placeholder: '0.5',
type: 'number',
additionalParams: true,
optional: true
}
]
inputs.push(...mmrInputParams)
}
packages/components/nodes/vectorstores/Weaviate/Weaviate.ts
+5 -13
View File
@@ -5,6 +5,7 @@ import { Document } from 'langchain/document'
import { Embeddings } from 'langchain/embeddings/base'
import { ICommonObject, INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface'
import { getBaseClasses, getCredentialData, getCredentialParam } from '../../../src/utils'
import { addMMRInputParams, resolveVectorStoreOrRetriever } from '../VectorStoreUtils'
class Weaviate_VectorStores implements INode {
label: string
@@ -23,12 +24,12 @@ class Weaviate_VectorStores implements INode {
constructor() {
this.label = 'Weaviate'
this.name = 'weaviate'
this.version = 1.0
this.version = 2.0
this.type = 'Weaviate'
this.icon = 'weaviate.png'
this.category = 'Vector Stores'
this.description =
'Upsert embedded data and perform similarity search upon query using Weaviate, a scalable open-source vector database'
'Upsert embedded data and perform similarity or mmr search using Weaviate, a scalable open-source vector database'
this.baseClasses = [this.type, 'VectorStoreRetriever', 'BaseRetriever']
this.badge = 'NEW'
this.credential = {
@@ -107,6 +108,7 @@ class Weaviate_VectorStores implements INode {
optional: true
}
]
addMMRInputParams(this.inputs)
this.outputs = [
{
label: 'Weaviate Retriever',
@@ -174,9 +176,6 @@ class Weaviate_VectorStores implements INode {
const weaviateTextKey = nodeData.inputs?.weaviateTextKey as string
const weaviateMetadataKeys = nodeData.inputs?.weaviateMetadataKeys as string
const embeddings = nodeData.inputs?.embeddings as Embeddings
const output = nodeData.outputs?.output as string
const topK = nodeData.inputs?.topK as string
const k = topK ? parseFloat(topK) : 4
const credentialData = await getCredentialData(nodeData.credential ?? '', options)
const weaviateApiKey = getCredentialParam('weaviateApiKey', credentialData, nodeData)
@@ -199,14 +198,7 @@ class Weaviate_VectorStores implements INode {
const vectorStore = await WeaviateStore.fromExistingIndex(embeddings, obj)
if (output === 'retriever') {
const retriever = vectorStore.asRetriever(k)
return retriever
} else if (output === 'vectorStore') {
;(vectorStore as any).k = k
return vectorStore
}
return vectorStore
return resolveVectorStoreOrRetriever(nodeData, vectorStore)
}
}
packages/components/nodes/vectorstores/Zep/Zep.ts
+7 -15
View File
@@ -5,6 +5,7 @@ import { Embeddings } from 'langchain/embeddings/base'
import { Document } from 'langchain/document'
import { ICommonObject, INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface'
import { getBaseClasses, getCredentialData, getCredentialParam } from '../../../src/utils'
import { addMMRInputParams, resolveVectorStoreOrRetriever } from '../VectorStoreUtils'
class Zep_VectorStores implements INode {
label: string
@@ -23,12 +24,12 @@ class Zep_VectorStores implements INode {
constructor() {
this.label = 'Zep'
this.name = 'zep'
this.version = 1.0
this.version = 2.0
this.type = 'Zep'
this.icon = 'zep.svg'
this.category = 'Vector Stores'
this.description =
'Upsert embedded data and perform similarity search upon query using Zep, a fast and scalable building block for LLM apps'
'Upsert embedded data and perform similarity or mmr search upon query using Zep, a fast and scalable building block for LLM apps'
this.baseClasses = [this.type, 'VectorStoreRetriever', 'BaseRetriever']
this.badge = 'NEW'
this.credential = {
@@ -88,6 +89,7 @@ class Zep_VectorStores implements INode {
optional: true
}
]
addMMRInputParams(this.inputs)
this.outputs = [
{
label: 'Zep Retriever',
@@ -144,9 +146,6 @@ class Zep_VectorStores implements INode {
const zepMetadataFilter = nodeData.inputs?.zepMetadataFilter
const dimension = nodeData.inputs?.dimension as number
const embeddings = nodeData.inputs?.embeddings as Embeddings
const output = nodeData.outputs?.output as string
const topK = nodeData.inputs?.topK as string
const k = topK ? parseFloat(topK) : 4
const credentialData = await getCredentialData(nodeData.credential ?? '', options)
const apiKey = getCredentialParam('apiKey', credentialData, nodeData)
@@ -165,14 +164,7 @@ class Zep_VectorStores implements INode {
const vectorStore = await ZepExistingVS.fromExistingIndex(embeddings, zepConfig)
if (output === 'retriever') {
const retriever = vectorStore.asRetriever(k)
return retriever
} else if (output === 'vectorStore') {
;(vectorStore as any).k = k
return vectorStore
}
return vectorStore
return resolveVectorStoreOrRetriever(nodeData, vectorStore)
}
}
@@ -210,7 +202,7 @@ class ZepExistingVS extends ZepVectorStore {
this.args = args
}
async initalizeCollection(args: IZepConfig & Partial<ZepFilter>) {
async initializeCollection(args: IZepConfig & Partial<ZepFilter>) {
this.client = await ZepClient.init(args.apiUrl, args.apiKey)
try {
this.collection = await this.client.document.getCollection(args.collectionName)
@@ -259,7 +251,7 @@ class ZepExistingVS extends ZepVectorStore {
const newfilter = {
where: { and: ANDFilters }
}
await this.initalizeCollection(this.args!).catch((err) => {
await this.initializeCollection(this.args!).catch((err) => {
console.error('Error initializing collection:', err)
throw err
})