add subquery engine

packages/components/nodes/chatmodels/ChatOpenAI/ChatOpenAI_LlamaIndex.ts

@@ -41,6 +41,14 @@ class ChatOpenAI_LlamaIndex_LLMs implements INode {
                         label: 'gpt-4',
                         name: 'gpt-4'
                     },
+                    {
+                        label: 'gpt-4-turbo-preview',
+                        name: 'gpt-4-turbo-preview'
+                    },
+                    {
+                        label: 'gpt-4-0125-preview',
+                        name: 'gpt-4-0125-preview'
+                    },
                     {
                         label: 'gpt-4-1106-preview',
                         name: 'gpt-4-1106-preview'

packages/components/nodes/embeddings/OpenAIEmbedding/OpenAIEmbedding_LlamaIndex.ts

@@ -32,6 +32,27 @@ class OpenAIEmbedding_LlamaIndex_Embeddings implements INode {
             credentialNames: ['openAIApi']
         }
         this.inputs = [
+            {
+                label: 'Model Name',
+                name: 'modelName',
+                type: 'options',
+                options: [
+                    {
+                        label: 'text-embedding-3-large',
+                        name: 'text-embedding-3-large'
+                    },
+                    {
+                        label: 'text-embedding-3-small',
+                        name: 'text-embedding-3-small'
+                    },
+                    {
+                        label: 'text-embedding-ada-002',
+                        name: 'text-embedding-ada-002'
+                    }
+                ],
+                default: 'text-embedding-ada-002',
+                optional: true
+            },
             {
                 label: 'Timeout',
                 name: 'timeout',
@@ -51,12 +72,14 @@ class OpenAIEmbedding_LlamaIndex_Embeddings implements INode {
 
     async init(nodeData: INodeData, _: string, options: ICommonObject): Promise<any> {
         const timeout = nodeData.inputs?.timeout as string
+        const modelName = nodeData.inputs?.modelName as string
 
         const credentialData = await getCredentialData(nodeData.credential ?? '', options)
         const openAIApiKey = getCredentialParam('openAIApiKey', credentialData, nodeData)
 
         const obj: Partial<OpenAIEmbedding> = {
-            apiKey: openAIApiKey
+            apiKey: openAIApiKey,
+            model: modelName
         }
         if (timeout) obj.timeout = parseInt(timeout, 10)
 

packages/components/nodes/engine/SubQuestionQueryEngine/SubQuestionQueryEngine.ts

@@ -0,0 +1,193 @@
+import { flatten } from 'lodash'
+import { ICommonObject, INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface'
+import {
+    TreeSummarize,
+    SimpleResponseBuilder,
+    Refine,
+    BaseEmbedding,
+    ResponseSynthesizer,
+    CompactAndRefine,
+    QueryEngineTool,
+    LLMQuestionGenerator,
+    SubQuestionQueryEngine,
+    BaseNode,
+    Metadata,
+    serviceContextFromDefaults
+} from 'llamaindex'
+import { reformatSourceDocuments } from '../EngineUtils'
+
+class SubQuestionQueryEngine_LlamaIndex implements INode {
+    label: string
+    name: string
+    version: number
+    description: string
+    type: string
+    icon: string
+    category: string
+    baseClasses: string[]
+    tags: string[]
+    inputs: INodeParams[]
+    outputs: INodeOutputsValue[]
+    sessionId?: string
+
+    constructor(fields?: { sessionId?: string }) {
+        this.label = 'Sub Question Query Engine'
+        this.name = 'subQuestionQueryEngine'
+        this.version = 1.0
+        this.type = 'SubQuestionQueryEngine'
+        this.icon = 'subQueryEngine.svg'
+        this.category = 'Engine'
+        this.description =
+            'Breaks complex query into sub questions for each relevant data source, then gather all the intermediate reponses and synthesizes a final response'
+        this.baseClasses = [this.type]
+        this.tags = ['LlamaIndex']
+        this.inputs = [
+            {
+                label: 'QueryEngine Tools',
+                name: 'queryEngineTools',
+                type: 'QueryEngineTool',
+                list: true
+            },
+            {
+                label: 'Chat Model',
+                name: 'model',
+                type: 'BaseChatModel_LlamaIndex'
+            },
+            {
+                label: 'Embeddings',
+                name: 'embeddings',
+                type: 'BaseEmbedding_LlamaIndex'
+            },
+            {
+                label: 'Response Synthesizer',
+                name: 'responseSynthesizer',
+                type: 'ResponseSynthesizer',
+                description:
+                    'ResponseSynthesizer is responsible for sending the query, nodes, and prompt templates to the LLM to generate a response. See <a target="_blank" href="https://ts.llamaindex.ai/modules/low_level/response_synthesizer">more</a>',
+                optional: true
+            },
+            {
+                label: 'Return Source Documents',
+                name: 'returnSourceDocuments',
+                type: 'boolean',
+                optional: true
+            }
+        ]
+        this.sessionId = fields?.sessionId
+    }
+
+    async init(): Promise<any> {
+        return null
+    }
+
+    async run(nodeData: INodeData, input: string, options: ICommonObject): Promise<string | object> {
+        const returnSourceDocuments = nodeData.inputs?.returnSourceDocuments as boolean
+        const embeddings = nodeData.inputs?.embeddings as BaseEmbedding
+        const model = nodeData.inputs?.model
+
+        const serviceContext = serviceContextFromDefaults({
+            llm: model,
+            embedModel: embeddings
+        })
+
+        let queryEngineTools = nodeData.inputs?.queryEngineTools as QueryEngineTool[]
+        queryEngineTools = flatten(queryEngineTools)
+
+        let queryEngine = SubQuestionQueryEngine.fromDefaults({
+            serviceContext,
+            queryEngineTools,
+            questionGen: new LLMQuestionGenerator({ llm: model })
+        })
+
+        const responseSynthesizerObj = nodeData.inputs?.responseSynthesizer
+        if (responseSynthesizerObj) {
+            if (responseSynthesizerObj.type === 'TreeSummarize') {
+                const responseSynthesizer = new ResponseSynthesizer({
+                    responseBuilder: new TreeSummarize(serviceContext, responseSynthesizerObj.textQAPromptTemplate),
+                    serviceContext
+                })
+                queryEngine = SubQuestionQueryEngine.fromDefaults({
+                    responseSynthesizer,
+                    serviceContext,
+                    queryEngineTools,
+                    questionGen: new LLMQuestionGenerator({ llm: model })
+                })
+            } else if (responseSynthesizerObj.type === 'CompactAndRefine') {
+                const responseSynthesizer = new ResponseSynthesizer({
+                    responseBuilder: new CompactAndRefine(
+                        serviceContext,
+                        responseSynthesizerObj.textQAPromptTemplate,
+                        responseSynthesizerObj.refinePromptTemplate
+                    ),
+                    serviceContext
+                })
+                queryEngine = SubQuestionQueryEngine.fromDefaults({
+                    responseSynthesizer,
+                    serviceContext,
+                    queryEngineTools,
+                    questionGen: new LLMQuestionGenerator({ llm: model })
+                })
+            } else if (responseSynthesizerObj.type === 'Refine') {
+                const responseSynthesizer = new ResponseSynthesizer({
+                    responseBuilder: new Refine(
+                        serviceContext,
+                        responseSynthesizerObj.textQAPromptTemplate,
+                        responseSynthesizerObj.refinePromptTemplate
+                    ),
+                    serviceContext
+                })
+                queryEngine = SubQuestionQueryEngine.fromDefaults({
+                    responseSynthesizer,
+                    serviceContext,
+                    queryEngineTools,
+                    questionGen: new LLMQuestionGenerator({ llm: model })
+                })
+            } else if (responseSynthesizerObj.type === 'SimpleResponseBuilder') {
+                const responseSynthesizer = new ResponseSynthesizer({
+                    responseBuilder: new SimpleResponseBuilder(serviceContext),
+                    serviceContext
+                })
+                queryEngine = SubQuestionQueryEngine.fromDefaults({
+                    responseSynthesizer,
+                    serviceContext,
+                    queryEngineTools,
+                    questionGen: new LLMQuestionGenerator({ llm: model })
+                })
+            }
+        }
+
+        let text = ''
+        let sourceDocuments: ICommonObject[] = []
+        let sourceNodes: BaseNode<Metadata>[] = []
+        let isStreamingStarted = false
+        const isStreamingEnabled = options.socketIO && options.socketIOClientId
+
+        if (isStreamingEnabled) {
+            const stream = await queryEngine.query({ query: input, stream: true })
+            for await (const chunk of stream) {
+                text += chunk.response
+                if (chunk.sourceNodes) sourceNodes = chunk.sourceNodes
+                if (!isStreamingStarted) {
+                    isStreamingStarted = true
+                    options.socketIO.to(options.socketIOClientId).emit('start', chunk.response)
+                }
+
+                options.socketIO.to(options.socketIOClientId).emit('token', chunk.response)
+            }
+
+            if (returnSourceDocuments) {
+                sourceDocuments = reformatSourceDocuments(sourceNodes)
+                options.socketIO.to(options.socketIOClientId).emit('sourceDocuments', sourceDocuments)
+            }
+        } else {
+            const response = await queryEngine.query({ query: input })
+            text = response?.response
+            sourceDocuments = reformatSourceDocuments(response?.sourceNodes ?? [])
+        }
+
+        if (returnSourceDocuments) return { text, sourceDocuments }
+        else return { text }
+    }
+}
+
+module.exports = { nodeClass: SubQuestionQueryEngine_LlamaIndex }

packages/components/nodes/engine/SubQuestionQueryEngine/subQueryEngine.svg

@@ -0,0 +1 @@
+<svg xmlns="http://www.w3.org/2000/svg" class="icon icon-tabler icon-tabler-filter-question" 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="M15 19l-6 2v-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 4.414" /><path d="M19 22v.01" /><path d="M19 19a2.003 2.003 0 0 0 .914 -3.782a1.98 1.98 0 0 0 -2.414 .483" /></svg>

packages/components/nodes/tools/QueryEngineTool/QueryEngineTool.ts

@@ -0,0 +1,68 @@
+import { INode, INodeData, INodeParams } from '../../../src/Interface'
+import { VectorStoreIndex } from 'llamaindex'
+
+class QueryEngine_Tools implements INode {
+    label: string
+    name: string
+    version: number
+    description: string
+    type: string
+    icon: string
+    category: string
+    tags: string[]
+    baseClasses: string[]
+    inputs?: INodeParams[]
+
+    constructor() {
+        this.label = 'QueryEngine Tool'
+        this.name = 'queryEngineToolLlamaIndex'
+        this.version = 1.0
+        this.type = 'QueryEngineTool'
+        this.icon = 'queryEngineTool.svg'
+        this.category = 'Tools'
+        this.tags = ['LlamaIndex']
+        this.description = 'Tool used to invoke query engine'
+        this.baseClasses = [this.type]
+        this.inputs = [
+            {
+                label: 'Vector Store Index',
+                name: 'vectorStoreIndex',
+                type: 'VectorStoreIndex'
+            },
+            {
+                label: 'Tool Name',
+                name: 'toolName',
+                type: 'string',
+                description: 'Tool name must be small capital letter with underscore. Ex: my_tool'
+            },
+            {
+                label: 'Tool Description',
+                name: 'toolDesc',
+                type: 'string',
+                rows: 4
+            }
+        ]
+    }
+
+    async init(nodeData: INodeData): Promise<any> {
+        const vectorStoreIndex = nodeData.inputs?.vectorStoreIndex as VectorStoreIndex
+        const toolName = nodeData.inputs?.toolName as string
+        const toolDesc = nodeData.inputs?.toolDesc as string
+        const queryEngineTool = {
+            queryEngine: vectorStoreIndex.asQueryEngine({
+                preFilters: {
+                    ...(vectorStoreIndex as any).metadatafilter
+                }
+            }),
+            metadata: {
+                name: toolName,
+                description: toolDesc
+            },
+            vectorStoreIndex
+        }
+
+        return queryEngineTool
+    }
+}
+
+module.exports = { nodeClass: QueryEngine_Tools }

packages/components/nodes/tools/QueryEngineTool/queryEngineTool.svg

@@ -0,0 +1 @@
+<svg xmlns="http://www.w3.org/2000/svg" class="icon icon-tabler icon-tabler-brand-google-big-query" 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="M17.73 19.875a2.225 2.225 0 0 1 -1.948 1.125h-7.283a2.222 2.222 0 0 1 -1.947 -1.158l-4.272 -6.75a2.269 2.269 0 0 1 0 -2.184l4.272 -6.75a2.225 2.225 0 0 1 1.946 -1.158h7.285c.809 0 1.554 .443 1.947 1.158l3.98 6.75a2.33 2.33 0 0 1 0 2.25l-3.98 6.75v-.033z" /><path d="M11.5 11.5m-3.5 0a3.5 3.5 0 1 0 7 0a3.5 3.5 0 1 0 -7 0" /><path d="M14 14l2 2" /></svg>

packages/components/nodes/vectorstores/Pinecone/Pinecone_LlamaIndex.ts

@@ -13,7 +13,7 @@ import {
 import { FetchResponse, Index, Pinecone, ScoredPineconeRecord } from '@pinecone-database/pinecone'
 import { flatten } from 'lodash'
 import { Document as LCDocument } from 'langchain/document'
-import { ICommonObject, INode, INodeData, INodeParams } from '../../../src/Interface'
+import { ICommonObject, INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface'
 import { flattenObject, getCredentialData, getCredentialParam } from '../../../src/utils'
 
 class PineconeLlamaIndex_VectorStores implements INode {
@@ -28,6 +28,7 @@ class PineconeLlamaIndex_VectorStores implements INode {
     baseClasses: string[]
     inputs: INodeParams[]
     credential: INodeParams
+    outputs: INodeOutputsValue[]
 
     constructor() {
         this.label = 'Pinecone'
@@ -93,6 +94,18 @@ class PineconeLlamaIndex_VectorStores implements INode {
                 optional: true
             }
         ]
+        this.outputs = [
+            {
+                label: 'Pinecone Retriever',
+                name: 'retriever',
+                baseClasses: this.baseClasses
+            },
+            {
+                label: 'Pinecone Vector Store Index',
+                name: 'vectorStore',
+                baseClasses: [this.type, 'VectorStoreIndex']
+            }
+        ]
     }
 
     //@ts-ignore
@@ -155,8 +168,10 @@ class PineconeLlamaIndex_VectorStores implements INode {
         }
 
         if (pineconeNamespace) obj.namespace = pineconeNamespace
+
+        let metadatafilter = {}
         if (pineconeMetadataFilter) {
-            const metadatafilter = typeof pineconeMetadataFilter === 'object' ? pineconeMetadataFilter : JSON.parse(pineconeMetadataFilter)
+            metadatafilter = typeof pineconeMetadataFilter === 'object' ? pineconeMetadataFilter : JSON.parse(pineconeMetadataFilter)
             obj.queryFilter = metadatafilter
         }
 
@@ -171,11 +186,21 @@ class PineconeLlamaIndex_VectorStores implements INode {
             serviceContext
         })
 
-        const retriever = index.asRetriever()
+        const output = nodeData.outputs?.output as string
-        retriever.similarityTopK = k
-        ;(retriever as any).serviceContext = serviceContext
 
-        return retriever
+        if (output === 'retriever') {
+            const retriever = index.asRetriever()
+            retriever.similarityTopK = k
+            ;(retriever as any).serviceContext = serviceContext
+            return retriever
+        } else if (output === 'vectorStore') {
+            ;(index as any).k = k
+            if (metadatafilter) {
+                ;(index as any).metadatafilter = metadatafilter
+            }
+            return index
+        }
+        return index
     }
 }
 

packages/components/nodes/vectorstores/SimpleStore/SimpleStore.ts

@@ -63,6 +63,18 @@ class SimpleStoreUpsert_LlamaIndex_VectorStores implements INode {
                 optional: true
             }
         ]
+        this.outputs = [
+            {
+                label: 'SimpleStore Retriever',
+                name: 'retriever',
+                baseClasses: this.baseClasses
+            },
+            {
+                label: 'SimpleStore Vector Store Index',
+                name: 'vectorStore',
+                baseClasses: [this.type, 'VectorStoreIndex']
+            }
+        ]
     }
 
     //@ts-ignore
@@ -114,10 +126,19 @@ class SimpleStoreUpsert_LlamaIndex_VectorStores implements INode {
         const storageContext = await storageContextFromDefaults({ persistDir: filePath })
 
         const index = await VectorStoreIndex.init({ storageContext, serviceContext })
-        const retriever = index.asRetriever()
-        retriever.similarityTopK = k
 
-        return retriever
+        const output = nodeData.outputs?.output as string
+
+        if (output === 'retriever') {
+            const retriever = index.asRetriever()
+            retriever.similarityTopK = k
+            ;(retriever as any).serviceContext = serviceContext
+            return retriever
+        } else if (output === 'vectorStore') {
+            ;(index as any).k = k
+            return index
+        }
+        return index
     }
 }
 

packages/server/marketplaces/chatflows/Context Chat Engine.json

@@ -181,6 +181,28 @@
                         "credentialNames": ["openAIApi"],
                         "id": "openAIEmbedding_LlamaIndex_0-input-credential-credential"
                     },
+                    {
+                        "label": "Model Name",
+                        "name": "modelName",
+                        "type": "options",
+                        "options": [
+                            {
+                                "label": "text-embedding-3-large",
+                                "name": "text-embedding-3-large"
+                            },
+                            {
+                                "label": "text-embedding-3-small",
+                                "name": "text-embedding-3-small"
+                            },
+                            {
+                                "label": "text-embedding-ada-002",
+                                "name": "text-embedding-ada-002"
+                            }
+                        ],
+                        "default": "text-embedding-ada-002",
+                        "optional": true,
+                        "id": "openAIEmbedding_LlamaIndex_0-input-modelName-options"
+                    },
                     {
                         "label": "Timeout",
                         "name": "timeout",
@@ -315,13 +337,29 @@
                 },
                 "outputAnchors": [
                     {
-                        "id": "pineconeLlamaIndex_0-output-pineconeLlamaIndex-Pinecone|VectorIndexRetriever",
+                        "name": "output",
-                        "name": "pineconeLlamaIndex",
+                        "label": "Output",
-                        "label": "Pinecone",
+                        "type": "options",
-                        "type": "Pinecone | VectorIndexRetriever"
+                        "options": [
+                            {
+                                "id": "pineconeLlamaIndex_0-output-retriever-Pinecone|VectorIndexRetriever",
+                                "name": "retriever",
+                                "label": "Pinecone Retriever",
+                                "type": "Pinecone | VectorIndexRetriever"
+                            },
+                            {
+                                "id": "pineconeLlamaIndex_0-output-retriever-Pinecone|VectorStoreIndex",
+                                "name": "vectorStore",
+                                "label": "Pinecone Vector Store Index",
+                                "type": "Pinecone | VectorStoreIndex"
+                            }
+                        ],
+                        "default": "retriever"
                     }
                 ],
-                "outputs": {},
+                "outputs": {
+                    "output": "retriever"
+                },
                 "selected": false
             },
             "selected": false,
@@ -367,6 +405,14 @@
                                 "label": "gpt-4",
                                 "name": "gpt-4"
                             },
+                            {
+                                "label": "gpt-4-turbo-preview",
+                                "name": "gpt-4-turbo-preview"
+                            },
+                            {
+                                "label": "gpt-4-0125-preview",
+                                "name": "gpt-4-0125-preview"
+                            },
                             {
                                 "label": "gpt-4-1106-preview",
                                 "name": "gpt-4-1106-preview"
@@ -672,6 +718,14 @@
                                 "label": "gpt-4",
                                 "name": "gpt-4"
                             },
+                            {
+                                "label": "gpt-4-turbo-preview",
+                                "name": "gpt-4-turbo-preview"
+                            },
+                            {
+                                "label": "gpt-4-0125-preview",
+                                "name": "gpt-4-0125-preview"
+                            },
                             {
                                 "label": "gpt-4-1106-preview",
                                 "name": "gpt-4-1106-preview"

packages/server/marketplaces/chatflows/Query Engine.json

@@ -163,13 +163,29 @@
                 },
                 "outputAnchors": [
                     {
-                        "id": "pineconeLlamaIndex_0-output-pineconeLlamaIndex-Pinecone|VectorIndexRetriever",
+                        "name": "output",
-                        "name": "pineconeLlamaIndex",
+                        "label": "Output",
-                        "label": "Pinecone",
+                        "type": "options",
-                        "type": "Pinecone | VectorIndexRetriever"
+                        "options": [
+                            {
+                                "id": "pineconeLlamaIndex_0-output-retriever-Pinecone|VectorIndexRetriever",
+                                "name": "retriever",
+                                "label": "Pinecone Retriever",
+                                "type": "Pinecone | VectorIndexRetriever"
+                            },
+                            {
+                                "id": "pineconeLlamaIndex_0-output-retriever-Pinecone|VectorStoreIndex",
+                                "name": "vectorStore",
+                                "label": "Pinecone Vector Store Index",
+                                "type": "Pinecone | VectorStoreIndex"
+                            }
+                        ],
+                        "default": "retriever"
                     }
                 ],
-                "outputs": {},
+                "outputs": {
+                    "output": "retriever"
+                },
                 "selected": false
             },
             "selected": false,
@@ -206,6 +222,28 @@
                         "credentialNames": ["openAIApi"],
                         "id": "openAIEmbedding_LlamaIndex_0-input-credential-credential"
                     },
+                    {
+                        "label": "Model Name",
+                        "name": "modelName",
+                        "type": "options",
+                        "options": [
+                            {
+                                "label": "text-embedding-3-large",
+                                "name": "text-embedding-3-large"
+                            },
+                            {
+                                "label": "text-embedding-3-small",
+                                "name": "text-embedding-3-small"
+                            },
+                            {
+                                "label": "text-embedding-ada-002",
+                                "name": "text-embedding-ada-002"
+                            }
+                        ],
+                        "default": "text-embedding-ada-002",
+                        "optional": true,
+                        "id": "openAIEmbedding_LlamaIndex_0-input-modelName-options"
+                    },
                     {
                         "label": "Timeout",
                         "name": "timeout",

packages/server/src/utils/index.ts

@@ -844,7 +844,7 @@ export const isFlowValidForStream = (reactFlowNodes: IReactFlowNode[], endingNod
         const whitelistAgents = ['openAIFunctionAgent', 'csvAgent', 'airtableAgent', 'conversationalRetrievalAgent']
         isValidChainOrAgent = whitelistAgents.includes(endingNodeData.name)
     } else if (endingNodeData.category === 'Engine') {
-        const whitelistEngine = ['contextChatEngine', 'simpleChatEngine', 'queryEngine']
+        const whitelistEngine = ['contextChatEngine', 'simpleChatEngine', 'queryEngine', 'subQuestionQueryEngine']
         isValidChainOrAgent = whitelistEngine.includes(endingNodeData.name)
     }