fix: upload document mid-thread does not work (#2504)

extensions/assistant-extension/src/node/index.ts

@@ -1,5 +1,5 @@
 import { getJanDataFolderPath, normalizeFilePath } from '@janhq/core/node'
-import { retrieval } from './tools/retrieval'
+import { retrieval } from './retrieval'
 import path from 'path'
 
 export function toolRetrievalUpdateTextSplitter(

extensions/assistant-extension/src/node/retrieval.ts

@@ -5,7 +5,7 @@ import { PDFLoader } from 'langchain/document_loaders/fs/pdf'
 import { HNSWLib } from 'langchain/vectorstores/hnswlib'
 
 import { OpenAIEmbeddings } from 'langchain/embeddings/openai'
-import { readEmbeddingEngine } from '../../engine'
+import { readEmbeddingEngine } from './engine'
 
 export class Retrieval {
   public chunkSize: number = 100

extensions/assistant-extension/src/tools/retrieval.ts

@@ -18,7 +18,7 @@ export class RetrievalTool extends InferenceTool {
     tool?: AssistantTool
   ): Promise<MessageRequest> {
     if (!data.model || !data.messages) {
-      return Promise.resolve(data)
+      return Promise.resolve(this.normalize(data))
     }
 
     const latestMessage = data.messages[data.messages.length - 1]
@@ -48,7 +48,7 @@ export class RetrievalTool extends InferenceTool {
     ) {
       // No document ingested, reroute the result to inference engine
 
-      return Promise.resolve(data)
+      return Promise.resolve(this.normalize(data))
     }
     // 2. Load agent on thread changed
     if (this.retrievalThreadId !== data.threadId) {
@@ -87,8 +87,14 @@ export class RetrievalTool extends InferenceTool {
             .replace('{QUESTION}', prompt)
     }
 
-    // Filter out all the messages that are not text
-    data.messages = data.messages.map((message) => {
+    // 4. Reroute the result to inference engine
+    return Promise.resolve(this.normalize(data))
+  }
+
+  // Filter out all the messages that are not text
+  // TODO: Remove it until engines can handle multiple content types
+  normalize(request: MessageRequest): MessageRequest {
+    request.messages = request.messages?.map((message) => {
       if (
         message.content &&
         typeof message.content !== 'string' &&
@@ -101,8 +107,6 @@ export class RetrievalTool extends InferenceTool {
       }
       return message
     })
-
-    // 4. Reroute the result to inference engine
-    return Promise.resolve(data)
+    return request
   }
 }

extensions/model-extension/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@janhq/model-extension",
-  "version": "1.0.27",
+  "version": "1.0.28",
   "description": "Model Management Extension provides model exploration and seamless downloads",
   "main": "dist/index.js",
   "module": "dist/module.js",

models/groq-llama2-70b/model.json

@@ -10,7 +10,9 @@
   "version": "1.0",
   "description": "Groq Llama 2 70b with supercharged speed!",
   "format": "api",
-  "settings": {},
+  "settings": {
+    "text_model": false
+  },
   "parameters": {
     "max_tokens": 4096,
     "temperature": 0.7,

models/groq-mixtral-8x7b-instruct/model.json

@@ -10,7 +10,9 @@
   "version": "1.0",
   "description": "Groq Mixtral 8x7b Instruct is Mixtral with supercharged speed!",
   "format": "api",
-  "settings": {},
+  "settings": {
+    "text_model": false
+  },
   "parameters": {
     "max_tokens": 4096,
     "temperature": 0.7,

web/utils/messageRequestBuilder.ts

@@ -91,7 +91,7 @@ export class MessageRequestBuilder {
         },
       ] as ChatCompletionMessageContent,
     }
-    this.messages = [message, ...this.messages]
+    this.messages = [...this.messages, message]
     return this
   }
 
@@ -113,7 +113,7 @@ export class MessageRequestBuilder {
       ] as ChatCompletionMessageContent,
     }
 
-    this.messages = [message, ...this.messages]
+    this.messages = [...this.messages, message]
     return this
   }