Iterate

api/src/controllers/discovery.js

@@ -69,7 +69,7 @@ export const chat = async (req, res, next) => {
     });
   } catch (error) {
     // Handle the exception
-    console.error("An error occurred:", error);
+    console.error("An error occurred:", error.message);
     res.status(500).json({ error: "Internal Server Error" });
   }
 };

api/src/libs/indexer.js

@@ -261,14 +261,7 @@ class Indexer {
     const chatRequest = {
       indexIds,
       input: {
-        question: `
-          Determine if the following information is relevant to the previous conversation.
-          If it is relevant, output a conversation simulating that you received real-time news for the user.
-          Use conversational output format suitable to data model, use bold texts and links when available.
-          Do not mention relevancy check, just share it as an update.
-          If it is not relevant, simply say "NOT_RELEVANT".
-          Information: ${JSON.stringify(item)}
-          `,
+        information: JSON.stringify(item),
         chat_history: messages,
       },
     };

indexer/src/app/modules/agent.module.ts

@@ -1,6 +1,7 @@
 import { ChatOpenAI, OpenAI, OpenAIEmbeddings } from '@langchain/openai';
 import { loadSummarizationChain } from 'langchain/chains';
 import {
+  RunnableConfig,
   RunnableLambda,
   RunnablePassthrough,
   RunnableSequence,
@@ -42,10 +43,11 @@ export class Agent {
     indexIds: string[],
     model_type: string = 'OpenAI',
     model_args: any,
+    prompt: any,
   ): Promise<any> {
     switch (chain_type) {
       case 'rag-v0':
-        return this.createRAGChain(indexIds, model_type, model_args);
+        return this.createRAGChain(indexIds, model_type, model_args, prompt);
 
       default:
         throw new Error('Chain type not supported');
@@ -133,7 +135,9 @@ export class Agent {
     chroma_indices: string[],
     model_type: string,
     model_args: any = { temperature: 0.0, max_tokens: 1000, max_retries: 4 },
+    prompt: any,
   ): Promise<any> {
+    console.log(prompt);
     // TODO: Add prior filtering for questions such as "What is new today?" (with date filter)
     // TODO: Add self-ask prompt for fact-checking
     const argv = model_args ?? {
@@ -181,8 +185,6 @@ export class Agent {
 
     const retriever = vectorStore.asRetriever();
 
-    const answerPrompt = await pull(process.env.PROMPT_ANSWER_TAG);
-
     const formatChatHistory = (chatHistory: string | string[]) => {
       if (Array.isArray(chatHistory)) {
         const updatedChat = chatHistory
@@ -206,7 +208,9 @@ export class Agent {
         context: RunnableSequence.from([
           {
             docs: async (input) => {
-              const docs = await retriever.getRelevantDocuments(input.question);
+              const docs = await retriever.getRelevantDocuments(
+                input.question || input.information,
+              );
               return docs;
             },
           },
@@ -219,8 +223,10 @@ export class Agent {
           },
         ]),
         question: (input) => {
-          Logger.log(input.question, 'ChatService:answerChain:inputQuestion');
-          return input.question;
+          if (input.question) {
+            Logger.log(input.question, 'ChatService:answerChain:inputQuestion');
+            return input.question;
+          }
         },
         chat_history: (input) => input.chat_history,
       },
@@ -235,16 +241,27 @@ export class Agent {
               return serialized;
             },
             question: (input) => {
-              Logger.log(
-                input.question,
-                'ChatService:answerChain:inputQuestion',
-              );
-              return input.question;
+              if (input.question) {
+                Logger.log(
+                  input.question,
+                  'ChatService:answerChain:inputQuestion',
+                );
+                return input.question;
+              }
+            },
+            information: (input) => {
+              if (input.information) {
+                Logger.log(
+                  input.information,
+                  'ChatService:answerChain:inputInformation',
+                );
+                return input.inormation;
+              }
             },
-            chat_history: (input) => formatChatHistory(input.chat_history),
             relation: (input) => '',
+            chat_history: (input) => formatChatHistory(input.chat_history),
           },
-          answerPrompt as any,
+          prompt as any,
           model,
           new StringOutputParser(),
         ]),

indexer/src/chat/schema/chat.schema.ts

@@ -76,7 +76,7 @@ export class RetrievalQuestionInput {
       chat_history: [],
     },
   })
-  input: { question: string; chat_history: [] };
+  input: { question?: string; information?: string; chat_history: [] };
 
   @ApiPropertyOptional({
     description: 'Model arguments',

indexer/src/chat/service/chat.service.ts

@@ -16,6 +16,7 @@ import {
 import { Agent } from 'src/app/modules/agent.module';
 import { RunnableSequence } from '@langchain/core/runnables';
 import { OpenAIEmbeddings } from '@langchain/openai';
+import { pull } from 'langchain/hub';
 
 @Injectable()
 export class ChatService {
@@ -37,20 +38,33 @@ export class ChatService {
 
     try {
       // Initialize the agent
+
+      const answerPrompt = await pull(
+        body.input.question
+          ? process.env.PROMPT_ANSWER_TAG
+          : process.env.PROMPT_RELEVANCY_CHECK_TAG,
+      );
       const chain: RunnableSequence = await this.agentClient.createAgentChain(
         body.chain_type,
         body.indexIds,
         body.model_type,
         body?.model_args,
+        answerPrompt,
       );
-
+      if (body.input.question) {
+        const stream = await chain.stream({
+          question: body.input.question,
+          chat_history: body.input.chat_history,
+        });
+        return stream;
+      } else if (body.input.information) {
+        const stream = await chain.stream({
+          information: body.input.information,
+          chat_history: body.input.chat_history,
+        });
+        return stream;
+      }
       // Invoke the agent
-      const stream = await chain.stream({
-        question: body.input.question,
-        chat_history: body.input.chat_history,
-      });
-
-      return stream;
     } catch (e) {
       Logger.log(
         `Cannot process ${body.input.question} ${e}`,

web-app/src/components/site/indexes/AskIndexes/index.tsx

@@ -182,7 +182,8 @@ const AskIndexes: FC<AskIndexesProps> = ({ chatID, sources }) => {
     return () => {
       ws.close();
     };
-  }, [chatID, setMessages, messages]);
+    // eslint-disable-next-line
+  }, [chatID]);
   if (leftSectionIndexes.length === 0) {
     return <NoIndexes tabKey={leftTabKey} />;
   }