Hello, I’m using the Pinecone vector store. However, it’s not working as expected and not retrieving the matching documents from the store, even though I have the matching query documents in the store.
import { StreamingTextResponse, LangChainAdapter } from 'ai';
import { ChatOpenAI } from '@langchain/openai';
import {
RunnablePassthrough,
RunnableSequence,
} from '@langchain/core/runnables';
import { formatDocumentsAsString } from 'langchain/util/document';
import {
ANSWER_GENERATION_CHAIN_PROMPT,
EMBEDDING_MODEL,
STANDALONE_QUESTION_PROMPT,
} from '@/constants/chat';
import { formatMessage } from '@/lib/utils';
import { AgentExecutor, createOpenAIToolsAgent } from 'langchain/agents';
import { tools } from './engine/tools';
import { StringOutputParser } from '@langchain/core/output_parsers';
import { PineconeStore } from '@langchain/pinecone';
export const dynamic = 'force-dynamic';
export async function POST(req: Request) {
try {
const body = await req.json();
const { messages, nameSpace } = body;
const formattedPreviousMessages = messages.slice(0, -1).map(formatMessage);
const currentMessageContent = messages[messages.length - 1].content;
const vectorStore = await PineconeStore.fromExistingIndex(EMBEDDING_MODEL, {
pineconeConfig: {
config: {
apiKey: process.env.PINECONE_API_KEY!,
},
indexName: process.env.PINECONE_INDEX!,
namespace: nameSpace,
},
onFailedAttempt: (error) => {
console.error('Failed attempt', error);
},
});
// Standalone question chain to retrieve to best documents
const standaloneQuestionChain = RunnableSequence.from([
RunnablePassthrough.assign({
conversation_history: (input) => input.chat_history,
question: (input) => input.question,
}),
STANDALONE_QUESTION_PROMPT,
new ChatOpenAI({
apiKey: process.env.OPENAI_API_KEY!,
model: 'gpt-4o-mini',
temperature: 0,
verbose: true,
}),
new StringOutputParser(),
]);
// Document retrieval chain
const retriever = vectorStore.asRetriever();
const documentRetrievalChain = RunnableSequence.from([
// get the question from the input
(input) => input.question,
// retrieve the document from the vector store using the question
retriever,
// format the retrieved document as a string
formatDocumentsAsString,
]);
// Create the LLM model
const llmModel = new ChatOpenAI({
apiKey: process.env.OPENAI_API_KEY!,
model: 'gpt-4o-mini',
// set the temperature to 0 to get deterministic results
temperature: 0,
streaming: true,
verbose: true,
});
// Create the OpenAI tools agent
const openAIToolsAgent = await createOpenAIToolsAgent({
llm: llmModel,
tools,
prompt: ANSWER_GENERATION_CHAIN_PROMPT,
});
// Create the agent executor
const openAIToolsAgentExecutor = new AgentExecutor({
agent: openAIToolsAgent,
tools,
});
const retrievalChain = RunnableSequence.from([
RunnablePassthrough.assign({
question: standaloneQuestionChain,
chat_history: (input) => input.chat_history,
original_message: (input) => input.question,
}),
{
context: documentRetrievalChain,
chat_history: (input) => input.chat_history,
question: (input) => input.original_message,
},
openAIToolsAgentExecutor,
{
content: (input) => {
return input.output;
},
},
]);
// Execute the retrieval chain
const stream = await retrievalChain.stream({
question: currentMessageContent,
chat_history: formattedPreviousMessages,
});
const aiStream = LangChainAdapter.toAIStream(stream);
// Respond with the stream
return new StreamingTextResponse(aiStream);
} catch (e: any) {
return Response.json({ error: e.message }, { status: e.status ?? 500 });
}
}
Query: What are stars and forks count?
