Pinecone Assistant

Next.js sample chat app that connects to a Pinecone Assistant for grounded answers over uploaded PDFs with citations and file references.

Github Open in a new window

Built with

Pinecone Assistant API
Next.js + tailwind
Node version 20 or higher

Run the sample app

The fastest way to get started is to use the create-pinecone-app CLI tool to get up and running:

Bash

npx -y create-pinecone-app@latest --template pinecone-assistant

Get your API key

You need an API key to make API calls to your Pinecone project:

Then copy your generated key:

PINECONE_API_KEY="{{YOUR_API_KEY}}"

Alternatively, follow these steps:

Open the Pinecone console.
Select your project.
Go to API Keys.
Copy your API key.

Create a Pinecone Assistant

You can create a Pinecone Assistant in the console, or by following the instructions here.

Start the project

Requires Node version 20+

Dependency installation

From the project root directory, run the following command:

Bash

cd pinecone-assistant && npm install

Make sure you have populated the .env file with relevant keys:

Bash

PINECONE_API_KEY="your-pinecone-api-key-here"
PINECONE_ASSISTANT_NAME="your-pinecone-assistant-name-here"
# Set this if you want users chatting with your assistant to be able to see
# and click into the files used as references in answers
SHOW_ASSISTANT_FILES=true

Start the app:

Bash

npm run dev

Project structure

This project uses a standard Next.js application structure with API routes for backend functionality.

Frontend client

The frontend uses Next.js, Tailwind CSS, and custom React components to power the chat interface.

Backend server

This project uses Next.js API routes to proxy requests to the Pinecone Assistant API.

Pinecone Assistant

Key features

Connect to existing Pinecone Assistant: Connect to an existing Pinecone Assistant to provide a chat experience that can be hosted privately or publicly.
Streaming responses: Ask questions of the assistant and get responses streamed to the frontend in real-time.
Reference highlighting: Documents that were used in answering user questions are highlighted as references.

Implementation details

Server action for chat

The server action creates a stream with Pinecone Assistants:

TypeScript

'use server'

import { createStreamableValue } from 'ai/rsc'
import { EventSource } from 'extended-eventsource';

type Message = {
  role: string;
  content: string;
}

export async function chat(messages: Message[]) {

  // Create an initial stream, which we'll populate with events from the Pinecone Assistants API
  const stream = createStreamableValue()

  // Construct the full URL to the Pinecone Assistant API for the specific assistant
  // indicated by the user
  const url = `${process.env.PINECONE_ASSISTANT_URL}/${process.env.PINECONE_ASSISTANT_NAME}/chat/completions`

  const eventSource = new EventSource(url, {
    method: 'POST',
    body: JSON.stringify({
      stream: true,
      messages,
    }),
    headers: {
      Authorization: `Bearer ${process.env.PINECONE_API_KEY}`,
      'X-Project-Id': process.env.PINECONE_ASSISTANT_ID!,
    },
    disableRetry: true,
  });

  // When we receive a new message from the Pinecone Assistant API, we update the stream
  // unless the Assistant is done, in which case we close the stream
  eventSource.onmessage = (event: MessageEvent) => {
    const message = JSON.parse(event.data)
    if (message?.choices[0]?.finish_reason) {
      eventSource.close();
      stream.done();
    } else {
      stream.update(event.data)
    }
  };

  eventSource.onerror = (error) => {
    console.error('EventSource error:', error);
    eventSource.close();
  };

  return { object: stream.value }
}

Chat functionality

The chat functionality in the Home component consumes the stream from the server action and updates the UI in real-time:

TypeScript

const handleChat = async () => {
    if (!input.trim()) return;

    const newUserMessage: Message = {
      id: uuidv4(), // Generate a unique ID
      role: 'user',
      content: input,
      timestamp: new Date().toISOString()
    };

    setMessages(prevMessages => [...prevMessages, newUserMessage]);
    setInput('');
    setIsStreaming(true);

    try {
      const { object } = await chat([newUserMessage]);
      let accumulatedContent = '';
      const newAssistantMessage: Message = {
        id: uuidv4(),
        role: 'assistant',
        content: '',
        timestamp: new Date().toISOString(),
        references: []
      };

      setMessages(prevMessages => [...prevMessages, newAssistantMessage]);

      // Process the response stream from the Assistant that is created in the ./actions.ts Server action
      for await (const chunk of readStreamableValue(object)) {
        try {
          const data = JSON.parse(chunk);
          const content = data.choices[0]?.delta?.content;

          if (content) {
            accumulatedContent += content;
          }

          setMessages(prevMessages => {
            const updatedMessages = [...prevMessages];
            const lastMessage = updatedMessages[updatedMessages.length - 1];
            lastMessage.content = accumulatedContent;
            return updatedMessages;
          });

        } catch (error) {
          console.error('Error parsing chunk:', error);
        }
      }

      // Extract references after the full message is received
      const extractedReferences = extractReferences(accumulatedContent);
      setReferencedFiles(extractedReferences);

    } catch (error) {
      console.error('Error in chat:', error);
      setError('An error occurred while chatting.');
    } finally {
      setIsStreaming(false);
    }
  };

Troubleshooting

Experiencing any issues with the sample app? Submit an issue, create a PR, or post in our community forum!