This project demonstrates the integration of OpenAI's Assistants API into a React/Next.js application. The application showcases a chat interface where users can interact with the AI assistant. This demo serves as an example of how OpenAI's powerful AI models can be leveraged within a web application to create interactive and engaging user experiences.
Users can send queries and receive responses from the AI assistant in real-time.
Utilizes Next.js API routes (Edge Functions) for server-side processing and interaction with OpenAI's Assistants API.
Messages are dynamically displayed in a chat-like interface, with different alignments and styles for user and assistant messages.
- Nx: For management of the monorepo.
- React/Next.js: For the front-end interface and server-side logic.
- OpenAI API: Specifically the Assistants API for generating responses based on user queries.
git clone git@github.com:mandarini/openai-assistant-demo.gitpnpm install- Create a
.envfile in theapps/assistantdirectory. - Add NX_OPENAI_KEY with your OpenAI API key.
- Add NX_ASSISTANT_ID with the ID of your OpenAI Assistant.
The createAssistant function demonstrates a powerful feature of the OpenAI API, allowing the creation of a custom AI assistant tailored to specific needs. This function is particularly useful for integrating a set of documentation files into the assistant's knowledge base, enabling it to provide accurate and source-cited responses based on the provided documentation. You can read more about how the Assistants API with Knowledge Retrieval works in the OpenAI Docs.
The function initializes by setting up the OpenAI client and creating a new assistant.
It reads a directory containing documentation files (in this case, related to Nx Dev Tools). Each file in the directory is uploaded to OpenAI's servers.
After uploading the documents, it creates a new assistant with a specific set of instructions. This assistant is configured to use the uploaded files as its primary knowledge base, ensuring that all responses are based on the provided documentation.
This function can be executed to create an assistant that is particularly suited for applications where responses need to be based on a controlled set of documents, like internal tools documentation, FAQs, or policy manuals. By uploading relevant documents, the assistant becomes a specialized resource, providing responses that are consistent with the organization's information and guidelines.
You can call this function with the following command:
npx nx run create-assistant:run-nodeand this will upload the files in the docs directory and create a new assistant with the all the files id's. You can then use the assistant ID to interact with the assistant in your application.
You can find all your uploaded files here: https://platform.openai.com/files
There, you can also upload new files and delete existing ones, using the GUI. You can read more about how to use the File API here: https://platform.openai.com/docs/api-reference/files.
npx nx serve assistant- Open the application in your browser.
- Type your Nx-related query in the input box and press 'Send' or hit Enter.
- View the AI assistant's response in the chat interface.
apps/assistant/app: Contains the React/Next.js pages including the main chat interface.apps/assistant/pages/api/chat: Includes the Edge Function for handling API requests and interactions with the OpenAI Assistant. Some of the reasons why we're using an edge function to handle the OpenAI calls is for security reasons (keep API key hidden) and to avoid CORS issues.libs/utils: Utility functions for API interactions and error handling.
- Text Streaming is not yet supported by the API so the responses are super slow
- We should show sources for the AI responses
- Improvement in error handling and user feedback mechanisms should be considered
The waitForRunCompletion function is crucial for handling the asynchronous nature of interactions with the OpenAI API, specifically when waiting for the run task to complete. The waitForRunCompletion function is designed to periodically check the status of a run and retrieve results once the run status is completed.
The function implements a polling mechanism where it periodically checks the status of a run by retrieving the Run object from the OpenAI API.
For each polling iteration, the function checks if the run has completed, failed, been cancelled, or expired.
Once the run is completed, the function fetches the results, which include both the final status of the run and the messages associated with the thread.
In case of failure, cancellation, expiration, or timeout, the function handles these scenarios by rejecting the promise with an appropriate error message.
As per the current OpenAI API documentation, there's no streaming support for run updates. Therefore, applications need to implement a polling mechanism to stay updated with the run status. waitForRunCompletion functionally addresses this need by providing a reliable method to wait for a run's completion and handle various statuses of the run.
OpenAI plans to introduce support for streaming in the future, which will simplify the process of tracking run status. However, until such features are implemented, this polling mechanism remains an essential part of interacting with the OpenAI API for asynchronous tasks.
Contributions to enhance this demo are welcome. Please follow the standard fork, branch, and pull request workflow.
MIT