articleschannelstagsspacestoolkit
Microsoft Dev Blogs

OpenAI Assistants

View Original
thumbnail

OpenAI Dev Day 2023: Introduction to Assistants API

  • OpenAI held their first Dev Day on November 6th, 2023, where they made several product announcements.
  • One of the key announcements was the introduction of the Assistants API, which simplifies the process of building chat-based applications.
  • In this blog post, we will focus on the Assistants API and compare it to the previously used Chat API in the Android JetchatAI sample app.

Key Features of Assistants API

  • Stateful: The history of the conversation is stored on the server, making it easier for the client app to track the thread identifier and send new messages.
  • Sliding window: The API automatically manages the context window of the model, reducing the need for manual management.
  • Code interpreter: The API allows for the generation and execution of Python code to answer questions, making it more versatile and powerful.

Implementing an Assistant with the Assistants API

  • The Assistants overview provides a detailed guide on how to implement an Assistant using the API in Python.
  • The blog post showcases an example of an Assistant prototype built in the Playground that can answer questions based on documentation sources.
  • The prototype demonstrates how the Assistant can provide responses grounded in uploaded documents, with relevant citations.

Comparison to Chat API

  • The Assistants API simplifies the process of building chat-based applications compared to the previously used Chat API.
  • With the Assistants API, the client code only needs to keep track of the thread identifier and exchange queries and responses with the server.
  • Tasks such as managing the conversation history, context window, chunking source data, and prompt tweaking are handled by the API.

Future with Kotlin

  • While the Assistants API is still in beta, it is expected that the implementation of Assistants in Kotlin will be much simpler compared to the Chat API.
  • Instead of complex code for conversation management, the Kotlin code will focus on exchanging queries and responses with the server.
  • The OpenAI Kotlin open-source client library is currently implementing the Assistants API, with progress visible in a pull request.

Resources and Feedback

  • More details on the OpenAI Dev Day announcements can be found on the OpenAI blog.
  • The Assistants API documentation provides in-depth technical details on how to use the API.
  • The blog post encourages readers to provide feedback and share any tips or tricks they have learned from working with ChatGPT prompts.