Is it possible to have streaming with OpenAI Assistant?

Hello,

My question is very simple: without using plugins, is it possible to receive streaming answers from my OpenAI Assistant? Not generic chat, but OpenAI Assisntant. I already have implemented streaming with generic chat, but now I’d like to know if I can go further.

Thank you

OpenAI is deprecating Assistant, so you’ll need to move to the responses API. Bubble API connector supports streaming, which must work with responses API.

Hello @angel1996, Yes, it’s possible to receive streaming responses from an OpenAI Assistant without using any plugins. OpenAI currently supports two approaches for assistants:

1. Responses API:

  • Designed for chat-style interactions.

  • Supports streaming out of the box.

  • You get structured events for text deltas and tool invocations.

  • Ideal if your assistant doesn’t heavily rely on threads or files.

  • Simpler to implement and future-facing.

2. Assistants API with Threads/Runs:

  • Used when you want persistent threads, file handling, or more complex multi-step workflows.

  • Supports streaming via Server-Sent Events (SSE).

  • You’ll receive events like:

    • response.output_text.delta → when text chunks arrive

    • thread.run.requires_action → when the assistant requests tool output

    • response.completed → when the run finishes

  • This is better if you’re using code interpreter, vector stores, or complex assistant behaviors.

1 Like

Thanks for your answer @harsh.tiwari

I’d like to confirm it because I don’t know if I expressed myself properly. I’m already using what you can see in the screenshot below (responses). In my chatbot, I’m using OpenAI’s AI as a generic chatbot, with no documents, and I’m receiving fast answers with streaming.

But I’d like to do is to receive OpenAI’s responses coming from something that has instructions and 2-3 pdf files (I thought that was only possible with the assistant), so the answers are based on those instructions and documents. And I’d like to receive that response in a fast way with streaming.

Is it possible to get what I described in the last paragraph? In case that it is possible, are there any guides, documents, videos or ideas of how to do it?

Thanks a lot for the answer :slight_smile:

I think it’s possible. But this is more about Open AI than Bubble. You should read the documentation or ask Chat GPT itself. From what I know, it is possible to create your own GPT, assign it a name and instruct it with documents and instructions of textual type or images. I also believe that then this can be made available through API.