AI Streaming | Server Side Events | WebSocket's

Questions
1

One feature that we are really missing in n8n is the ability to work with streaming APIs.

Is this something n8n are working on?

I use n8n for everything, I love using it for AI but I find myself not using n8n’s AI features due to a lack of SSE capabilities.

Right now, I build workflows for AI applications but I generally have to code my own AI cloud functions and use n8n to orchestrate them, if they require real-time user interaction.

For example, I have an AI app that triggers an n8n workflow, which does lots of data gathering before sending a request to my cloud function, the cloud function then interacts with the AI steaming APIs, and streams responses back to my client via a web socket. When complete, the HTTP node gets the final message returned for further actions and processing.

Of course, creating a cloud function of this nature and also having to code clients to work with the WebSocket is quite a bit of work.

I’d love it if n8n supported SSE and Web Sockets. Competitors like Buildship have this functionality, we need it too! :slight_smile:

2

It looks like your topic is missing some important information. Could you provide the following if applicable.

  • n8n version:
  • Database (default: SQLite):
  • n8n EXECUTIONS_PROCESS setting (default: own, main):
  • Running n8n via (Docker, npm, n8n cloud, desktop app):
  • Operating system:
4

Hi Wayne,

So we do have an SSE trigger. Have you tried to use it yet? If yes, were there any challenges or feature gaps?

As for the Web Sockets to stream the response back; since n8n nodes “execute” and then output their step; this paradigm would either have to change OR the AI node itself (e.g. AI Agent node) would need to have ability to stream response back. That sounds relatively non-trivial. If you have any other details on this web sockets usecase; please do share them and I’ll try to get a more technical colleague familiar with Web Sockets to weigh in on feasibility here (which of course would inform how soon we might be able to add such a feature).

Thanks!