Honestly, I've fallen asleep with this trend. Question, have any of you built conversational chatbots for companies using n8n? I've been making them for a while, but just for practice and demos. But at a production level, I haven't done any until now that I have the opportunity to test one. Still, I'm not sure if n8n is ultimately the best option for it. This chatbot needs to answer questions about the company and its programs, and schedule appointments. And I already have a large part of it built. But relying on a single prompt to do all that doesn't seem effective to me. But if I split it into several agents, then it's too many calls, which makes it not very cost-effective. So I'm in that dilemma.

I am using AI Agents in n8n and needed to output JSON. Of course, LLM outputs text, which I then parse to JSON. However, Even with clear prompt (output JSON only) I noticed inconsistent results. Example - I need from AI to go to link and parse content - Create article TEXT from that content - Create article TITLE from that content - Assign meaningful tags (i.e. from Google Sheet predefined tags list) - Output this as JSON ({contenwt: "...", title: "...", tags: ["..."]}). What is your approach in case you need JSON output from LLM? Should I use different agents for each part (title, content, tags,...) and then Merge it?

n8n does not have native, first-class multi-tenancy support (as of now), and the "Hard Isolation" is the recommended way as i can see. Basically, create copy of workflow. However this cause maintenance harder. On the other way, "Soft Isolation" is supported more via hacks than natively supported in N8N (and other known tools). Am I wrong? How do you deal with multi-tenancy?

It might be really hard but missing feature from the N8N is the way for workflow to output costs of each run so you can easier calculate pricing. This is even more important for multitenancy solutions. I know this depends on support of integrated APIs but would be extremely useful to know this at least for LLM models.

I'm working on a workflow where I need my agent to generate images not just from text prompts, but also based on existing images that are provided as input. My current setup only handles text-to-image generation. Could anyone share some insights or point me in the right direction for achieving image-to-image generation with an agent?