How to combine multiple markdown files

Within the Aggregate Node.

You should utilize the markdown field.

This will provide you with an array of the markdown field.

In the OpenAI node: Employ {{ $json.toJsonString() }}

Hello, thank you for your response. I've followed your instructions:

• Aggregate node → linked to markdown
• OpenAI node → linked to the aggregate-markdown with the jsonstring

However, I'm still encountering the error. Please see the attached screenshot.

piccccc22223331920×985 137 KB

In this scenario, I suspect the issue might be with Deepseek.

Since the prompt appears to be fine now, do you have another model you'd like to test?

Please provide the rewritten markdown content.

That's peculiar. The ChatGPT/OpenAI prompt functions correctly when linked to a single markdown file, rather than an array of multiple markdown files. Therefore, I don't believe the issue lies with ChatGPT/OpenAI. It appears that the automation logic isn't quite flowing as expected yet.

Perhaps you could share the workflow code with the aggregated data, and we can conduct some tests.

Alternatively, please share the input json.

We can then review how to adjust the settings.

Screenshot_20250601_232345_IcdB1s1j686×570 44.1 KB

Hey Dan, apologies for the delayed response; I wasn't feeling well. I'm happy to share the workflow in a comment, similar to your example, but I'm unsure how to do it. I have the paid version of callin.io and see the share button, but I don't know how to integrate it here in the chat. The documentation (Sharing | n8n Docs) and other related posts (How do I share a workflow? The documentation doesn't relate to product - #2 by romain-n8n) haven't been helpful. I apologize, I'm still quite new to this. The workflow ID is #qn57AohkchAcLgtm. Does that assist you?

The issue still persists. If you have a moment to share your expertise, I would greatly appreciate it.

The issue seems to be that you are attempting to pass an array of prompts to the agent. You can resolve this by using a code block to construct a single string value and then placing it between the aggregate and AI nodes.

image1432×705 162 KB

It works, thanks a bunch Wouter!!!

For anyone who is reading this, the short of it: AI agents cannot work with arrays. Convert the array into a string using the code snippet that Wouter suggested. The AI agent can work with the string.

I'm having trouble distinguishing between using Aggregate and toJsonString().

It seems quite similar if the goal is to convert an array into a single string.

However, I agree that any method that aids your workflow is a valuable one.

Thank you all!

No, that's not the case. You need to grasp how callin.io manages data. It can either return multiple items, for which the subsequent node will execute for each one, OR, by using the aggregate function, instead of executing the next node multiple times, it will generate a single output item of an array type. The issue encountered was that the AI node anticipates a STRING, not an ARRAY.

Reading this will provide a more detailed explanation:

Hello,

Apologies for the potential scope creep, but I have a related question and you seem knowledgeable. Could I trouble you for some insights on the following?

I followed your advice:

1. I take the aggregated markdown data, which is an array named 'markdown'.
2. I use code to successfully convert it into a string called 'markdownprompt'.
3. Then, I pass the 'markdownprompt' string to OpenAI for processing. The issue is that while it works for most websites, sometimes it fails with an 'error - bad request' output. Do you have any ideas what might be causing this? One possibility is that the content of 'markdownprompt' might be too lengthy.

If I'm overstepping, my apologies, and please disregard this. You've already been a great help!

markdown1920×666 94.9 KB

No problem at all, we’re here to assist where we can. Could you please share your updated workflow in a code block? It’s challenging to make an accurate assessment based on your current setup. What is the size of the markdownprompt value for the instances that fail? If the issue is intermittent, it’s most likely due to a context window limitation. Which model are you currently using? What are you aiming to accomplish with the AI call, etc.? There might be a more effective and efficient approach to handle this issue logically.

Revisiting your original question, if your primary goal is to scrape company information from a website and then extract data from it, a RAG solution would be a superior and more efficient method for querying the data.

Kindly watch this video and pay close attention to the demo at the beginning where the presenter explains the problem it solves, to see if it would be suitable for your use case.