AI chat with different steps and vector stores

[Joachim_Brindeau]

Hello!

I have a question regarding building an AI chat.
I try to suggest classification of business ideas using a list of terms.

My list is made of 10k terms that I plan on bumbing to 80k terms.
They are categorized in 45 categories.

I have managed to:

• make a tools agent that outputs the JSON of the categories (from vector store A) to filter from (let’s call it Output A).
• make a tools agent that outputs the final selection of terms (from vector store B), filtering on a metadata that matches the category ID from Output A.

But would anyone have an example of how to chain them?

I want the user to use the tchat to get a list of categories, and once he validates these categories, he can get the list of terms without going to another tchat.

I saw some workflows with routes but I don’t understand how that works.
If anyone could point me to a tutorial or some explanation, I would appreciate it.

Thank you very much

[n8n]

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:

[Jim_Le]

Hey @Joachim_Brindeau
It’s possible you can merge the two chat agents into one: 1x agent + 2x vector store tool. But this is a little bit tricky with structured outputs for both tools and the necessarily prompts to coordinate could get complex really quickly.

I don’t have a definitive solution but you could try the following instead. It’s a simpler implementation mostly to explain how the custom workflow tool works.