What are `tool calls`?

Questions
1

I have used Cole Medin’s local-ai-packaged n8n workflow, customised to use local file uploads.

I have 2 small, simple text documents now uploaded to my Qdrant vector store, just to test accuracy, etc.

Strangely, the first Chat response is always:

< python_tag >tool call: get_christmas_plans()

If I repeat the same Chat question / prompt, it seems to behave “normally”.

When asking a new question/prompt, the 1st response always seems to include some leading text along the lines of (and this was 1 of the responses):

Tool call output: John’s present ideas are:

  1. Lego Creator Expert - Ferrari 250 GT Berlinetta
  2. Lego Ideas - Old Fishing Store

This is actually wrong, and if I resubmit the prompt, the leading text is not there and the correct information is pulled from my text files.

I made a mistake earlier. The correct output from the tool call is Ferrari Lego Technic and iPad Mini.

What are these tool calls? Why do they only seem to happen on the 1st question/prompt?

Information on your n8n setup

  • **n8n version: 1.67.1
  • **Database (default: SQLite): Postgres + Qdrant
  • **n8n EXECUTIONS_PROCESS setting (default: own, main): ???
  • **Running n8n via (Docker, npm, n8n cloud, desktop app): Docker
  • **Operating system: Macbook M1 with macOS 18.1
1 Like
2

Hi @roastbullish

The output of the Agent will depend on the model you’re using and the instructions you’re giving.

In your workflow, the agent doesn’t get any instructions so it will just reply according to what the base model has been instructed, which is apparently to call the get_christmas_plans().
Which model are you using?

The tool calling is the underlying functionality of the agent to execute a connected tool (in your case the qdrant vector store) and returning the output it got from the tool. However, the agent needs to be instructed on how to use the tool correctly, otherwise you may get small surprises like this :christmas_tree: :wink:
If you want the agent to call the vector store tool correctly, you need to make sure the name and description are descriptive enough for the model to know when to call it.

You can read more about the concept of tool calling here

3

This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.