Chat agent with a country capitals tool and a date tool

Chat agent with a country capitals tool and a date tool

An n8n workflow that shows how to give an AI agent two custom tools. A chat message goes to an AI Agent that uses an
OpenAI chat model. One tool is a code tool that returns the current date and time. The other calls a sub-workflow in the
same file, which lists a set of fictional countries or returns the capital of one of them.

What is inside

• workflow.json: the workflow as n8n exports it, 13 working nodes and two sticky notes. It holds no credential.
• README.md: these notes.

What we changed from the original copy

• The agent was an OpenAI Assistant node tied to an assistant in its author's own OpenAI account. OpenAI's deprecations
  page, read on 12 November 2026, says the Assistants API shut down on 26 August 2026, and names the Responses API and
  the Conversations API as its replacements. The node is now AI Agent (version 1.7) with an OpenAI Chat Model
  (version 1.2) set to gpt-5.6-terra. The assistant's own instructions lived in OpenAI's dashboard and were not in the
  file, so the agent has a short system message that names the two tools.
• The country lookup never worked. Mapping data1 supplies the country list to the merge, and it had no incoming line,
  so it never ran, and every lookup returned Capital not found. List countries? now feeds it from its false line too.
  In an offline test of the sub-workflow, Narnia returns Cair Paravel.
• The two sticky notes are reworded, and the pinned test item on Execute Workflow Trigger is removed.
• The assistant ID and the credential reference are removed, so you pick your own credential.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• An OpenAI API key that can use gpt-5.6-terra.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Add an OpenAI credential and select it in OpenAI Chat Model.
3. Save the workflow. Tool to call the workflow below calls this same workflow by its own ID, and a new workflow has
   none until it is saved.
4. Open the chat from the button at the bottom of the screen and ask, for example, what the capital of Narnia is.

How a run goes

1. A chat message starts On new manual Chat Message, which hands it to AI Agent as chatInput.
2. AI Agent asks OpenAI Chat Model what to do, and can call a tool up to 10 times, which is the node's default.
3. Tool: Get current date and time, named date_tool, returns the current time as an ISO timestamp.
4. Tool to call the workflow below, named country_capitals_tool, runs the sub-workflow in this same file with the
   agent's text as query. Execute Workflow Trigger starts it, and List countries? checks whether the query is
   exactly list.
5. For list, Mapping data makes one item for each of the ten countries, Concatenate country names joins the names
   with line breaks, and Return country list returns them in response.
6. For anything else, Mapping data1 makes the same ten items. Get the matching country's details matches the query
   with the country field, and Return specific capital returns the capital in response, or Capital not found.
7. The agent turns the tool's answer into a chat reply.

Change the look

• The agent's behavior: the system message in AI Agent.
• What the agent knows about a tool: the name and description in Tool to call the workflow below and
  Tool: Get current date and time. The model reads them to decide when to call the tool.
• The countries: the list in Mapping data and the same list in Mapping data1. Keep the two equal.
• The model: OpenAI Chat Model.
• Memory: connect a memory node to AI Agent to keep earlier messages.

Good to know

• The Assistants API is retired, so the original node could not run today. gpt-5.6-terra is a current OpenAI chat model.
• The agent has no memory, as in the original. Each message is answered on its own.
• A lookup needs the country spelled and capitalized exactly as in the list. narnia finds nothing and returns
  Capital not found. The tool description tells the model to call list first when it is unsure.
• The country list is in two Code nodes. Change both, or lookups and the list will disagree.
• The tool calls its own workflow by ID, so it works after you save the imported copy.
• OpenAI bills one model call for each step of the agent, and an answer that uses a tool needs at least two.
• Your chat messages go to OpenAI.