Gmail agent that picks or creates labels with OpenAI

What it is

This n8n workflow labels every new Gmail email with an AI agent. The agent reads the email, looks at the labels you already have and adds the best one. If none fits, it creates a new label, as a sub label of an existing one when it can. Nine nodes do the work, beside four sticky notes.

Who it suits

People whose Gmail labels grow over time and who want the labelling done for them, new labels included. Skip it if you want a fixed list of labels, since this agent may create new ones.

You need a little n8n: import a file, add two credentials and keep a few labels in Gmail for the agent to follow.

Where to use it

Any n8n that can reach Gmail and OpenAI will run it. It works on mail as it arrives, so you do not need to open n8n.

• n8n Cloud: import the file into a new workflow there and switch it on
• Your own n8n server: the same file and steps
• Narrower scope: add a Search filter to the trigger so that only some emails are labelled
• Your own rules: rewrite the System Message to say how your labels should be used

Setup

Download saves a zip with workflow.json and Mindket’s setup notes in your downloads folder. A few existing labels help, because the agent copies their style.

1. In n8n, create a workflow, open its menu, choose Import from File and pick workflow.json.
2. Add a Gmail credential and select it in Gmail Trigger and in the four tools, which are Gmail read labels, Gmail get message, Gmail add label to message and Gmail create label.
3. Add an OpenAI credential and select it in OpenAI Chat Model.
4. Read the System Message in Gmail labelling agent and adjust it to your labels.
5. Switch the workflow on, then send yourself a test email.

How to use it

Once it is on, new email is labelled within about five minutes. Open Gmail to see the label on each message. The agent can also create a label, so a new one may appear in your label list. It never sends or deletes mail.

Example

A consultant receives a message from a new client about a project kickoff. The agent reads your label list and finds a Clients label, but nothing for this client.

It creates a sub label under Clients for the new client and adds it to the message. When the next message from that client arrives, the agent finds the label and reuses it.

For newcomers and experts

If you are new. An agent is an AI that can use tools. Here the tools are four Gmail actions, and the agent decides which to use and in what order. A Gmail label is a tag you can filter on.

For experts. The trigger item goes to the agent as JSON, and the agent reads the message itself through Gmail get message. Window Buffer Memory uses the email id as its session key, so each email has its own short memory.

Good to know

The original named no model, and its prompt asked for something no tool could do.

• Models: the original set no model, so n8n’s default applied. OpenAI’s deprecations page, read on 12 November 2026, lists gpt-5-mini, that default, as shutting down on 11 December 2026 with gpt-5.6-terra as the replacement. Mindket’s copy sets gpt-5.6-terra, which supports function calling, and leaves out the original’s limit of 4,096 output tokens
• The prompt: it also told the agent to take unimportant mail out of the inbox, but no tool in the file can remove a label. Mindket’s copy drops that line, and you can bring it back by adding a Gmail tool with the Remove Label operation
• Labels grow: the agent creates labels, and its prompt says to put a new label under an AI parent label when no parent exists
• Every new email counts: the trigger has no filter, so each email, newsletters included, goes to the agent
• Steps per email: one email can take several model calls, because the agent reads labels, reads the message and adds or creates a label
• Memory: the window memory lives in n8n’s own memory and is kept per email
• Your data: the email goes to OpenAI
• Credentials: the JSON holds none, so add the Gmail and OpenAI credentials yourself