workflows.fit
Back to n8n workflows
n8n templateFreeBy Albert Ho

Intelligent Gmail label management with AI & Discord notifications

How it works - This workflow will trigger upon new mail in your gmail inbox and will automatically apply existing labels OR identify, create, then apply up to 5 new labels. - This is currently set up to run using a Op...

CommunicationHITLDevelopmentCore NodesAILangchainChain LlmGmail Trigger
Loading interactive preview...

Template notes

How it works - This workflow will trigger upon new mail in your gmail inbox and will automatically apply existing labels OR identify, create, then apply up to 5 new labels. - This is currently set up to run using a OpenAI endpoint which I have set up connected to my local llama.cpp server running. - After your email is done labeling, the workflow will send you a nice notification in your discord channel!

Set up Instructions - This workflow requires a Gmail OAuth account credential with the ability to create labels and label messages - [OPTIONAL] You can create a Discord Bot Account which connects with a Bot Token which will summarize the email and show you which labels were created and added in the workflow. - You might want to tweak the prompts a bit based on how detailed you want your labels. I've found after running it for a bit that the labels that get created mostly encompass everything I want.

Workflow Hints - The workflow works in 5 phases -- Phase 1: Find existing Gmail labels as options for the LLM to choose from. -- Phase 2: LLM chooses labels or defines new ones -- Phase 3: Create additional Gmail labels if needed -- Phase 4: Label the message with newly created and existing labels -- Phase 5: Rewrite and send a message notification via Discord.

Screenshot 1: ![Screenshot 20250831 at 9.06.29 PM.png](fileId:2291) Note: colors are my own

![Screenshot 20250831 at 9.09.01 PM.png](fileId:2292) Screenshot of notification sent via discord

Model Tips - I set this up to run completely using offline models. I had good success with Llama.cpp server with gpt-oss-20B and with a bit more tweaking could get it to work with a smaller gemma3-4b model mostly likely as well. - The whole workflow runs in about 5-10 seconds on my Blackwell Nvidia GPU. Probably would be a few seconds slower on an older piece of hardware.