n8n Workflow automation

AI PERSONAL MULTIAGENT ASSISTANT

Step 1: Webhook Message Intake

• Users send a POST request containing:

message (text or voice-transcribed input)
inputType (text or voice)

• The webhook waits until the full AI response is generated before replying.

Step 2: Central AI Agent Processes the Request

• The AI Agent (Cohere-powered) reads a detailed system prompt defining when and how to use each tool.
• It selects the right tools, chains them in the correct order, and executes all steps before responding.

Step 3: Tool Execution

• The Google Sheets Tool looks up contacts (Name, Email, Phone, Company) from a live Google Sheet.
• The Email Agent Tool calls the Email sub-workflow to read Gmail messages or create draft emails.
• The Calendar Agent Tool calls the Calendar sub-workflow to create, get, update, or delete Google Calendar events.
• The Research Agent Tool calls the Research sub-workflow to search the web via SerpAPI or fetch Hacker News articles.

Step 4: Session Memory

• A Simple Memory node stores conversation history keyed by the user’s IP address.
• This allows follow-up queries like “tell me more about him” to resolve correctly without repeating context.

Step 5: Output Cleaning & Response

• A JavaScript Code node cleans the AI output — removing markdown, fixing newlines, and formatting bullets.
• The Respond to Webhook node returns a JSON object with output, timestamp, and inputType.

⇒ Research Agent – How It Works

• Receives a search query from the Main Workflow and classifies it into one of 7 types: Brief, Moderate, Detailed, Articles, Comparison, How-To, or News.
• Uses SerpAPI for general web searches → facts, trends, news, and definitions.
  Uses Hacker News when the user explicitly asks for articles about a topic.
• Runs 1–6 searches depending on query depth and returns a fully synthesized structured response.

⇒ Calendar Agent – How It Works

• Receives a natural-language calendar query and immediately calculates the current IST date and time via a JavaScript Code node.
• Blocks past-time CREATE requests before they reach the AI Agent, asking the user for a new time instead.
• Supports four operations:

CREATE → validates title and future time, handles overnight events, creates the event, checks for conflicts.

GET → resolves date ranges like “today”, “this week”, “next Monday” and lists all events in that window.

UPDATE → performs a fresh GET first to get the current eventId, then applies the changes.

DELETE → performs a fresh GET first, shows the event to the user, asks for confirmation, then deletes.

⇒ Email Agent – How It Works

• Receives a query from the Main Workflow with a resolved email address and full context.
• Identifies whether the action is READ (grab, get, fetch, find) or WRITE (send, write, draft, compose)
• FOR READ:

Fetches emails filtered by sender address with a configurable limit (default 10).
Returns a numbered list with subject, human-readable date, and a 2–3 sentence preview per email.

• FOR WRITE:

Extracts recipient email, topic, tone hints, and any included research content.
Generates a specific subject line and writes a fully-formed email — no placeholders, ever.
Creates a Gmail draft and always signs off as: Harsh.