Generic AI is fine for drafting emails. It's useless when you need something that knows your business, your clients, and how you actually operate.
The fix isn't a better prompt. It's a better system. And n8n is one of the best tools I've found for building that system without writing a ton of custom backend code.
This post walks through how to build a personal AI assistant that's wired into your actual business data, not just a chat window floating in the cloud.
Why Generic AI Falls Short
When you open ChatGPT or Claude and ask a question, those tools don't know anything about you. They don't know your client roster, your pricing tiers, your project statuses, your SOPs. Every session starts cold.
That's fine for one-off tasks. But if you're running a business, you need an assistant that carries context. One that can tell you what stage a client is in, surface the right template, or fire off a follow-up without you explaining everything from scratch every time.
n8n makes that possible by acting as the connective tissue between your AI model and your actual data sources.
The Stack We Use
Here's a simple version of what we run internally:
- n8n as the workflow engine and orchestrator
- Claude or GPT-4o as the brain doing the reasoning
- Airtable or Notion as the business knowledge base
- Slack or Telegram as the input interface (how you talk to it)
- Google Docs or Drive for document retrieval when needed
You don't need all of these to start. A Telegram bot connected to n8n connected to one Airtable base is enough to get something useful running in a few hours.
Step 1: Pick Your Input Method
Most people overthink this. You just need a way to send a message and get a response. Telegram is my go-to for personal assistants because the bot API is dead simple to set up inside n8n and you already have Telegram on your phone.
Set up a Telegram bot through BotFather, grab the token, drop it into n8n's Telegram trigger node. That's your input layer done.
If you're building something team-facing, Slack works great. Same concept, slightly more setup.
Step 2: Connect Your Business Context
This is the part that makes it actually useful. Before you send anything to the AI, you pull in relevant context from wherever you store your business data.
Here's the basic logic in n8n:
- Message comes in from Telegram
- n8n parses the message content
- It runs a lookup against your Airtable base (clients, projects, whatever makes sense)
- That data gets injected into the prompt as context before it hits the AI node
- AI responds with that full context already loaded
- Response goes back to Telegram
The result is an assistant that can answer "what's the status on the Dario project?" because you actually fed it that data before asking the question. It's not magic, it's plumbing.
Step 3: Build a Memory Layer
Out of the box, n8n workflows don't persist conversation history between runs. Each trigger fires a fresh execution. So you need to handle memory yourself.
Simplest approach: log every exchange to an Airtable table or a Google Sheet. Store the timestamp, the input, and the AI's output. Then when a new message comes in, pull the last N rows and prepend them to the system prompt.
This gives your assistant short-term memory. It won't remember a conversation from three months ago, but it'll track the last few exchanges and maintain context within a session.
For longer-term memory, look at vector storage options like Pinecone or Supabase with pgvector. n8n has nodes for both. That's a bigger build, but it's worth it if you want the assistant to actually learn how you work over time.
Step 4: Give It Tools, Not Just Text
The difference between a chatbot and an assistant is that an assistant can do things. With n8n, you can give your AI actual actions it can trigger based on what you ask.
Examples of tools we've wired in:
- Create a new project record in Airtable when you say "start a new project for X"
- Draft and send a follow-up email through Gmail when you ask it to follow up on a proposal
- Pull today's calendar from Google Calendar and summarize what's ahead
- Search your document library and return the most relevant SOP or template
In n8n, this works by using the AI's response to trigger conditional branches in your workflow. You can either parse structured JSON output from the AI (tell it to always respond with a JSON action field), or use a router node to direct the flow based on keywords.
The JSON output approach is cleaner. Tell the model in your system prompt to always respond in a format like:
{"action": "create_project", "response": "Got it, creating the project now.", "data": {...}}
Then your n8n workflow reads the action field and routes accordingly. Simple, reliable, easy to debug.
What This Looks Like in Practice
I send a message: "What did I last discuss with Marcus and is there anything I said I'd follow up on?"
The workflow pulls the last three logged exchanges that mention Marcus, injects them as context, sends to Claude, and returns a clean summary with any action items I mentioned. Takes about four seconds.
That's not something a generic chat window can do. That's a tool that knows my business.
Common Mistakes to Avoid
- Injecting too much context. Bigger context windows don't mean you should fill them. Be selective. Pull only what's relevant to the question, not your entire database.
- No error handling. If your Airtable lookup fails, your whole workflow breaks. Add error branches from the start.
- Skipping the system prompt. The system prompt is where you define the assistant's behavior, tone, and limitations. Don't skip it. Spend real time on it.
- Building too much before testing. Get one flow working end to end before you add tools and memory. Keep it simple until it's reliable.
The system prompt breakdown above, including the exact template and the section-by-section logic, is inside Inner Circle. That's where we share the actual internals of how we build, not just the concepts.
Keep building
- Cursor: Generate Realistic Seed Data for Local Dev from the prompt library
- Gemini Long Context: Codebase Onboarding Brief from the prompt library
- How I Run My Entire Business by Talking to It
- The full Vaylo prompt library, new drops weekly
