I have a personal AI assistant that lives in my Telegram. I message it throughout the day — research requests, quick drafts, questions about my own notes, reminders I want processed. It knows my writing style, my business context, and the kinds of tasks I do regularly.
Building it took one focused day. Maintaining it takes almost nothing — it runs on a $5 VPS with n8n and I haven’t touched the core setup in months.
This guide is about building exactly that: a personal AI assistant that’s genuinely useful to you specifically, not a generic chatbot.
What makes an AI assistant actually personal
A personal AI assistant is different from Claude.ai or ChatGPT in three ways:
1. It knows your context. Your role, your business, your preferences, your ongoing projects. You don’t have to re-explain yourself every time. The system prompt carries this context permanently.
2. It can access your tools. A general AI assistant knows nothing about your calendar, your emails, your notes, or your documents unless you paste them in. A personal assistant can query these directly.
3. It lives where you work. Not in a separate tab you forget to open — in Telegram, Slack, or wherever you already spend time. The best interface is the one with the lowest friction to use.
Building these three properties into an assistant is what today’s guide covers.
Choosing your interface
Telegram — my recommendation
Telegram bots are easy to set up (free, takes 10 minutes with @BotFather), available on every device, and support rich formatting. The Telegram node in n8n makes connecting your AI backend straightforward. Most people already have Telegram installed. This is what I use.
Slack — best if you’re already in Slack
If you live in Slack for work, a Slack bot means your assistant is always one @mention away. n8n has a Slack trigger node. Slightly more setup than Telegram but deeply integrated into a workflow if Slack is your primary tool.
Custom web UI
A simple chat interface built in Bolt or React. More setup, but gives you full control over the UI and makes it easy to share with others or access from a browser. Good option if you want a more polished experience or plan to expand into a product.
Email-based
Send an email to a specific address, get a reply from your assistant. Lowest-friction to set up (n8n’s email trigger + Gmail sender), but the asynchronous nature means it’s better for batch tasks than quick questions.
The technical setup
For a Telegram-based personal assistant in n8n:
What you need:
- n8n (self-hosted or cloud)
- A Telegram bot token (from @BotFather)
- An AI API key (Anthropic or OpenAI)
- Optionally: a simple database for memory (Supabase free tier works)
The core workflow structure:
- Telegram Trigger — fires on any message to your bot
- Set node — extract the message text and user ID
- Optional: Memory lookup — retrieve relevant past context from your database
- AI node — system prompt + conversation history + current message → AI response
- Optional: Memory save — save the exchange to your database
- Telegram node — send the response back to the chat
The minimum viable version (no memory, no tool integrations) can be built in about 2 hours. Add memory and tools in the afternoon.
Writing your system prompt
This is the most important investment you’ll make in your assistant. A well-written system prompt is what separates a useful personal assistant from a generic AI wrapper.
My system prompt structure:
Part 1: identity and context
“You are [your name]’s personal AI assistant. [Your name] is [brief description of role and work]. You help with [primary task areas]. You have access to [tools/integrations if any].”
Part 2: communication style
“Communicate in [tone: direct/casual/professional]. Keep responses concise unless asked for detail. When something is unclear, ask one clarifying question rather than assuming. Use [specific formatting preferences: bullet points for lists, bold for key terms, etc.].”
Part 3: task-specific instructions
“For research requests: summarize findings in bullet points with sources. For writing tasks: match [your name]’s writing style — direct, no jargon, short sentences. For scheduling: always confirm before treating anything as final.”
Part 4: constraints
“Do not: make up information you don’t have, take actions without confirmation, or access tools that haven’t been connected. If you can’t help with something, say so directly and suggest an alternative.”
This prompt gets refined over time as you interact with the assistant. I add new instructions whenever I notice it doing something in a way I don’t want.
Adding memory
Without memory, your assistant treats every message as a fresh conversation. This is fine for one-off tasks but limits its usefulness for anything ongoing.
Short-term memory: conversation history
For multi-turn conversations, pass the recent message history to the AI node with each request. n8n’s AI Agent node handles this natively with a built-in memory buffer. For a custom workflow, store the last N messages in a variable and include them in each prompt.
Long-term memory: a notes database
For things you want the assistant to remember permanently — your preferences, ongoing projects, key facts about your business — use a simple Supabase table as a knowledge store. When you tell the assistant something it should remember, it writes it to the database. Future queries can optionally retrieve relevant records.
Simple implementation: a Supabase table with columns for content, category, and timestamp. A “remember this” command triggers a write. A “what do you know about X” command queries relevant records. Not a full vector database — just structured notes that the AI can reason over.
Full vector memory: for power users
For more sophisticated memory — semantic search over large bodies of notes, referencing your documents — you’d use a vector database like Pinecone or Supabase’s pgvector extension. This is a half-day project in itself and is probably overkill for day one. Build basic memory first.
Adding tool integrations
The real power of a personal assistant comes from connecting it to your actual tools. Here are the integrations I’ve added over time, in order of usefulness:
Web search (high value, easy)
Connect a search API (Serper.dev or Brave Search API, both have cheap paid tiers) so the assistant can look things up when asked. Add an HTTP Request node that calls the search API when the assistant indicates it needs to search, then feed the results back to the AI node for synthesis.
Calendar read access (high value, moderate complexity)
Connect Google Calendar via n8n’s Google Calendar node. Let the assistant answer questions like “what do I have this week” or “when am I next free on a Monday?” Read-only is fine for day one — adding write access (creating events) requires extra care and confirmation steps.
Note-taking app (high value, varies by tool)
If you use Notion, Obsidian, or Bear, connecting your notes gives the assistant access to your knowledge base. Ask it to find notes on a topic, summarize something you wrote, or build on ideas you’ve already captured.
Email read (useful, use carefully)
Read access to your inbox lets the assistant summarize email threads, find specific messages, or draft replies. Be deliberate about what you’re comfortable with here — a misconfigured email integration can cause problems. Start with read-only and test carefully.
Three workflows my assistant handles daily
1. Research and summarize
I message: “Research [topic] and give me a 5-bullet summary with the key points and 2-3 good sources.” The assistant searches the web, reads the top results, and replies with a structured summary. Replaces about 20 minutes of manual research per query.
2. Draft emails and messages
I message: “Draft a reply to [brief description of email context]. Tone: professional but warm. The key point I want to make is [X].” The assistant drafts a reply I can review and edit. Cuts email writing time roughly in half.
3. Process my notes
I send rough meeting notes as a message. The assistant structures them into: decisions made, action items with owners, key points discussed, and open questions. Eliminates the 15-minute cleanup I used to do after every meeting.
What to build next
- → [LINK: How to Build an AI Agent with n8n] — go deeper on n8n automation
- → [LINK: n8n vs Make vs Zapier — Tool Comparison]
- → Back to the Build with AI pillar
FAQ
What’s the difference between a personal AI assistant and just using chatgpt?
Context, tools, and interface. ChatGPT starts fresh every conversation and doesn’t know anything about you unless you tell it. A personal assistant built on n8n knows your context permanently, can access your tools, and lives where you already work. The output quality is similar; the friction is dramatically lower.
Is this secure? my personal data goes through n8n and an AI API.
If you self-host n8n, your data stays on your server until it’s sent to the AI API. Both Anthropic and OpenAI have enterprise-grade data handling and opt-out options for training. For sensitive business data, review each provider’s data processing terms. If this is a concern, self-hosted open-source models via Ollama are an alternative that keeps data fully local.
Can i use a free AI model for my personal assistant?
Yes. Groq offers fast inference on open-source models (Llama 3, Mixtral) on a generous free tier. Ollama lets you run models locally for zero cost but requires a capable machine. For a personal assistant with moderate usage, the cost on Claude or GPT-4o APIs is usually under $5/month anyway.
How long does it take to set up?
The minimum viable version — Telegram bot connected to an AI node via n8n — takes 2 to 3 hours. Adding memory takes another 2 hours. Tool integrations vary — a web search integration is 1 hour; calendar access is 2–3 hours depending on the tool. A full-featured personal assistant is a one-day project.
Can i share my personal assistant with others?
Yes, with modifications. The system prompt would need to remove personal context and add shared knowledge. The interface could be a custom web UI rather than a personal Telegram bot. This is essentially how simple SaaS AI products are built — a personal assistant for one person is architecturally similar to a shared tool for many.