Every "build your own AI assistant" tutorial ends the same way: connect this $30/month tool to that $20/month tool, pay for API credits, wire it all together in some automation platform. By the time you're done you're paying $80+ a month and maintaining plumbing that breaks every other week.
I built mine differently. Everything lives inside Claude. No external tools, no automation platforms, no API costs. The only thing I pay for is the Claude subscription I already had. Net new spend: $0.
Follow my Substack to get fresh AI alpha:
[substack.com/@0xmorty](//substack.com/@0xmorty)
It handles my research, my writing, my inbox planning, my daily priorities, and even builds me custom tools on demand. Here's exactly how it's put together - copy it piece by piece.
What "$0" means here:
Not the free tier - I mean no spend
*beyond*
Claude itself. Most people stack 4-5 paid tools to get an "assistant." This entire build uses features already included in your Claude subscription. The marginal cost is nothing.
1. The Home Base - One Project
Everything starts here. This is the "body" your assistant lives in.
Your assistant needs a permanent home - not a random new chat each time. That's a Project. It holds a system prompt (custom instructions) and your context files, and every conversation inside it starts with full context.
Think of the Project as the assistant's job description and desk. Set it up once, and you never re-explain who you are or what you want again.
SET IT UP
- Sidebar → Projects → New Project → name it "My Assistant"
- You'll add the system prompt (Step 2) and files (Step 4) here
- From now on, you only talk to your assistant *inside* this Project
✓ A permanent workspace that remembers your setup forever
2. The Brain - The Master Prompt
The custom instructions that define how your assistant behaves.
This is the most important part. The master prompt is your assistant's operating manual - it defines its role, how it talks to you, what format it uses, and what it should never do. Paste this into the Project's custom instructions.
MASTER PROMPT - PASTE INTO PROJECT INSTRUCTIONS
You are my personal assistant. You know me through the files in this
project. Read them before responding.
YOUR JOB:
- Help me think, plan, research, write, and stay organized
- Be proactive: if you spot something I'm missing, say it
- Match my preferences (in preferences.txt) in everything you produce
HOW YOU RESPOND:
- Direct and concise. No filler, no "Certainly!", no preamble.
- When I ask for a plan, give me steps - not an essay
- When I ask for writing, use my voice (see my samples)
- When something needs a decision from me, flag it clearly with 🔴
DAILY MODES (I'll name one):
- "brief me" → summarize what matters today
- "plan my day" → top 3 priorities from my goals
- "research [X]" → search + structured findings
- "build me [X]" → make it as an artifact
If my request is vague, ask one sharp question before guessing.✓ One paste = your assistant now has a personality and a job
3. The Identity - Memory + Context Files
What turns a generic chatbot into YOUR assistant.
The master prompt tells the assistant *how* to behave. The context files and memory tell it *who you are*. This is the difference between a stranger and someone who's worked with you for a year.
Upload three simple text files to your Project. Update them as your life changes - that's the entire maintenance.
THE 3 FILES TO UPLOAD
about_me.txt
Name, what I do, current projects, what I'm working toward.
preferences.txt
- I like concise answers, no fluff
- Writing voice: [paste 3-5 of your best posts here]
- People whose messages always matter: [names]
- Topics I care about / topics to ignore
goals.txt
This quarter I'm focused on: [2-3 goals]
My biggest constraint right now: [time / money / energy]
What "a good week" looks like for me: [...]On top of the files, Claude's Memory accumulates context automatically as you use it. The combination means your assistant gets sharper every week - no extra work from you.
✓ The assistant knows you - no re-explaining, ever
4. The Senses - Web Search
So your assistant knows what's happening now, not just what it was trained on.
An assistant that can't see the current world is half-useful. Turn on web search and your assistant can research live - news in your niche, current prices, what happened today, fact-checking before you publish.
This is what makes "brief me on what's happening in AI today" a real command instead of a guess from stale training data.
ENABLE + USE
- Make sure web search is toggled on (near the chat input or in settings)
- "Research the latest on \[topic\] and give me 3 takeaways"
- "What's the current state of \[thing\]? Search before answering."
- "Fact-check this claim before I post it: \[claim\]"
✓ A research assistant that's always up to date
5. The Hands - Artifacts
Your assistant doesn't just talk. It builds you tools on demand.
This is the part people forget. Your assistant can build functional little tools for you - a tracker, a calculator, a planner, a dashboard - right in the chat, no code from you. You describe the tool, it builds it, you keep it.
Mine has built me a daily content tracker, a "what should I post" decision helper, and a simple revenue calculator. Each took one sentence to request.
THINGS TO ASK YOUR ASSISTANT TO BUILD
"Build me a daily habit tracker with a streak counter."
"Make a simple planner for my top 3 tasks each day."
"Build a calculator that shows my revenue per follower."
"Create a decision matrix to compare 3 options on 5 criteria."
→ Each one becomes a real, working tool you can use and reuse.✓ Personal software on demand - the assistant has hands, not just a mouth
**
How I Actually Use It Every Day**
Here's what a normal day with the assistant looks like - every interaction happens inside the one Project:
- Morning: "brief me" → it searches my niche, checks my goals, hands me what matters in 3 minutes
- Planning: "plan my day" → top 3 priorities pulled from goals.txt, in order
- Working: "research \[X\]" or "draft \[Y\] in my voice" → it already knows my style and context
- On demand: "build me \[tool\]" → a working artifact in under a minute
- Evening: "what did I not get to, and what should move to tomorrow?"
The unlock:
Because it's all one Project with persistent context, I never start from scratch. The assistant on day 30 is dramatically better than day 1 - not because Claude changed, but because the context compounded.
The Honest Limitations
So you're not surprised - here's what this $0 setup *doesn't* do, and how I work around it:
- It doesn't run on its own. No external automation means I trigger it manually. That's a feature for me - I stay in the loop. If you want true scheduled automation, that's where paid tools would come in.
- It doesn't act in other apps by itself unless you connect them. If you add connectors (Drive, Calendar, etc.) it can reach into those - still no extra spend.
- Usage limits exist. Heavy days can hit them. For most people one assistant Project stays well within normal use.
The point isn't "this replaces every paid tool."
It's that 90% of what people pay $80/month of stacked tools to do, you can do inside one Claude subscription with zero extra spend. Start there. Add paid plumbing only when you've proven you actually need it.
The Build - Quick Reference
Save this.
Build the Project today, paste the master prompt, upload three files. 20 minutes, $0 in new spend, and you have a real personal assistant by tonight.