Meet ARIA: Building a Personal AI Executive Assistant
Part of: aria-progress
Every developer I know has a graveyard of productivity tools they tried and abandoned. Notion boards that went stale. Linear tickets nobody updated. Slack reminders that got snoozed to infinity. I was no different.
What I actually wanted was simple: something that knew my context, could give me a real briefing in the morning, and didn’t require me to update it manually. Something that worked the way my brain works, not the way some SaaS product manager thinks it should.
So I built ARIA.
What is ARIA?
ARIA stands for Assistant for Reports, Insights & Agenda. It’s my personal AI executive assistant, built on top of Claude Code and a collection of MCP servers I’ve been developing over the past few months.
The short version: ARIA is a CLI-first assistant that lives in my terminal and understands my entire work context — projects, finances, infrastructure, tasks, calendar. When I type /aria in the morning, I get a real briefing, not a template.
/aria # Morning briefing
/aria evening # End of day report
/aria task "title" # Create a task in Hub
/aria health # Docker + infra status
/aria vps # Production server healthIt’s not a product. It’s not trying to be one. It’s a system I designed for exactly one user: me.
How it Started
The seed was frustration.
I run Aethos Tech, where I build micro SaaS products and client projects simultaneously. On any given week I might be pushing a feature on a client’s platform, running a CI/CD fix on a different service, and validating a new product idea. The context-switching tax is brutal.
My mornings used to look like this:
- Open browser, check Linear → which tasks are due?
- Open another tab, check bank → did the Neon bill hit?
- SSH into VPS → is everything running?
- Check Slack/WhatsApp → anything urgent?
- Open VS Code → okay, now where was I?
That’s 20–30 minutes of overhead before writing a single line of code. And since it was fragmented across tools, I was constantly losing context mid-research.
The first version of ARIA was just a Claude Code slash command that pulled project git status and formatted it. Honestly pretty basic. But it saved me 10 minutes every morning, and that was enough to keep building.
The Architecture
ARIA is built on three layers:
1. Claude Code + Slash Commands
Claude Code supports custom slash commands via ~/.claude/ config files. Each command loads a “skill” — a detailed prompt that gives the model full context on what to do, which tools to call, and how to format output.
This means /aria isn’t a hardcoded script. It’s a reasoning system that adapts: if it’s morning, it runs the full briefing flow; if it’s afternoon, it does a lighter check; if I ask for something specific, it handles that too.
2. MCP Servers
The real power comes from the MCP (Model Context Protocol) servers running on my machine. MCP lets you expose any data source or API as tools that Claude can call directly.
I’ve built and connected:
| MCP Server | What it does |
|---|---|
| ARIA MCP | Core: project scan, tasks, briefings, insights |
| Neutron | Personal finance — P&L, budget, recurring payments |
| Docker MCP | Container health, logs, start/stop |
| Rastro Pop MCP | Client project monitoring (my microservices stack) |
| Google Calendar | Events and scheduling |
| WhatsApp MCP | Receive/send messages programmatically |
| Memory MCP | Persistent knowledge graph across sessions |
When I run /aria, Claude orchestrates calls across these servers to build a coherent picture of my day — without me lifting a finger.
3. Hub (The Backend)
Hub is a private Next.js + PostgreSQL app I run on my VPS. It’s the central data store for:
- Tasks (replaces Linear for my personal workflow)
- Insights and learnings captured during sessions
- Briefing history (so ARIA can see what it told me yesterday)
- Project activity aggregation
The interesting thing about Hub is that ARIA can still work when Hub is offline — it falls back to direct MCP calls. Queue any writes locally, sync when back online.
The 4-Tier Model Routing System
One thing I’m particularly happy with: ARIA routes different tasks to different AI models based on complexity and cost.
| Tier | Model | Cost | Used for |
|---|---|---|---|
| 0 | Ollama llama3.2:3b (local) | Free | Commits, summaries, translations |
| 1 | Claude Haiku | Cheapest | Quick lookups |
| 2 | Claude Sonnet | Moderate | Coding, features — default |
| 3 | Claude Opus | Expensive | Architecture, audits |
Writing a commit message? That goes to Ollama — free, instant, runs locally. Security audit? That escalates to Opus. Day-to-day coding stays on Sonnet.
This keeps costs low without sacrificing quality where it matters. I’m spending roughly $15-20/month on API calls for what would otherwise require multiple SaaS subscriptions.
What It Actually Feels Like
Here’s a real morning briefing, lightly edited:
☀️ GOOD MORNING, CÉSAR
📅 FRIDAY, FEB 20 — 09:14
📋 TASKS (4 active)
□ Configure Abacate Pay credentials — Menthos [HIGH]
□ Blog first post — Aethos Blog [MEDIUM]
□ Review Rastro Pop deploy — [HIGH]
□ VPS disk cleanup — [LOW]
🚀 PROJECTS
🟢 aethos-blog — 2h ago (build: added UX fixes)
🟡 menthos — 1d ago (deploy: Vercel live)
🔴 listai-shopee — 12d ago (waiting on client creds)
💰 NEUTRON: R$ 3,200 receita / R$ 1,100 despesa | +R$ 2,100
⚠️ Software 82% of budget
🐳 DOCKER: 7 containers running, all healthyThat’s my entire context in under 10 seconds. No tabs, no switching, no copy-pasting. Just /aria and go.
What’s Next for ARIA
ARIA is alive and evolving. Upcoming experiments:
- Proactive alerts — WhatsApp notification when a container goes down or a task deadline is today
- Session learning — ARIA capturing patterns from sessions and surfacing them as insights over time
- Financial forecasting — Neutron integration deep enough to project next month’s cash flow
- Voice interface — Dictate tasks and capture ideas while away from keyboard
I’ll be writing about each of these as they get built. This blog is as much a changelog for ARIA as it is a place for broader thoughts on building tools and products.
Why Build It Yourself?
Because the off-the-shelf tools don’t know you.
Notion doesn’t know that aethos-blog had commits 2 hours ago. Linear doesn’t know your Docker containers are healthy. Your calendar app doesn’t know your Neutron budget is running hot.
ARIA knows all of that. Because I built it to.
That’s the premise. Let’s see where it goes.
ARIA is not open source yet — it’s deeply tied to my specific stack and workflows. But if you’re curious about building something similar, I’m happy to talk architecture. Find me on GitHub or ping me at aethostech.com.br.