The Moment

You’ve seen the headlines. Self-hosted AI assistants are suddenly everywhere - people running them on old laptops, dedicated hardware, home servers. Despite the security issues, despite the complex setup, despite needing a machine running 24/7, people are doing it. The demand is real and urgent.

This is the iPhone moment for personal AI. Not because of any one product, but because of what it reveals: people are ready for AI that’s always there, always reachable, integrated into how they already communicate.

I saw this coming. I’ve been living it.

Why I Built This

I have ADHD. I can’t keep track of things. My whole life I’ve felt like someone who doesn’t execute on things - I’d start to-do lists knowing they’d be abandoned, attempt projects knowing they’d stall. That learned helplessness becomes a self-fulfilling prophecy.

Then Claude Code changed everything. I started using it as a prosthetic brain - not just for coding, but for thinking. I kept a to-do list in Claude Code and realized I never got stuck anymore. The agent helped me power through each thing. Suddenly the to-do list was worth keeping because I could actually complete what was on it.

I’ve decided to be a high functioning person. I think that’s possible now.

But Claude Code is trapped on my laptop. When I close the lid, it’s gone. I imagined using my phone, my watch, the texting button in my car to throw ideas at an AI - "hold onto this for me" - and having it carry that load. If I could hire an assistant to follow me around and be my second brain, that would work great. Obviously impractical as a human employee. But I’ve been deep in agents, I know what they can do, and I realized: holy shit, the technology is there to have an agent do this.

So I built it.

What It Is

Stumpy is always-on AI agents that live in the cloud. You create an agent, connect it to your channels - Slack, SMS, Telegram, email, web chat - and talk to it. That’s it. No hardware. No complex setup. Just have a conversation about what you need it to do.

Each agent gets:

• File storage for memory, logs, whatever it needs
• A SQLite database (agents are fantastic at SQL)
• Scheduled tasks via cron
• Google Calendar access
• A Linux sandbox for running code
• The ability to generate and manipulate images

And because I spent years on a trust and safety team and hate spam with a passion, it’s opt-in by design. Agents can only message people who’ve explicitly connected. No surprise messages to random people.

The security model is real. Incoming messages from untrusted sources get scanned for prompt injection. Outgoing messages get checked for data exfiltration. You control exactly who your agent can talk to.

It’s still a work in progress - things flake out under certain conditions, we’re fixing bugs constantly. But the core is solid.

How It Works

Here’s what I actually use it for:

I have several software engineer agents and one engineering manager. When I want something done, I talk to the manager. It maintains to-dos, knows which developers are busy, finds a free one and assigns the task. If they’re all busy, it buffers work and hands it off as developers become available. I just hit the manager with my thoughts and tasks as fast as I can, and it keeps all the developer agents humming.

My personal assistant was getting overwhelmed - to-do list, shopping list, email, calendar. So I gave it a team: a shopping specialist, an email agent that scans my inbox periodically. The personal assistant delegates to specialists.

The multi-agent stuff just works. You put agents on a team, talk to them about how to collaborate, and they figure it out. They can even modify their own roles - negotiating who’s in charge of what.

Try It

Three weeks of building. An LLC. Payment processing. The whole thing.

It’s live at stumpy.ai.

Budget-based pricing, not pay-per-use - pick a tier, that’s your rate limit. No surprise bills.

Go make an agent.