Most AI you can buy is a single assistant that answers one question at a time. Smart Node is the opposite: a team of AI workers that runs inside a business, on hardware the business owns. It is far enough along now to show what it does.
The problem it solves
A small business that wants AI working inside its operations has two real options today, and both are bad.
You can wire up the AI provider APIs yourself. It starts simple, then you find you also need memory across sessions, an audit trail, approval gates, role separation and error handling. Each one becomes its own engineering project, and most small teams do not have the bandwidth for that.
Or you can rent a managed agent platform. Faster to start, but your workflows, agent logic and customer history all live on someone else's cloud, billed per seat, and moving off it later means rebuilding everything.
Smart Node is a third option: you own the whole layer. The agents, the data, the workflows and the audit trail sit on your own server, and you stay free to swap the underlying AI model whenever you want.
The idea behind it
Picture an empty Linux server. You install Smart Node, then you talk to it the way you would brief a new hire. Here is our daily routine. Here are our customers. Here is how we handle a request. It builds the workflows, connects to your tools, and from then on works as part of the team, around the clock, under your control.
I think of it as the next layer of small-business infrastructure. Twenty years ago a small business installed a mail server. Then a file server. Then a CRM. Now, an AI server. Smart Node is that server. The full write-up of the vision lives here.
What it does
Smart Node lets you stand up a crew of AI specialists, each owning a piece of the work: research, customer replies, scheduling, pricing, admin. They share memory, so they hold context across every job instead of starting cold each time. You place approval gates wherever you want a human in the loop, and let the rest run on its own. Every action the agents take is written to an audit log, so you can always see what they did.
It also does not lock you to one AI provider. Claude, GPT, Gemini, or a local model on your own machine, swappable without rewriting your agents or workflows. The model is a part you can change. The system around it is yours.
The 60-second version
That is the concept. The real question is whether it holds up on an actual business, with real calls and real stakes, instead of a scripted demo.
A real call, start to finish
So I set it up for a working distillery, the kind of place where one or two people answer every call and a busy afternoon means missed orders and a voicemail box nobody gets back to. Exactly the load a crew of agents should be able to take off their hands.
On the call you can watch the work move between agents. The first one answers and works out what the caller actually wants. From there it hands the pieces off: a stock question to the agent that knows the inventory, a price to the one that handles quoting, a tour request to the one that owns the calendar. They do not step on each other, and each reports back what it found.
The owner is never cut out of it. Where a decision matters, the work pauses for a quick approval instead of running blind, and every step is written down, so you can replay exactly what happened.
What stands out on the recording is how ordinary it sounds. No hold music, no "let me transfer you", no obvious script. The caller asks, the answer comes back, the booking lands, and all the routing between agents happens quietly in the background.
Here is that call in full, showing how all of this actually happens.
Where it is going
A single Smart Node lives in one company's network. The part I am most interested in is what happens when many of them connect. An agency's Smart Node sharing a workspace with its client's. A holding company coordinating with each subsidiary. Two independent businesses working together without either one handing its data to a third party. Most agent frameworks assume one organization, one deployment. Smart Node is built peer-to-peer from the start: each instance is a node, not a hub.
Where it stands today
Some components are already open source and useful on their own right now. The rest, along with the integration layer that turns them into one system, is in active development. I am building it in the open.
FAQ
What is Smart Node? Smart Node is a self-hosted crew of AI agents that runs inside a business on its own server. Specialists handle real work, share memory, follow the approval gates you set, and call whichever AI model you choose.
What does self-hosted AI mean here? The agents, their memory, the workflows and the audit log all run on your own server instead of a vendor's cloud. Only the calls to an external AI model leave your machine, and you can point those at a local model if you want nothing to leave at all.
Which AI providers does it work with? Any of them. Claude, GPT, Gemini, OpenRouter, or a local model through Ollama. Models are reached through one gateway, so you can switch providers without rewriting your agents or workflows.
Can a small business really run AI agents on its own server? Yes. That is who it is for: teams of one to a few hundred, not enterprises. You install it on a VPS or a server you already have, and your data and workflows stay with you.
How is this different from a managed AI agent platform? On a managed platform your workflows, agent logic and customer history live on someone else's cloud, priced per seat. With Smart Node you own that layer, audit every action, and change AI providers without re-implementing anything. The trade-off is that you run the server.
Is it available to use today? Some components are open source and usable on their own right now. The rest, plus the layer that ties them into one system, is in active development.
If you run a small business and this is the kind of help you want, I am taking on a few pilots. I will set it up for you, hands-on, and tune it to how you work. Want to try it? Write to me.
