Why We Publish the Tools Our AI Agents Run On
worldinmotion reaches 1.0, and mcp-patchbay and mcp-frontdesk join it. All three are public under Apache-2.0, and all three come from our own workbench.

A customer should have to take our word for as little as possible. That is why we publish our code: fourteen projects now, from the District AI apps to the tools behind them.
All three tools in this launch, worldinmotion, mcp-patchbay and mcp-frontdesk, started in our own work with AI agents.
worldinmotion 1.0: now the phone rings
worldinmotion is a phone line for AI agents.
It lets AI agent sessions message each other in real time, across machines and between agents from different makers. The messages are encrypted from end to end. The relay in the middle carries them and cannot read them. There is no account to create, and it sends no telemetry.
We use it all day. Our agents on three machines hand work to each other through it: building, reviewing and reporting on every deploy.
Until now there was a gap. An agent that had finished its work and gone quiet did not always hear a new message. It found out when it next finished a task, a little like checking voicemail.
In 1.0 the phone rings. One command starts an agent that wakes the moment a message arrives:
wim run gemini
It works through the Agent Client Protocol, an open standard. We have run it with Gemini CLI and Antigravity CLI, and it includes commands for Cursor, Copilot CLI and Codex, built from their makers' documentation.
Waking an agent comes with two rules. Only a person can start an agent this way, never another agent. And worldinmotion never says yes to anything beyond its own messaging tools. Everything else a woken agent may do is what you allowed beforehand, and nothing more.
To set it up on a machine, one line is enough:
npx worldinmotion setup
A connector is a door
The two new tools look after a different part of an agent's work: what it can reach.
An assistant that can read your call log can tell you which customers you missed last week. The piece that lets it reach something is called a connector. The standard behind it is the Model Context Protocol, or MCP, and it is open: anyone can build on it.
A connector is a door, and a door needs two things. You should be able to keep it closed until it is needed. And you should know who is coming through. We needed both for our own work, so we built both.
mcp-patchbay keeps every door closed until it is needed
A patchbay is the panel in a recording studio where you plug in only the lines you need. Ours does the same job for an AI assistant.
Our assistants can reach about forty connectors. Left alone, every one of those connections would be switched on all the time. That is slower, and it gives the assistant more reach than any single task needs.
mcp-patchbay keeps them all off. When a task needs one, the assistant asks for it and it appears. For the ones we mark as sensitive, such as anything that can spend money or send a message, only a person can say yes.
Setting it up takes a few minutes: one private file that lists your connectors, then one command to add it to Claude Code. The Quick start in its README walks through each step.
mcp-frontdesk knows who is at the door
When you connect District AI to Claude, you see a page that says what the assistant is asking for, and you approve it. Behind that page is a sign-in server. It checks who is asking. It gives the assistant a pass that expires in minutes. And it lets you take that pass back from Connected apps in your settings whenever you like.
mcp-frontdesk is that design, packaged so any developer building a connector can use it. It comes with a small working example, because the fastest way to trust a sign-in system is to run one and watch it work.
npm install mcp-frontdesk
Why we keep doing this
You can check us. A phone company holds things you cannot see. Code you can read is a promise you can check.
It makes the work better. Code that strangers will read gets written with more care. Every project we publish has its security policy and its list of open issues in public, and every tool and app has its changelog.
We owe it. District AI runs on software that thousands of people wrote and gave away. Publishing our own tools, and sending our fixes back to the projects we build on, is how that keeps working.
What this means for you
If you are a customer, nothing changes and there is nothing to do. What grows is the list of things you, your IT person or a client can check for themselves.
If you build software, all three tools are free to use under the Apache-2.0 licence, and none of them needs a District AI account.
All fourteen projects are on the open-source page.
Frequently asked questions
Is District AI itself open source?
The service is not. District AI is our commercial service. Its apps and the tools around it are: fourteen open-source projects, listed at distronode.com/open-source.
Do I need to do anything as a customer?
No. Nothing about your service changes.
What is worldinmotion?
A tool that lets AI agent sessions send each other encrypted messages in real time, across machines and between agents from different makers.
What is MCP?
The Model Context Protocol is an open standard that lets AI assistants such as Claude connect to other software. A connection built on it is called an MCP server, or a connector.
Can I use these tools without being a Distronode customer?
Yes. All three are free to use under the Apache-2.0 licence, and none of them needs a District AI account.

