Agent mesh: our agents hand work between machines
A broker that lets agents on a Mac, a Windows PC and a server pass tasks to whichever machine suits them.
- AI agent
- In-house R&D, prototype
- In-house prototype
At a glance
Client OutGrow
Client
OutGrow
Industry AI infrastructure
Industry
AI infrastructure
Year 2025
Year
2025
Status In-house prototype
Status
In-house prototype
Every case on this site answers the same four questions, so you can compare them.
Before
Each agent was stuck on one machine. It couldn't reach desktop files, start work on a server, or borrow a GPU somewhere else.
What it does now
A broker on a VPS keeps a registry of connected machines, with heartbeats.
A small worker runs on each Mac, Windows PC or Linux server.
An agent splits a goal into steps and sends each to the right machine.
Heavy model work goes to the Windows GPU, and the result comes back.
A dashboard shows which workers are live.
Where a person decides
It is an internal tool. We start every workflow ourselves and watch each step on the dashboard.
Result
In-house prototype. No outcome figures yet.
How we built it
Built the broker with Node.js, Express and WebSockets.
Added the agent registry and heartbeats.
Wrote the headless worker for Mac, Windows and Linux.
Wrote the routing skill agents use, and the live dashboard.
Stack
- Node.js
- WebSockets
- Express
- SQLite
- TypeScript
- systemd
From the project
More work like this
Every inquiry read, scored and routed, at any hour
BigDeal AI: a real estate CRM with an agent that asks before it acts
Our whole operation, readable and writable by any AI model
Your case study starts with an audit.
One to two weeks inside your operations. You get a map of where time and money leak, the return on each fix, and a fixed scope before we build anything.
Most businesses we meet have paid for an AI tool nobody uses. We start from your numbers, so the first build is chosen by payback.
Zain Khan and Anwar Sheikh
Founders, OutGrow
Outgrow
Intelligence
Studios.