Agent Hub
Task queue coordinating AI agents across two machines via GitHub.
- Role
- Builder
- When
- 2026
- Stack
- GitHub Issues, Windows, macOS
Breakdown
I run AI agents on a Mac and on a Windows PC. There was no safe, visible way to hand work from one machine to the other without copying private data around.
- Two machines, no shared queue
- Copying whole workspaces leaks private data
- Crashed jobs must not block the queue
Use GitHub Issues as the task bus. The PC checks the queue every five minutes, claims exactly one job, and releases anything stuck for more than two hours. Labels show state, priority and where a task came from.
- One job per run keeps it predictable
- Stale jobs recover automatically
- Everything visible and auditable in GitHub
Built
- Dispatcher running on the Windows machine
- First task completed end to end
- Workspace transfer through GitHub rejected on purpose to protect private data
Not yet
- Direct connection between the machines prepared but not switched on
How it works
A task is created as a GitHub issue with labels.
The PC picks up exactly one queued task.
The agent runs the task on that machine.
Result posted back to the issue, stale jobs released.