BMD
A work queue for your AI agents.
Keep project context, queues and agent runs in one Windows workspace. Review the saved result before you ship it.
Free Windows beta. No card.
Try BMD
BMD Pat LLC · Nashville, TN
Patrick Hughes
Veteran-Owned Small Business
I build AI workflows and business software.
I want to leave work running and come back to a result I can inspect. That means keeping the instructions, the run and the evidence together. BMD grows out of that work.
Build useful things. Keep time for life.
My background is IT infrastructure, software engineering and AI. I help businesses and IT partners build AI workflows, connect systems, and develop business software. Azure is a specialty, from migrations to cloud build-outs and repairs. I also subcontract with Microsoft consultancies and managed IT providers. I care about what happens after a demo: a failed run, an unexpected bill, an unclear result, or work that needs a person to decide.
I run BMD Pat with AI agents. They help with research, code and recurring checks. I own the product decisions and the standards for what gets published. Software should give me more time with my family, not another system that needs constant attention.
I write about the work as I build it, including the failures. I also play Dota.
The workshop
What I build
AgentGuard, showwork and BMD grew out of my own engineering work. You can try each one without adopting the others.
AgentGuard
Put limits around agent operations.
A Python library for budgets, call limits and timing checks at the points where you invoke the guard.
A timeout check does not cancel an in-flight request. Recorded cost can exceed a limit on the request that crosses it.
Try AgentGuardshowwork
Check what a completion claim proves.
Declare the result you expect, run a check, and keep its evidence with the work.
A passing check proves only what it tests. Per-result checks inside BMD are still planned; the standalone tool works separately.
Try a showwork checkHow I work
A queue makes the next action clear. A saved result makes review possible. A check gives me a specific claim to accept or reject.
Define the result
Write down what should change and how to check it before starting a run.
Review the evidence
Run the relevant tests. Check the actual page or output. Keep failures visible.
Make the next run better
Turn a repeated problem into a clearer instruction, a test or a smaller task.
More of the work
The local AI tools and published GPU measurements remain here for people choosing models for their own hardware. They are a separate path from BMD.
Contact
Talk to me directly
Have a workflow, integration or cloud project in mind? Start with the kind of help you need. I will reply personally about the goal and next step.