Has anyone here had a chance to work with this and have freedom to share thoughts?
4 Likes
Fascinating - would love to participate
1 Like
No I havent used it yet, but if you want to register to test this: https://www.modelhardwarestandard.com/
most of this is just leveraging pylabrobot + MHS, at least the Hamilton portions of the video. You can get real close right now with pylabrobot
4 Likes
@cwehrhan @yhs.allen the difference between your responses is actually where my question comes from. I have written a few drivers easily enough that they didn’t seem worth contributing (admittedly for pretty simple devices) so programmatic control given docs seems pretty easy for an ideal AI agent to achieve for most capabilities that matter. I’m wondering what else this standard does.
@yhs.allen - Could you describe it in a bit more detail? I’m very curious too.
Specifically, I’d really like to hear your take on how it differs from something like SiLA. My basic understanding (from the anthropic blog posts) of MHS is that it’s an abstraction layer that combines the device-specific communication and MCP / CLI library.
SiLA doesn’t natively create an MCP library, but it’s easy to build in as a custom feature because the device-specific communication/commands are wrapped into SiLA’s gRPC communications. So a consumer/client/agent could read the structured gRPC command definitions, or you could also wrap and expose them as an MCP library. Is there more in MHS?
@pud - I think you’re right, the device communication is becoming “solved” (still hard though, and solved only through the constant work of this community). But the next level up - scheduling, coordination of multiple devices, protocols & system-state awareness - those are still real issues.
1 Like