Automated Performance Profiling with MicroProfiler and Assistant

This topic was automatically opened after 10 minutes.

This is a great feature, as using the microprofiler is sometimes tedious for fully understanding issues. However, most of my microprofiler diagnosis is not in Studio but is instead via creating a dump from a live game instance. Does support for uploading a microprofiler dump from your files to assis

Yes! You can use MicroProfiler’s top menu → Dump → Dump in binary format , then tell your agent: “Analyze this MicroProfiler dump. Here’s a SKILL file for it.” It will download Lute and LibMP (if they haven’t been downloaded already) and parse the data for you. In the future, you’ll also be able to

So, I capture a thing. Not sure how, I guess I time enter button or something. But let’s say I give AI a capture. Now, will the AI know what all these labels mean, like the Engine ones that aren’t documented or will it make assumptions Once I wanted to document the missing labels, but it was rejecte

ZenMa1n: Under the hood, this utilizes the MicroProfiler API (LibMP) that we recently announced. With this API, you can programmatically access every piece of data in the MicroProfiler. Be sure to check it out if you’re an advanced developer looking to dive deeper! So what’s the point of releasing L

It will follow the public documentation, so it doesn’t have any additional secret knowledge beyond that. We do need to add more documentation about the scope names and their meanings so this automated approach works better and developers have more insight as well.

ZenMa1n: We do need to add more documentation about the scope names and their meanings so this automated approach works better and developers have more insight as well. But they said https://github.com/Roblox/creator-docs/pull/1495 For the MP, we’re not trying to catalog every label since many or ev

Kitsune: What I’m asking is which one do I use for which use case? They seem like they are the same thing so I’m not sure which to use for what There is no difference - they’re the same. LibMP is the MicroProfiler API. Since it can be used both inside and outside the Engine, it was given a separate

I usually dump it onto my pc & then export as JSON, gets the job done + any AI can evaluate it

Yep, now imagine capturing an entire 20-minute gameplay session as a 4 GB capture on disk (oh, the dream). Do you really want to export it to JSON, inflate its size even further, and then feed it into an LLM? Exporting to text works well for aggregated data analysis (without in-depth digging), or fo

this is a nice usage of ai, less chores, higher quality development process

I think the main benefits of AI is its ability to read a bunch of junk data a developer would take far too long analyzing, and put meaning behind it, multiple times I’ve just over printed every single thing in a script, and slapped the result into an LLM so it can find meaning in it, and tell me wha

Awesome! Is there any plan to allow MCP to spin up separate test servers akin to how “server & clients” does? Basically so multiple claude sessions can spin up test sessions in parallel?

ZenMa1n: We do need to add more documentation about the scope names and their meanings so this automated approach works better and developers have more insight as well. Yes please! This is exactly what I was hoping to read in this post.

Assistant do really live up to their name especially in a very tedious task like looking for issues in microprofiler. Amazing work.