Opening
Hi.
I'm Haruyoshi Hattori, founder of coiai, in my first year running the company.
coiai is an IT company that builds systems and products with technology at the core.
Every Monday, I use this show to share what running a startup is really like.
This week's tech news
This week's tech news
entire
Announced on February 10th by GitHub CEO Thomas Dohmke.
The idea seems to be that sessions with AI agents get saved into Git as well.
Personally, I've been saving my AI sessions in a directory I made called "diary," but I hadn't thought to tie them together with Git commits, so I honestly thought this was a great idea.
References
https://zenn.dev/aromarious/articles/20260223-entire-io-review
https://github.com/entireio/cli
AI Pyramid
A friend of mine bought an AI Pyramid, and we tried installing Openclaw on it.
The AI Pyramid is built on top of the M5Stack hardware, which is aimed at geeks who want to tinker with hardware easily. I won't go into the details of M5Stack since that's a tangent, but think of it as being talked about in the same breath as Arduino — that should give you a sense of it. It's the kind of gadget where you keep adding modules to expand it, which tickles that inner-kid curiosity.
Back to the point — the AI Pyramid appears to be an M5Stack with an added NPU for machine learning.
Because of that, its main feature is being able to run an LLM locally.
What we struggled with for a while was memory allocation. Not just installing Openclaw — even installing Git, pnpm, or npm would freeze up, and we kept thinking, something's off here…?
Turns out that, by default, only 2GB of RAM is allocated (the rest of the memory is reserved for machine learning via something called CMM), which made things run heavily.
So, for the initial Openclaw setup, we decided to allocate 7GB of RAM to work with instead.
Another thing we struggled with — not sure if this is a hardware issue — was picking a "channel," meaning a chat service to use for conversation. Services like Discord kept showing up as "not configured, plugin disabled."
We solved this by installing the plugin separately for each one.
We finally got it up and running, but we didn't manage to get the local LLM working as an API before we ran out of time and had to call it a night.
I want to give it another shot next time we get together!
References
https://docs.m5stack.com/ja/stackflow/ai_pyramid/memory_adjustment
Ending
All right, that about wraps up this week.
coiai works on web, XR, core systems, and on-premise AI adoption support.
We can take on pretty much anything, so if you've been turned down elsewhere or gotten a quote that felt too high, feel free to reach out — quotes are free!
See you again next Monday! Subscribe to the channel and I hope you'll come listen again next week!
Thank you!