Now I don't have to wonder if my Local SI rig is working on the prompt or I have to ask it to continue.
Idle = yellow
Working = green
Good to see what each GPU is doing.
Rewatching Project Hail Mary tonight after seeing it only once in theaters.
Held off until it could be watched on a bigger, better TV.
Got a working alpha of Telegram replacement.
Had Grok and local SI combine agent chats with my dashboard.
Using Cloudflare Tunnel
Qwen 2.1 made an interesting tray from different angles in a single image.
Still trying to get my agent to make one based on reference.
I need to look into a better vision model, see if that helps.
Check your harness prompts.
I had thousands of wasted tokens that agents would go through on every interaction with tools and general instructions.
I told kiki to design a more unique tray.
Now I'm having Tessel model it in Blender. We'll see how that goes.
When I had AI setup my Intel B60s it failed to use SYCL
So my tok/s were atrocious. Now less so.
4-10-17
Haha. My agents comment after making this.
"My internal QA model (a 3B vision model with very conservative taste) flagged that it "doesn't look like a typical album cover because it's dark and eerie" — which, for a giant monster behind an abandoned asylum, is exactly the point. Tool would be proud."
Local SI rig -> Telegram message to Kiki (agent running on openclaw, Qwen 3.8 q4 model) to create an image.
Kiki then makes a prompt and sends it to qwen 2.1 image running on same rig. Sending the results back when done.
Grok imagine animated
🤖 Made with AI











