v1.1.9 — x86_64 (AppImage, DEB, RPM)
| OS | Ubuntu 22.04+ / Debian 12+ / Fedora 38+ or any modern 64-bit Linux distro |
| Architecture | x86_64 (64-bit) |
| RAM | 4 GB minimum — 8 GB recommended |
| Disk space | 300 MB free |
| USB | USB port for serial communication |
| Dependencies | WebKit2GTK 4.1 (preinstalled on most distros; if missing: sudo apt install libwebkit2gtk-4.1-0) |
The serial terminal, macros, triggers, scripting, and programmer work without any GPU. Running your own local model through AI Assistant (BYOK) is the demanding part and needs a high-VRAM GPU. During testing, models below roughly 20B parameters could not drive the assistant's tools reliably, 20-30B is the lower edge of usable, and confident results start above 100B.
| GPU | Any Vulkan-capable GPU — NVIDIA, AMD (recent), Intel Arc |
| Entry (below usable) |
16 GB VRAM — 20B-class models (Q4); below the range where the assistant works reliably |
| Recommended | 24 GB VRAM — 20-30B models (Q4-Q5), longer context; lower edge of usable |
| High-end | 48 GB+ VRAM (single card or two of them) — 70B and larger; 100B+ (Q4) needs about 64 GB |
| Drivers | A graphics driver with Vulkan support — NVIDIA proprietary driver, or Mesa for AMD and Intel |
CPU-only is technically possible but extremely slow at these sizes — a dedicated GPU is strongly recommended. During testing, models below roughly 20B parameters could not drive the assistant's tools reliably.
The assistant can be wrong. Language models misread frames, invent register numbers and write plausible code that does the wrong thing. Read what it proposes before you send it to a device — a bad Modbus write or a wrong firmware flash can damage equipment or stop a machine.