Skip to content

AI Command Interface — Project Overview

🧪 Partially validated — project status

The Windows prototype implements local chat, telemetry, and authentication. Selected Proxmox and HTTPS integrations have been exercised in a nested lab; physical-host validation remains outstanding.

The AI Command Interface is a unified dashboard, backend API, and Discord bot designed to manage local AI models, system telemetry, GPU inference, and multi-agent workflows. It is optimized for AMD hardware (RX 6800 XT) and intended to run on a dedicated homelab server.

Components

  • Frontend Dashboard (HTML/CSS/JS)
  • FastAPI Backend (Python)
  • Discord Bot (Python)
  • Model Registry
  • System Monitor
  • AI Interface (single admin-shared server console)
  • AI Chat (private conversations and web research)
  • Local LLM Chat (llama.cpp test integration)

Current Status

The Windows build/prototype connects dashboard AI Chat to a local llama.cpp Qwen2.5 1.5B Instruct model and reports live RX 6800 XT load and VRAM telemetry. Dashboard authentication supports local accounts and explicitly mapped Proxmox identities, with protected sessions and role-gated APIs. Proxmox API integration and HTTPS proxy access have been exercised in the nested lab; physical-host validation remains outstanding. Installed GGUF files are discovered and can be selected for model switching; the backend confirms which model is active and attempts recovery if a switch fails. The deployment plan now proposes a resident Qwen2.5 1.5B Utility AI alongside a Household or Maintenance profile model. This multi-model residency and request routing are not implemented; the current model manager runs one llama.cpp model and restarts it to switch. Discord heartbeat/status, operational logs, restart, command listing, and model selection are connected to the optional bot service. Vision and image generation still use placeholders. The VM panel has a backend-provided, read-only Proxmox VM/container inventory when configured; VM management, game-server, and network-service integrations remain unimplemented. Linux AMD GPU telemetry has a kernel-sysfs provider with deterministic tests; readings still need validation on Linux hardware. ROCm runtime support remains unverified. Windows remains the active build environment for shared application behavior. An additional Proxmox VE 9.1-1 deployment/integration lab is running inside a nested Hyper-V virtual machine on the Windows PC; the dashboard and backend have been deployed there with loopback-only service ports and an NGINX HTTPS proxy, tested from LAN clients as well as through operator access. This lab can support Linux/Proxmox implementation work, but GPU/ROCm, physical hardware, production networking, and physical-host recovery behavior still require validation on the eventual server. Nested-lab boot recovery and LAN HTTPS proxy checks have passed; see the transition runbook for scope. The AI Interface has a single admin-shared, attributed transcript; private AI Chat conversations remain scoped to each account.

Goals

  • Provide a unified control interface for local AI systems
  • Support multiple models and inference backends
  • Extend real GPU telemetry and inference to the Linux/Proxmox host
  • Provide Discord bot control and monitoring
  • Run on a dedicated AMD-powered server