Local AI Manager
Sets up open models on your own GPU PC or Mac, keeps them running, routes your chat, code, image and voice jobs to them, and radars new models monthly.
Overview
Run your own AI, done right. Local AI Manager sets up open models on your own hardware, keeps them running, and handles the day-to-day usage of your AI: you tell Grok Bot what you want to do, and Grok Bot communicates directly with your machine and your local model to do it for you. Chat, code, images, video, and voice all run on a stack you own: private by default, yours to tinker with, and ready for the next model the day it drops. What you get: - Day-to-day AI operator: tell Grok Bot what you want done (summarize private files, draft replies, run batch jobs, or chat), and it talks directly to your machine and your local model to run the work and bring back the results - A read-only hardware scan with live Hugging Face and Ollama model matching for light (25%), balanced (50%), or max (90%) load, tailored to what you describe wanting to build or run - A guided install of the right runtime and model, one approved step at a time, with a clean uninstall and a copy-paste local OpenAI API snippet - A Monthly Model Radar on the 1st of each month that evaluates newly released open-weight models against your hardware and use case, plus on-demand side-by-side A/B model bake-offs on your own prompt - Image, video, and voice generation set up for your GPU - Task handoff to a job folder on your computer, with a weekly estimate of the cloud usage you saved - 1-click Pause / Game Mode to unload GPU VRAM and snooze background jobs and alerts for 2 hours when you want your machine back for gaming or heavy work - Watchdog alerts and crash recovery if the model goes offline or runs hot: it reads the logs, diagnoses what broke, and repairs it after your yes - A sleek, private, dark-themed dashboard on 127.0.0.1 that you can check from your phone through Tailscale Serve - Lock-down rules you set: local-only or not, which folders it can touch, and what needs approval Built for local-AI tinkerers, privacy-minded users, and small businesses who already own capable hardware. You need an NVIDIA GPU PC, an Apple Silicon Mac, or a DGX Spark-class machine, registered with Grok Bot. Every command asks first. Monitoring and the dashboard are new. The bot tells you plainly which parts are still being tested.
Contents
- skill local-ai-manager-getting-started Use when the owner sends any message while setup is unfinished (even "hi"), says "set me up", "resume setup" or "start over", or asks what their computer can run. Runs the resumable one-question-at-a-time setup, hardware scan, live public model evaluation, and 25/50/90% fit report.
- skill local-ai-manager-containment Use before the first install or download, when the user asks what the local AI may access, wants to change privacy or approval rules, and at the start of every turn that touches their computer (to load and obey the saved rules).
- skill local-ai-manager-install Use when the user approves a recommendation, or asks to install, switch, upgrade, or uninstall a local runtime or model (LLM, image, video, or voice), or asks for the right settings or prompt format for a model.
- skill local-ai-manager-local-chat-and-handoff Use when the user wants to talk to their local model, send it a task (summaries, sorting, drafts, private files), check a handed-off job, set up or change the job runner, or add a new feature or skill on top of the local model.
- skill local-ai-manager-dashboard Use when the user wants a local dashboard for their AI (status, temperatures, VRAM, tokens per second, jobs, alerts, Start/Stop/Restart), wants it on their phone, or wants it changed or removed.
- skill local-ai-manager-monitoring Use after install to offer monitoring and the monthly model radar, when the scheduled health-check or monthly model-radar routine fires, when the watchdog webhook routine receives an alert, or when the user asks to check, restart, or stop monitoring.
- skill local-ai-manager-weekly-savings-report Use when the weekly savings routine fires, or the user asks how much cloud usage their local model saved, wants the report set up, changed, or paused, or wants to set the cloud price used for the estimate.
- skill local-ai-manager-troubleshooting Use when the local model is slow, fails to load, runs out of memory, gives empty or truncated answers, ignores the GPU, or when ComfyUI, vLLM, or a video job errors.
Dependencies
No plugin dependencies are declared.
Attachments
No separate attachments are published for this record.