Added a live vote on kernelgarage.com for what to build next — an LLM-written summary of site activity, a Jarvis-style voice assistant, an AI reviewer for pull requests, and more. What do you think, any of these worth building, or any other suggestions? Reply and let me know.

Follow-up to the MCP usage-report tool: it now logs the exact PromQL queries + raw values behind every finding, a claim like "queue wait jumped to ~6.7min" is checkable now, not just asserted. Kept it read-only no new way to touch the Pi.

#selfhosted #mcp

I tested my first MCP server, built to report on my Pi's usage. Hooked it into Claude Code and just asked. A follow-up for the last week made it notice a real queue problem on its own, unprompted.

#python #selfhosted #llm #opensource #homelab #mcp #claudecode

I built agent-monitor.

Send a burst of requests and watch it drain back down. Real token counts too.

The API sits between the call and Ollama.

What do you use to watch your own local models?

Code: github.com/kernelgarage/pi-monitoring

#SelfHosted #Observability #LLM #Homelab #MLOps #Grafana

Full Grafana stack (Prometheus, Loki, Tempo) watching the Pi at idle before I load anything else on it — ~1% CPU, under 50°C baseline.

Curious what people use for hardware monitoring vs. LLM/agent monitoring (MLflow has some built in) — what are you running?

#selfhosted #observability

Been running a Pi 5 as my Home Assistant server for a while. Now trying to turn it into a local AI assistant too. What's everyone actually using for this? Is OpenClaw still a thing, or has something better shown up?

#selfhosted #localai