We’ve been building atop the .agents standard and have written some tools around this under the ai-outfitter org few interesting repos.
- outfitter compose different agent profile w, eg skills, mcp, context and wrap pi,Claude,codex
- agent-operator run these agents into kubernetes
- actions run pi during investigations on failed ci or weekly updates
- channels give pi agents an inbox for slack/email/websocket access, calendar wake ups
- deepwork build long running workflows with verification gates
The best example of different profiles is having a prod/nonprod bot that has grafana mcp for incident investigation
One of my favorite joys is going to the library in the Museum of fine arts and reading (usually with the help of Google Translate) really really old books. One of which was “La science des medailles” written between 1600-1700. It talks about the science of medals (which are much like coins!)
Seeing how people wrote back then and what they where thinking about (without the internet) always impresses me!
At a career fair in an auditorium I was showing a sensor suite with OLED displaying current readings, one of which was the SCD41 CO2 sensor (they cost more than the esp32!)
I watched the sensor rise from 800 to 1700 PPM by the time the last group left the
It’s quite easy to build one and deploy with esphome and breadboard with stuff you can order on Amazon and have an LLM walk your through hardware and setup.
It is interesting where the rate of speech quickens as the co2 rises and the body starts to notice the co2, or maybe that was just the coffee.
I have been building a similar concept into my custom NixOS distribution, Keystone, where agents operate within their own user accounts with dedicated emails and SSH access.
> It utilizes the Claude, Gemini, and Ollama CLIs. Because it is built directly into the OS, it seamlessly integrates with native notes and records calls. Furthermore, an AI agent can access Immich to deduce my context by analyzing image metadata and tagged faces. It features dedicated calendars for task scheduling and native PDF extraction capabilities. The entire system is declarative via NixOS, allowing it to provision itself almost entirely automatically.
I kinda suspected this was an early way to catch AI generated content. It ironically broke stalwart/himilaya somewhere along the lines when I had an ai generate a status report to email to me
- outfitter compose different agent profile w, eg skills, mcp, context and wrap pi,Claude,codex - agent-operator run these agents into kubernetes - actions run pi during investigations on failed ci or weekly updates - channels give pi agents an inbox for slack/email/websocket access, calendar wake ups - deepwork build long running workflows with verification gates
The best example of different profiles is having a prod/nonprod bot that has grafana mcp for incident investigation