Kernel
My personal AI operating system — running on my Mac, live in my Slack, in daily use for the past four months. It spawns workers, files research to Obsidian, reads my calendar and email, and surfaces what matters. Not a prototype. Not a concept. This one I actually depend on.
The Problem
AI tools are great at individual tasks but terrible at continuity. You can ask Claude to write code, but it doesn't know about the plan you made yesterday, the bug you filed this morning, or the three things you need to ship by Friday. Work happens across Slack, terminals, files, and your head. Nothing ties it together.
The Build
Kernel is a Claude SDK session running as a persistent process on my Mac. It receives messages via Slack, processes them through an intent classifier, and dispatches work to Claude Code worker processes. Workers run autonomously — building code, running tests, reading books, filing research to a knowledge graph. Results surface back in Slack. The whole system runs unattended: nightly reading sessions, morning briefings, GitHub cleanup, demo testing — all without my involvement. 120+ days of daily use, 1,000+ workers run, 148-node knowledge vault.
What Makes It Different
It's not a chatbot wrapper. It's infrastructure. Kernel spawns autonomous workers, manages their lifecycle, tracks costs, logs decisions, and maintains operational memory across days and weeks. The security model handles unattended agent execution — not just 'I'm at the keyboard' assumptions. It's the difference between a tool and a system. The most interesting part for Anthropic: it's Claude orchestrating Claude. The Kernel session (Claude SDK) spawns Claude Code workers, monitors their lifecycle, reads their output, and chains follow-up workers when needed. It's a real multi-agent system that's been running in production for 4 months.