Agent Kernel

Agent Kernel is a minimal way to make any AI coding agent stateful, using three markdown files and a git repository instead of a framework or database. You clone the repository, start a coding agent inside it and talk to it; the agent reads the kernel, recognizes that it is new, asks who it should be, and from then on remembers between sessions, takes notes and builds on past work. It is aimed at developers and individuals who want a persistent personal agent, such as a homelab, investing or health assistant, built on tools they already use. It works with OpenCode, Claude Code, Codex, Cursor, Windsurf or any coding agent that reads AGENTS.md as project instructions, since the kernel uses that file to teach the agent how to remember. The memory structure is plain text: AGENTS.md holds the generic kernel, IDENTITY.md describes who the agent is, KNOWLEDGE.md indexes a knowledge folder of mutable facts that the agent updates as reality changes, and a notes folder keeps append only daily session logs of decisions, actions and open items. Each agent is its own repository, so several agents can share the same kernel with different identities and knowledge. Agent Kernel is free to clone from GitHub, and the repository does not declare a license. The only cost is the coding agent or model you already pay for. The same author publishes kern, a companion runtime that runs a kernel based agent as a long lived process reachable from the terminal, Telegram, Slack and other chat channels, with daemon mode, user pairing and backup and restore.