oguzbilgic/agent-kernel: Minimal kernel to make any AI coding agent stateful. Clone, point your agent, go. · GitHub

The easiest way to create an AI agent. Clone, start, talk.

Your agent remembers between sessions, takes notes, and builds on past work. No framework, no database — just three markdown files and a git repo.

Works with any AI coding agent: OpenCode, Claude Code, Codex, Cursor, Windsurf, etc.

git clone https://github.com/oguzbilgic/agent-kernel.git my-agent
cd my-agent

Start your agent:

opencode     # or claude, codex, cursor, etc.

That’s it. The agent reads the kernel, realizes it’s new, asks who you want it to be. You tell it. It remembers.

AGENTS.md          ← kernel (generic, don't edit)
IDENTITY.md        ← who this agent is (agent maintains)
KNOWLEDGE.md       ← index of knowledge files (agent maintains)
knowledge/         ← facts about the world (mutable)
notes/             ← daily session logs (append-only)

Two kinds of memory:

• knowledge/ — State. Facts about how things are right now. The agent updates these when reality changes.
• notes/ — Narrative. What happened each session — decisions, actions, open items. Append-only. Never modified after the day ends.

AI agents already read AGENTS.md (or CLAUDE.md, .cursorrules, etc.) as project instructions. This kernel uses that mechanism to teach the agent how to remember.

The agent doesn’t need a database, a vector store, or a custom framework. It just needs:

• A file that says “you are stateful, here’s how”
• A git repo to store memory in
• Plain markdown files

Each agent is its own repo. To create another:

git clone https://github.com/oguzbilgic/agent-kernel.git another-agent
cd another-agent
opencode     # or claude, codex, etc.

Same kernel, different identity, different knowledge. You can have a homelab agent, an investing agent, a health agent — all running the same OS.

MIT