Agent Kernel gives coding agents memory

This is the right answer to agent amnesia: use git and markdown before reaching for a database or framework.

• –The memory model is elegantly simple: `IDENTITY.md`, `KNOWLEDGE.md`, `notes/`, and `knowledge/` make state easy to inspect, diff, and back up.
• –Supporting OpenCode, Claude Code, Codex, Cursor, and Windsurf lowers adoption friction because it fits tools people already use.
• –The split between mutable facts and append-only session logs is a smart compromise between durability and auditability.
• –The trade-off is obvious too: repo-based memory scales best for solo builders and small teams, while larger agent fleets will eventually need permissions, schema, and conflict handling.
• –It solves a real pain point in coding agents: every new session should not feel like a total reset.