Cogni Code — AI skill for Claude Code

Cogni-Code

Every time you start a new session, your agent starts from zero. It forgets your preferences, your decisions, the bug you spent two hours on yesterday, the deployment strategy you settled on last week. Cogni-Code fixes that.

It is a persistent, inspectable memory system for Claude Code, Codex CLI, OpenCode, pi, and any MCP-compatible agent. Your agent learns how you work across sessions and gets sharper every time you use it. Memory lives as plain files on your disk. You can read it, edit it, diff it, and back it up with git.

---

Contents

• What makes it different
• Install
• Quick start
• How it works
• The memory model
• The background pipeline
• Notion sync
• Skills
• Dashboard
• Tool reference
• Slash commands
• Where memory lives
• Project structure
• Read next

---

What makes it different

Most agent memory falls into one of three camps:

• Built-in memory (ChatGPT memory, Claude's saved context) is opaque, vendor-controlled, and locked to one product.
• Vector-DB memory (mem0, Letta, Zep) stores embeddings in a database. Powerful retrieval, but you cannot read what your agent knows without a UI.
• Hand-written context files (CLAUDE.md, AGENT.md, .cursorrules) give you full control but require you to write and maintain them by hand.

Cogni-Code is a fourth