MyJarbis

Persistent memory + per-project workflow + scoped skills for Claude Code.
Two slash commands: open and close. Everything else is a conversation.

Run myjarbis start from anywhere → pick a project → it opens a terminal already running Claude + /jarbis.

---

In plain terms

AI coding assistants forget everything when you close them. Every morning you re-explain the same things: how this project is built, the rules the team follows, what you were in the middle of yesterday.

MyJarbis is the notebook the assistant never had. It remembers:

• the decisions — what you chose and why, so nobody re-opens it later
• the traps — the thing that broke once, so it doesn't break twice
• where you stopped — so tomorrow starts with "here's where you were", not a blank page
• how your team works — the review rules, the release checklist, the order things get done in

It also *enforces* that last one. If the team's process says "someone else approves before this merges", the assistant stops and asks instead of deciding on its own.

Everything is stored on your own machine, in a plain database file. Nothing is uploaded anywhere.

The rest of this page is for developers.

---

What it is

Claude Code has no project memory. You repeat conventions, re-explain architecture, and lose "what was I doing yesterday."

**MyJarbis is a persistent-memory layer over MCP (Model Context Protocol).** It turns Claude Code into a project-aware orchestrator that remembers what you decided, where you left off, and which workflow applies to each part of the project — without dumping everything into context every session.

Every read/write goes through standard MCP tools + a local SQLite DB, so **any MCP-co