Memcrate memcrate

Early days. Memcrate ships Claude Code integration today. Cursor, Cowork, Aider, and an MCP server are on the roadmap — see the repo for status.

The vault

The Memcrate vault is a folder of markdown on your disk. That’s it — no database, no server, no proprietary format. Every AI tool reads it through its own integration; the source of truth is a folder you own.

Shape

vault/
├── Core/
│   ├── Context/
│   │   ├── Profile.md          # who you are
│   │   ├── Projects.md         # what you're building
│   │   └── Current State.md    # this week
│   └── Sessions/
│       ├── 2026-05-13-foo.md   # session logs
│       └── 2026-05-12-bar.md
└── .memcrate                   # vault marker

Five files plus a marker. That’s the minimum viable vault. You can add optional folders (Projects/, Daily/, Tasks/, Inbox/) alongside Core/ — the canonical files live in Core/ so the spec is small and stable.

The canonical files

Core/Context/Profile.md

Who you are. Role, expertise, working style, tools you use, anti-goals. Stable across weeks and months. This is what /load reads first.

Core/Context/Projects.md

Every project you’re working on, with status, stack, and where it lives. One H2 heading per project. Stable across days and weeks — bump when something material changes (new project, status flip, archived).

Core/Context/Current State.md

This week’s focus, active deadlines, recent decisions, open questions. Stable for days. This is the one file most likely to be stale; /pin tends to write here.

Core/Sessions/*.md

Compressed session logs. Each file is a record of one work session — decisions made, files touched, what was pending at the end. /save writes these. Filename convention: YYYY-MM-DD-slug.md.

Frontmatter

Every file has YAML frontmatter. The CLI and skills read these fields when relevant.

For context files:

---
title: Profile
last_updated: 2026-05-13
---

For session logs:

---
type: session
date: 2026-05-13
projects: [Memcrate, BradTraversyDev]
topics: [docs, deployment]
tool: claude-code
outcome: Shipped the docs site to Vercel.
---

projects and topics are arrays of labels; scoped loads (/load memcrate) use them to find relevant past sessions.

Where the vault lives

The CLI and skills look for vaults in this order:

  1. An explicit path passed to the CLI (memcrate init ~/myvault)
  2. The current working directory (if it has a .memcrate marker)
  3. ~/vault (the default)
  4. A scan of ~/*/Core/Context/Profile.md (catches custom paths)

So if you want a custom location, just run memcrate init ~/mything once and the skills will find it.

Designed to be tool-agnostic

The vault format doesn’t depend on any AI tool. You can:

  • Open it in Obsidian for backlinks and search
  • Edit it in VS Code, Vim, or nano
  • Sync it with Obsidian Sync, iCloud, Dropbox, or git
  • Grep it with normal shell tools
  • Point any AI tool at the canonical files — even one without a Memcrate integration will benefit

The skills (Claude Code today; more later) are conveniences. The vault is the contract.