The verbs

Three commands, three jobs. Run them inside any supported AI tool; the syntax is the same everywhere.

/save

Compress the current session into a structured log.

When you’re wrapping up a work session, run /save. The AI reviews the conversation, extracts what was decided / learned / left pending, and writes a markdown file to Core/Sessions/ with a standard schema.

A session log includes:

• Frontmatter — date, projects, topics, tool, outcome
• Quick reference — the one-line summary
• Decisions made — choices worth remembering, with the reasoning
• Key learnings — non-obvious things you discovered
• Files touched — what changed and why
• Pending / next actions — the open todo list
• Raw log — narrative summary of how the session went

Filename: YYYY-MM-DD-slug.md.

When to use it: at the end of any session where you’d want to remember what happened. Skip it for trivial sessions (a quick answer, a typo fix).

/pin

Promote an insight into permanent context.

/save records what happened. /pin decides which parts are worth keeping forever.

Use /pin when something from the conversation belongs in your long-term context — a new project, a status change, a preference you want every future session to know about. The AI figures out which canonical file the insight belongs in (Profile, Projects, or Current State), edits it, and bumps the last_updated field.

When to use it:

• You just decided to start a new project → pin it to Projects
• Your role or working style changed → pin it to Profile
• Something this week’s plan changed → pin it to Current State
• You learned a non-obvious working preference you want remembered → pin it to Profile

The threshold is “will I want a fresh AI session next week to know this?” If yes, /pin it.

/load

Load operating context at the start of a fresh session.

When you open a new AI session, run /load first. The AI reads:

1. Core/Context/Profile.md
2. Core/Context/Projects.md
3. Core/Context/Current State.md
4. The three most recent session logs from Core/Sessions/

Then it gives you a short oriented summary: who you are, what you’re working on, what happened last session, what’s pending.

You can scope a load with an argument:

• /load 10 — pull the last 10 sessions instead of 3 (deeper history)
• /load memcrate — scoped load: only the slice relevant to one project or topic

Scoped loads grep the session logs for tagged matches (projects: or topics: in frontmatter) and pull just the relevant context. Useful when you want to focus on one thing without the rest of your work as background noise.

When to use it: every time you start a new session and want the AI to know you. Skip it for one-shot tasks where context doesn’t matter.

The pattern

These three verbs together form a memory loop:

1. /load at the start of a session pulls you up to speed
2. You work
3. /pin along the way preserves anything worth keeping
4. /save at the end records the session itself

Next session, /load reads what you wrote, and the loop closes.

Today: only Claude Code

The verbs ship as Claude Code skills today (memcrate install claude-code). Integrations for Cursor, Cowork, Aider, and MCP-aware tools are in development. Until then, the vault format itself is the portable layer — even tools without a Memcrate integration can be pointed at the canonical files and benefit.